chore: update lance dependency to v2.0.1-rc.1

2026-03-28 03:20:39 +00:00 · 2026-02-12 21:16:38 +00:00
53 changed files with 2082 additions and 4083 deletions
--- a/.github/workflows/codex-fix-ci.yml
+++ b/.github/workflows/codex-fix-ci.yml
@@ -1,173 +0,0 @@
-name: Codex Fix CI
-
-on:
-  workflow_dispatch:
-    inputs:
-      workflow_run_url:
-        description: "Failing CI workflow run URL (e.g., https://github.com/lancedb/lancedb/actions/runs/12345678)"
-        required: true
-        type: string
-      branch:
-        description: "Branch to fix (e.g., main, release/v2.0, or feature-branch)"
-        required: true
-        type: string
-      guidelines:
-        description: "Additional guidelines for the fix (optional)"
-        required: false
-        type: string
-
-permissions:
-  contents: write
-  pull-requests: write
-  actions: read
-
-jobs:
-  fix-ci:
-    runs-on: warp-ubuntu-latest-x64-4x
-    timeout-minutes: 60
-    env:
-      CC: clang
-      CXX: clang++
-    steps:
-      - name: Show inputs
-        run: |
-          echo "workflow_run_url = ${{ inputs.workflow_run_url }}"
-          echo "branch = ${{ inputs.branch }}"
-          echo "guidelines = ${{ inputs.guidelines }}"
-
-      - name: Checkout Repo
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ inputs.branch }}
-          fetch-depth: 0
-          persist-credentials: true
-
-      - name: Set up Node.js
-        uses: actions/setup-node@v4
-        with:
-          node-version: 20
-
-      - name: Install Codex CLI
-        run: npm install -g @openai/codex
-
-      - name: Install Rust toolchain
-        uses: dtolnay/rust-toolchain@stable
-        with:
-          toolchain: stable
-          components: clippy, rustfmt
-
-      - uses: Swatinem/rust-cache@v2
-
-      - name: Install system dependencies
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y protobuf-compiler libssl-dev
-
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: '3.11'
-
-      - name: Install Python dependencies
-        run: |
-          pip install maturin ruff pytest pyarrow pandas polars
-
-      - name: Set up Java
-        uses: actions/setup-java@v4
-        with:
-          distribution: temurin
-          java-version: '11'
-          cache: maven
-
-      - name: Install Node.js dependencies for TypeScript bindings
-        run: |
-          cd nodejs
-          npm ci
-
-      - name: Configure git user
-        run: |
-          git config user.name "lancedb automation"
-          git config user.email "robot@lancedb.com"
-
-      - name: Run Codex to fix CI failure
-        env:
-          WORKFLOW_RUN_URL: ${{ inputs.workflow_run_url }}
-          BRANCH: ${{ inputs.branch }}
-          GUIDELINES: ${{ inputs.guidelines }}
-          GITHUB_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-          GH_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-          OPENAI_API_KEY: ${{ secrets.CODEX_TOKEN }}
-        run: |
-          set -euo pipefail
-
-          cat <<EOF >/tmp/codex-prompt.txt
-          You are running inside the lancedb repository on a GitHub Actions runner. Your task is to fix a CI failure.
-
-          Input parameters:
-          - Failing workflow run URL: ${WORKFLOW_RUN_URL}
-          - Branch to fix: ${BRANCH}
-          - Additional guidelines: ${GUIDELINES:-"None provided"}
-
-          Follow these steps exactly:
-
-          1. Extract the run ID from the workflow URL. The URL format is https://github.com/lancedb/lancedb/actions/runs/<run_id>.
-
-          2. Use "gh run view <run_id> --json jobs,conclusion,name" to get information about the failed run.
-
-          3. Identify which jobs failed. For each failed job, use "gh run view <run_id> --job <job_id> --log-failed" to get the failure logs.
-
-          4. Analyze the failure logs to understand what went wrong. Common failures include:
-             - Compilation errors
-             - Test failures
-             - Clippy warnings treated as errors
-             - Formatting issues
-             - Dependency issues
-
-          5. Based on the analysis, fix the issues in the codebase:
-             - For compilation errors: Fix the code that doesn't compile
-             - For test failures: Fix the failing tests or the code they test
-             - For clippy warnings: Apply the suggested fixes
-             - For formatting issues: Run "cargo fmt --all"
-             - For other issues: Apply appropriate fixes
-
-          6. After making fixes, verify them locally:
-             - Run "cargo fmt --all" to ensure formatting is correct
-             - Run "cargo clippy --workspace --tests --all-features -- -D warnings" to check for issues
-             - Run ONLY the specific failing tests to confirm they pass now:
-               - For Rust test failures: Run the specific test with "cargo test -p <crate> <test_name>"
-               - For Python test failures: Build with "cd python && maturin develop" then run "pytest <specific_test_file>::<test_name>"
-               - For Java test failures: Run "cd java && mvn test -Dtest=<TestClass>#<testMethod>"
-               - For TypeScript test failures: Run "cd nodejs && npm run build && npm test -- --testNamePattern='<test_name>'"
-               - Do NOT run the full test suite - only run the tests that were failing
-
-          7. If the additional guidelines are provided, follow them as well.
-
-          8. Inspect "git status --short" and "git diff" to review your changes.
-
-          9. Create a fix branch: "git checkout -b codex/fix-ci-<run_id>".
-
-          10. Stage all changes with "git add -A" and commit with message "fix: resolve CI failures from run <run_id>".
-
-          11. Push the branch: "git push origin codex/fix-ci-<run_id>". If the remote branch exists, delete it first with "gh api -X DELETE repos/lancedb/lancedb/git/refs/heads/codex/fix-ci-<run_id>" then push. Do NOT use "git push --force" or "git push -f".
-
-          12. Create a pull request targeting "${BRANCH}":
-              - Title: "ci: <short summary describing the fix>" (e.g., "ci: fix clippy warnings in lancedb" or "ci: resolve test flakiness in vector search")
-              - First, write the PR body to /tmp/pr-body.md using a heredoc (cat <<'PREOF' > /tmp/pr-body.md). The body should include:
-                - Link to the failing workflow run
-                - Summary of what failed
-                - Description of the fixes applied
-              - Then run "gh pr create --base ${BRANCH} --body-file /tmp/pr-body.md".
-
-          13. Display the new PR URL, "git status --short", and a summary of what was fixed.
-
-          Constraints:
-          - Use bash commands for all operations.
-          - Do not merge the PR.
-          - Do not modify GitHub workflow files unless they are the cause of the failure.
-          - If any command fails, diagnose and attempt to fix the issue instead of aborting immediately.
-          - If you cannot fix the issue automatically, create the PR anyway with a clear explanation of what you tried and what remains to be fixed.
-          - env "GH_TOKEN" is available, use "gh" tools for GitHub-related operations.
-          EOF
-
-          printenv OPENAI_API_KEY | codex login --with-api-key
-          codex --config shell_environment_policy.ignore_default_excludes=true exec --dangerously-bypass-approvals-and-sandbox "$(cat /tmp/codex-prompt.txt)"
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -3072,9 +3072,8 @@ checksum = "42703706b716c37f96a77aea830392ad231f44c9e9a67872fa5548707e11b11c"

 [[package]]
 name = "fsst"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "5f9e5c0b1c67a38cb92b41535d44623483beb9511592ae23a3bf42ddec758690"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow-array",
 "rand 0.9.2",
@@ -4405,9 +4404,8 @@ dependencies = [

 [[package]]
 name = "lance"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2b7f07b905df393a5554eba19055c620f9ea25a3e40a013bda4bd8dc4ca66f01"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow",
 "arrow-arith",
@@ -4472,9 +4470,8 @@ dependencies = [

 [[package]]
 name = "lance-arrow"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "100e076cb81c8f0c24cd2881c706fc53e037c7d6e81eb320e929e265d157effb"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -4493,9 +4490,8 @@ dependencies = [

 [[package]]
 name = "lance-bitpacking"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "588318d3d1ba0f97162fab39a323a0a49866bb35b32af42572c6b6a12296fa27"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrayref",
 "paste",
@@ -4504,9 +4500,8 @@ dependencies = [

 [[package]]
 name = "lance-core"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6fa01d1cf490ccfd3b8eaeee2781415d0419e6be8366040e57e43677abf2644e"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -4543,9 +4538,8 @@ dependencies = [

 [[package]]
 name = "lance-datafusion"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ef89a39e3284eef76f79e63f23de8881a0583ad6feb20ed39f47eadd847a2b88"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow",
 "arrow-array",
@@ -4575,9 +4569,8 @@ dependencies = [

 [[package]]
 name = "lance-datagen"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fc2a60eef5c47e65d91e2ffa8e7e1629c52e7190c8b88a371a1a60601dc49371"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow",
 "arrow-array",
@@ -4595,9 +4588,8 @@ dependencies = [

 [[package]]
 name = "lance-encoding"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "95ce4a6631308aa681b2671af8f2a845ff781f8d4e755a2a7ccd012379467094"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow-arith",
 "arrow-array",
@@ -4634,9 +4626,8 @@ dependencies = [

 [[package]]
 name = "lance-file"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e2d4d82357cbfaa1a18494226c15b1cb3c8ed0b6c84b91146323c82047ede419"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow-arith",
 "arrow-array",
@@ -4668,9 +4659,8 @@ dependencies = [

 [[package]]
 name = "lance-geo"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a7183fc870da62826f0f97df8007b634da053eb310157856efe1dc74f446951c"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "datafusion",
 "geo-traits",
@@ -4684,9 +4674,8 @@ dependencies = [

 [[package]]
 name = "lance-index"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "20e9c5aa7024a63af9ae89ee8c0f23c8421b7896742e5cd4a271a60f9956cb80"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow",
 "arrow-arith",
@@ -4753,9 +4742,8 @@ dependencies = [

 [[package]]
 name = "lance-io"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "c7d2af0b17fb374a8181bcf1a10bce5703ae3ee4373c1587ce4bba23e15e45c8"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow",
 "arrow-arith",
@@ -4795,9 +4783,8 @@ dependencies = [

 [[package]]
 name = "lance-linalg"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "5125aa62696e75a7475807564b4921f252d8815be606b84bc00e6def0f5c24bb"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -4813,9 +4800,8 @@ dependencies = [

 [[package]]
 name = "lance-namespace"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "70545c2676ce954dfd801da5c6a631a70bba967826cd3a8f31b47d1f04bbfed3"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow",
 "async-trait",
@@ -4827,9 +4813,8 @@ dependencies = [

 [[package]]
 name = "lance-namespace-impls"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "92519f9f27d62655030aac62ea0db9614b65f086ebe651c1b0a96e351b668022"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow",
 "arrow-ipc",
@@ -4872,9 +4857,8 @@ dependencies = [

 [[package]]
 name = "lance-table"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b06ad37bd90045de8ef533df170c6098e6ff6ecb427aade47d7db8e2c86f2678"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow",
 "arrow-array",
@@ -4913,9 +4897,8 @@ dependencies = [

 [[package]]
 name = "lance-testing"
-version = "2.0.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "cd7f13b0f2b6337af015dcb1519645388dca08c970037aa77aff517687c4019f"
+version = "2.0.1-rc.1"
+source = "git+https://github.com/lance-format/lance.git?tag=v2.0.1-rc.1#230f285e1c0a72d0ccb723721591e78dd871a419"
 dependencies = [
 "arrow-array",
 "arrow-schema",
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -15,20 +15,20 @@ categories = ["database-implementations"]
 rust-version = "1.88.0"

 [workspace.dependencies]
-lance = { "version" = "=2.0.1", default-features = false }
-lance-core = "=2.0.1"
-lance-datagen = "=2.0.1"
-lance-file = "=2.0.1"
-lance-io = { "version" = "=2.0.1", default-features = false }
-lance-index = "=2.0.1"
-lance-linalg = "=2.0.1"
-lance-namespace = "=2.0.1"
-lance-namespace-impls = { "version" = "=2.0.1", default-features = false }
-lance-table = "=2.0.1"
-lance-testing = "=2.0.1"
-lance-datafusion = "=2.0.1"
-lance-encoding = "=2.0.1"
-lance-arrow = "=2.0.1"
+lance = { "version" = "=2.0.1-rc.1", default-features = false, "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-core = { "version" = "=2.0.1-rc.1", "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-datagen = { "version" = "=2.0.1-rc.1", "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-file = { "version" = "=2.0.1-rc.1", "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-io = { "version" = "=2.0.1-rc.1", default-features = false, "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-index = { "version" = "=2.0.1-rc.1", "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-linalg = { "version" = "=2.0.1-rc.1", "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-namespace = { "version" = "=2.0.1-rc.1", "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-namespace-impls = { "version" = "=2.0.1-rc.1", default-features = false, "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-table = { "version" = "=2.0.1-rc.1", "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-testing = { "version" = "=2.0.1-rc.1", "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-datafusion = { "version" = "=2.0.1-rc.1", "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-encoding = { "version" = "=2.0.1-rc.1", "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
+lance-arrow = { "version" = "=2.0.1-rc.1", "tag" = "v2.0.1-rc.1", "git" = "https://github.com/lance-format/lance.git" }
 ahash = "0.8"
 # Note that this one does not include pyarrow
 arrow = { version = "57.2", optional = false }
--- a/java/pom.xml
+++ b/java/pom.xml
@@ -28,7 +28,7 @@
    <properties>
        <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
        <arrow.version>15.0.0</arrow.version>
-        <lance-core.version>2.0.1</lance-core.version>
+        <lance-core.version>2.0.1-rc.1</lance-core.version>
        <spotless.skip>false</spotless.skip>
        <spotless.version>2.30.0</spotless.version>
        <spotless.java.googlejavaformat.version>1.7</spotless.java.googlejavaformat.version>
--- a/nodejs/src/connection.rs
+++ b/nodejs/src/connection.rs
@@ -13,7 +13,6 @@ use crate::header::JsHeaderProvider;
 use crate::table::Table;
 use crate::ConnectionOptions;
 use lancedb::connection::{ConnectBuilder, Connection as LanceDBConnection};
-
 use lancedb::ipc::{ipc_file_to_batches, ipc_file_to_schema};

 #[napi]
--- a/python/.bumpversion.toml
+++ b/python/.bumpversion.toml
@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.30.0-beta.0"
+current_version = "0.29.2"
 parse = """(?x)
    (?P<major>0|[1-9]\\d*)\\.
    (?P<minor>0|[1-9]\\d*)\\.
--- a/python/Cargo.toml
+++ b/python/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-python"
-version = "0.30.0-beta.0"
+version = "0.29.2"
 edition.workspace = true
 description = "Python bindings for LanceDB"
 license.workspace = true
--- a/python/python/lancedb/permutation.py
+++ b/python/python/lancedb/permutation.py
@@ -9,7 +9,7 @@ import json
 from ._lancedb import async_permutation_builder, PermutationReader
 from .table import LanceTable
 from .background_loop import LOOP
-from .util import batch_to_tensor, batch_to_tensor_rows
+from .util import batch_to_tensor
 from typing import Any, Callable, Iterator, Literal, Optional, TYPE_CHECKING, Union

 if TYPE_CHECKING:
@@ -333,11 +333,7 @@ class Transforms:
    """

    @staticmethod
-    def arrow2python(batch: pa.RecordBatch) -> list[dict[str, Any]]:
-        return batch.to_pylist()
-
-    @staticmethod
-    def arrow2pythoncol(batch: pa.RecordBatch) -> dict[str, list[Any]]:
+    def arrow2python(batch: pa.RecordBatch) -> dict[str, list[Any]]:
        return batch.to_pydict()

    @staticmethod
@@ -691,17 +687,7 @@ class Permutation:
            return

    def with_format(
-        self,
-        format: Literal[
-            "numpy",
-            "python",
-            "python_col",
-            "pandas",
-            "arrow",
-            "torch",
-            "torch_col",
-            "polars",
-        ],
+        self, format: Literal["numpy", "python", "pandas", "arrow", "torch", "polars"]
    ) -> "Permutation":
        """
        Set the format for batches
@@ -710,18 +696,16 @@ class Permutation:

        The format can be one of:
        - "numpy" - the batch will be a dict of numpy arrays (one per column)
-        - "python" - the batch will be a list of dicts (one per row)
-        - "python_col" - the batch will be a dict of lists (one entry per column)
+        - "python" - the batch will be a dict of lists (one per column)
        - "pandas" - the batch will be a pandas DataFrame
        - "arrow" - the batch will be a pyarrow RecordBatch
-        - "torch" - the batch will be a list of tensors, one per row
-        - "torch_col" - the batch will be a 2D torch tensor (first dim indexes columns)
+        - "torch" - the batch will be a two dimensional torch tensor
        - "polars" - the batch will be a polars DataFrame

        Conversion may or may not involve a data copy.  Lance uses Arrow internally
-        and so it is able to zero-copy to the arrow and polars formats.
+        and so it is able to zero-copy to the arrow and polars.

-        Conversion to torch_col will be zero-copy but will only support a subset of data
+        Conversion to torch will be zero-copy but will only support a subset of data
        types (numeric types).

        Conversion to numpy and/or pandas will typically be zero-copy for numeric
@@ -734,8 +718,6 @@ class Permutation:
        assert format is not None, "format is required"
        if format == "python":
            return self.with_transform(Transforms.arrow2python)
-        if format == "python_col":
-            return self.with_transform(Transforms.arrow2pythoncol)
        elif format == "numpy":
            return self.with_transform(Transforms.arrow2numpy)
        elif format == "pandas":
@@ -743,8 +725,6 @@ class Permutation:
        elif format == "arrow":
            return self.with_transform(Transforms.arrow2arrow)
        elif format == "torch":
-            return self.with_transform(batch_to_tensor_rows)
-        elif format == "torch_col":
            return self.with_transform(batch_to_tensor)
        elif format == "polars":
            return self.with_transform(Transforms.arrow2polars())
--- a/python/python/lancedb/util.py
+++ b/python/python/lancedb/util.py
@@ -419,22 +419,3 @@ def batch_to_tensor(batch: pa.RecordBatch):
    """
    torch = attempt_import_or_raise("torch", "torch")
    return torch.stack([torch.from_dlpack(col) for col in batch.columns])
-
-
-def batch_to_tensor_rows(batch: pa.RecordBatch):
-    """
-    Convert a PyArrow RecordBatch to a list of PyTorch Tensor, one per row
-
-    Each column is converted to a tensor (using zero-copy via DLPack)
-    and the columns are then stacked into a single tensor.  The 2D tensor
-    is then converted to a list of tensors, one per row
-
-    Fails if torch or numpy is not installed.
-    Fails if a column's data type is not supported by PyTorch.
-    """
-    torch = attempt_import_or_raise("torch", "torch")
-    numpy = attempt_import_or_raise("numpy", "numpy")
-    columns = [col.to_numpy(zero_copy_only=False) for col in batch.columns]
-    stacked = torch.tensor(numpy.column_stack(columns))
-    rows = list(stacked.unbind(dim=0))
-    return rows
--- a/python/python/tests/test_permutation.py
+++ b/python/python/tests/test_permutation.py
@@ -664,20 +664,23 @@ def test_iter_basic(some_permutation: Permutation):
    expected_batches = (950 + batch_size - 1) // batch_size  # ceiling division
    assert len(batches) == expected_batches

-    # Check that all batches are lists of dicts (default python format)
-    assert all(isinstance(batch, list) for batch in batches)
+    # Check that all batches are dicts (default python format)
+    assert all(isinstance(batch, dict) for batch in batches)

    # Check that batches have the correct structure
    for batch in batches:
-        assert "id" in batch[0]
-        assert "value" in batch[0]
+        assert "id" in batch
+        assert "value" in batch
+        assert isinstance(batch["id"], list)
+        assert isinstance(batch["value"], list)

    # Check that all batches except the last have the correct size
    for batch in batches[:-1]:
-        assert len(batch) == batch_size
+        assert len(batch["id"]) == batch_size
+        assert len(batch["value"]) == batch_size

    # Last batch might be smaller
-    assert len(batches[-1]) <= batch_size
+    assert len(batches[-1]["id"]) <= batch_size


 def test_iter_skip_last_batch(some_permutation: Permutation):
@@ -696,11 +699,11 @@ def test_iter_skip_last_batch(some_permutation: Permutation):
    if 950 % batch_size != 0:
        assert len(batches_without_skip) == num_full_batches + 1
        # Last batch should be smaller
-        assert len(batches_without_skip[-1]) == 950 % batch_size
+        assert len(batches_without_skip[-1]["id"]) == 950 % batch_size

    # All batches with skip_last_batch should be full size
    for batch in batches_with_skip:
-        assert len(batch) == batch_size
+        assert len(batch["id"]) == batch_size


 def test_iter_different_batch_sizes(some_permutation: Permutation):
@@ -717,12 +720,12 @@ def test_iter_different_batch_sizes(some_permutation: Permutation):
    # Test with batch size equal to total rows
    single_batch = list(some_permutation.iter(950, skip_last_batch=False))
    assert len(single_batch) == 1
-    assert len(single_batch[0]) == 950
+    assert len(single_batch[0]["id"]) == 950

    # Test with batch size larger than total rows
    oversized_batch = list(some_permutation.iter(10000, skip_last_batch=False))
    assert len(oversized_batch) == 1
-    assert len(oversized_batch[0]) == 950
+    assert len(oversized_batch[0]["id"]) == 950


 def test_dunder_iter(some_permutation: Permutation):
@@ -735,13 +738,15 @@ def test_dunder_iter(some_permutation: Permutation):

    # All batches should be full size
    for batch in batches:
-        assert len(batch) == 100
+        assert len(batch["id"]) == 100
+        assert len(batch["value"]) == 100

    some_permutation = some_permutation.with_batch_size(400)
    batches = list(some_permutation)
    assert len(batches) == 2  # floor(950 / 400) since skip_last_batch=True
    for batch in batches:
-        assert len(batch) == 400
+        assert len(batch["id"]) == 400
+        assert len(batch["value"]) == 400


 def test_iter_with_different_formats(some_permutation: Permutation):
@@ -756,7 +761,7 @@ def test_iter_with_different_formats(some_permutation: Permutation):
    # Test with python format (default)
    python_perm = some_permutation.with_format("python")
    python_batches = list(python_perm.iter(batch_size, skip_last_batch=False))
-    assert all(isinstance(batch, list) for batch in python_batches)
+    assert all(isinstance(batch, dict) for batch in python_batches)

    # Test with pandas format
    pandas_perm = some_permutation.with_format("pandas")
@@ -775,8 +780,8 @@ def test_iter_with_column_selection(some_permutation: Permutation):

    # Check that batches only contain the id column
    for batch in batches:
-        assert "id" in batch[0]
-        assert "value" not in batch[0]
+        assert "id" in batch
+        assert "value" not in batch


 def test_iter_with_column_rename(some_permutation: Permutation):
@@ -786,9 +791,9 @@ def test_iter_with_column_rename(some_permutation: Permutation):

    # Check that batches have the renamed column
    for batch in batches:
-        assert "id" in batch[0]
-        assert "data" in batch[0]
-        assert "value" not in batch[0]
+        assert "id" in batch
+        assert "data" in batch
+        assert "value" not in batch


 def test_iter_with_limit_offset(some_permutation: Permutation):
@@ -807,14 +812,14 @@ def test_iter_with_limit_offset(some_permutation: Permutation):
    assert len(limit_batches) == 5

    no_skip = some_permutation.iter(101, skip_last_batch=False)
-    row_100 = next(no_skip)[100]["id"]
+    row_100 = next(no_skip)["id"][100]

    # Test with both limit and offset
    limited_perm = some_permutation.with_skip(100).with_take(300)
    limited_batches = list(limited_perm.iter(100, skip_last_batch=False))
    # Should have 3 batches (300 / 100)
    assert len(limited_batches) == 3
-    assert limited_batches[0][0]["id"] == row_100
+    assert limited_batches[0]["id"][0] == row_100


 def test_iter_empty_permutation(mem_db):
@@ -837,7 +842,7 @@ def test_iter_single_row(mem_db):
    # With skip_last_batch=False, should get one batch
    batches = list(perm.iter(10, skip_last_batch=False))
    assert len(batches) == 1
-    assert len(batches[0]) == 1
+    assert len(batches[0]["id"]) == 1

    # With skip_last_batch=True, should skip the single row (since it's < batch_size)
    batches_skip = list(perm.iter(10, skip_last_batch=True))
@@ -855,7 +860,8 @@ def test_identity_permutation(mem_db):

    batches = list(permutation.iter(10, skip_last_batch=False))
    assert len(batches) == 1
-    assert len(batches[0]) == 10
+    assert len(batches[0]["id"]) == 10
+    assert len(batches[0]["value"]) == 10

    permutation = permutation.remove_columns(["value"])
    assert permutation.num_columns == 1
@@ -898,10 +904,10 @@ def test_transform_fn(mem_db):
    py_result = list(permutation.with_format("python").iter(10, skip_last_batch=False))[
        0
    ]
-    assert len(py_result) == 10
-    assert "id" in py_result[0]
-    assert "value" in py_result[0]
-    assert isinstance(py_result, list)
+    assert len(py_result) == 2
+    assert len(py_result["id"]) == 10
+    assert len(py_result["value"]) == 10
+    assert isinstance(py_result, dict)

    try:
        import torch
@@ -909,11 +915,9 @@ def test_transform_fn(mem_db):
        torch_result = list(
            permutation.with_format("torch").iter(10, skip_last_batch=False)
        )[0]
-        assert isinstance(torch_result, list)
-        assert len(torch_result) == 10
-        assert isinstance(torch_result[0], torch.Tensor)
-        assert torch_result[0].shape == (2,)
-        assert torch_result[0].dtype == torch.int64
+        assert torch_result.shape == (2, 10)
+        assert torch_result.dtype == torch.int64
+        assert isinstance(torch_result, torch.Tensor)
    except ImportError:
        # Skip check if torch is not installed
        pass
@@ -946,16 +950,17 @@ def test_custom_transform(mem_db):
 def test_getitems_basic(some_permutation: Permutation):
    """Test __getitems__ returns correct rows by offset."""
    result = some_permutation.__getitems__([0, 1, 2])
-    assert isinstance(result, list)
-    assert "id" in result[0]
-    assert "value" in result[0]
-    assert len(result) == 3
+    assert isinstance(result, dict)
+    assert "id" in result
+    assert "value" in result
+    assert len(result["id"]) == 3


 def test_getitems_single_index(some_permutation: Permutation):
    """Test __getitems__ with a single index."""
    result = some_permutation.__getitems__([0])
-    assert len(result) == 1
+    assert len(result["id"]) == 1
+    assert len(result["value"]) == 1


 def test_getitems_preserves_order(some_permutation: Permutation):
@@ -965,40 +970,38 @@ def test_getitems_preserves_order(some_permutation: Permutation):
    # Get the same rows in reverse order
    reverse = some_permutation.__getitems__([4, 3, 2, 1, 0])

-    assert [r["id"] for r in forward] == list(reversed([r["id"] for r in reverse]))
-    assert [r["value"] for r in forward] == list(
-        reversed([r["value"] for r in reverse])
-    )
+    assert forward["id"] == list(reversed(reverse["id"]))
+    assert forward["value"] == list(reversed(reverse["value"]))


 def test_getitems_non_contiguous(some_permutation: Permutation):
    """Test __getitems__ with non-contiguous indices."""
    result = some_permutation.__getitems__([0, 10, 50, 100, 500])
-    assert len(result) == 5
+    assert len(result["id"]) == 5

    # Each id/value pair should match what we'd get individually
    for i, offset in enumerate([0, 10, 50, 100, 500]):
        single = some_permutation.__getitems__([offset])
-        assert result[i]["id"] == single[0]["id"]
-        assert result[i]["value"] == single[0]["value"]
+        assert result["id"][i] == single["id"][0]
+        assert result["value"][i] == single["value"][0]


 def test_getitems_with_column_selection(some_permutation: Permutation):
    """Test __getitems__ respects column selection."""
    id_only = some_permutation.select_columns(["id"])
    result = id_only.__getitems__([0, 1, 2])
-    assert "id" in result[0]
-    assert "value" not in result[0]
-    assert len(result) == 3
+    assert "id" in result
+    assert "value" not in result
+    assert len(result["id"]) == 3


 def test_getitems_with_column_rename(some_permutation: Permutation):
    """Test __getitems__ respects column renames."""
    renamed = some_permutation.rename_column("value", "data")
    result = renamed.__getitems__([0, 1])
-    assert "data" in result[0]
-    assert "value" not in result[0]
-    assert len(result) == 2
+    assert "data" in result
+    assert "value" not in result
+    assert len(result["data"]) == 2


 def test_getitems_with_format(some_permutation: Permutation):
@@ -1029,8 +1032,8 @@ def test_getitems_identity_permutation(mem_db):
    perm = Permutation.identity(tbl)

    result = perm.__getitems__([0, 5, 9])
-    assert [r["id"] for r in result] == [0, 5, 9]
-    assert [r["value"] for r in result] == [0, 5, 9]
+    assert result["id"] == [0, 5, 9]
+    assert result["value"] == [0, 5, 9]


 def test_getitems_with_limit_offset(some_permutation: Permutation):
@@ -1039,12 +1042,12 @@ def test_getitems_with_limit_offset(some_permutation: Permutation):

    # Should be able to access offsets within the limited range
    result = limited.__getitems__([0, 1, 199])
-    assert len(result) == 3
+    assert len(result["id"]) == 3

    # The first item of the limited permutation should match offset 100 of original
    full_result = some_permutation.__getitems__([100])
    limited_result = limited.__getitems__([0])
-    assert limited_result[0]["id"] == full_result[0]["id"]
+    assert limited_result["id"][0] == full_result["id"][0]


 def test_getitems_invalid_offset(some_permutation: Permutation):
--- a/python/python/tests/test_torch.py
+++ b/python/python/tests/test_torch.py
@@ -4,7 +4,6 @@
 import pyarrow as pa
 import pytest
 from lancedb.util import tbl_to_tensor
-from lancedb.permutation import Permutation

 torch = pytest.importorskip("torch")

@@ -17,26 +16,3 @@ def test_table_dataloader(mem_db):
    for batch in dataloader:
        assert batch.size(0) == 1
        assert batch.size(1) == 10
-
-
-def test_permutation_dataloader(mem_db):
-    table = mem_db.create_table("test_table", pa.table({"a": range(1000)}))
-
-    permutation = Permutation.identity(table)
-    dataloader = torch.utils.data.DataLoader(permutation, batch_size=10, shuffle=True)
-    for batch in dataloader:
-        assert batch["a"].size(0) == 10
-
-    permutation = permutation.with_format("torch")
-    dataloader = torch.utils.data.DataLoader(permutation, batch_size=10, shuffle=True)
-    for batch in dataloader:
-        assert batch.size(0) == 10
-        assert batch.size(1) == 1
-
-    permutation = permutation.with_format("torch_col")
-    dataloader = torch.utils.data.DataLoader(
-        permutation, collate_fn=lambda x: x, batch_size=10, shuffle=True
-    )
-    for batch in dataloader:
-        assert batch.size(0) == 1
-        assert batch.size(1) == 10
--- a/python/src/connection.rs
+++ b/python/src/connection.rs
@@ -121,8 +121,7 @@ impl Connection {

        let mode = Self::parse_create_mode_str(mode)?;

-        let batches: Box<dyn arrow::array::RecordBatchReader + Send> =
-            Box::new(ArrowArrayStreamReader::from_pyarrow_bound(&data)?);
+        let batches = ArrowArrayStreamReader::from_pyarrow_bound(&data)?;

        let mut builder = inner.create_table(name, batches).mode(mode);

--- a/python/src/table.rs
+++ b/python/src/table.rs
@@ -296,8 +296,7 @@ impl Table {
        data: Bound<'_, PyAny>,
        mode: String,
    ) -> PyResult<Bound<'a, PyAny>> {
-        let batches: Box<dyn arrow::array::RecordBatchReader + Send> =
-            Box::new(ArrowArrayStreamReader::from_pyarrow_bound(&data)?);
+        let batches = ArrowArrayStreamReader::from_pyarrow_bound(&data)?;
        let mut op = self_.inner_ref()?.add(batches);
        if mode == "append" {
            op = op.mode(AddDataMode::Append);
--- a/rust/lancedb/examples/bedrock.rs
+++ b/rust/lancedb/examples/bedrock.rs
@@ -3,12 +3,13 @@

 use std::{iter::once, sync::Arc};

-use arrow_array::{Float64Array, Int32Array, RecordBatch, StringArray};
+use arrow_array::{Float64Array, Int32Array, RecordBatch, RecordBatchIterator, StringArray};
 use arrow_schema::{DataType, Field, Schema};
 use aws_config::Region;
 use aws_sdk_bedrockruntime::Client;
 use futures::StreamExt;
 use lancedb::{
+    arrow::IntoArrow,
    connect,
    embeddings::{bedrock::BedrockEmbeddingFunction, EmbeddingDefinition, EmbeddingFunction},
    query::{ExecutableQuery, QueryBase},
@@ -66,7 +67,7 @@ async fn main() -> Result<()> {
    Ok(())
 }

-fn make_data() -> RecordBatch {
+fn make_data() -> impl IntoArrow {
    let schema = Schema::new(vec![
        Field::new("id", DataType::Int32, true),
        Field::new("text", DataType::Utf8, false),
@@ -82,9 +83,10 @@ fn make_data() -> RecordBatch {
    ]);
    let price = Float64Array::from(vec![10.0, 50.0, 100.0, 30.0]);
    let schema = Arc::new(schema);
-    RecordBatch::try_new(
+    let rb = RecordBatch::try_new(
        schema.clone(),
        vec![Arc::new(id), Arc::new(text), Arc::new(price)],
    )
-    .unwrap()
+    .unwrap();
+    Box::new(RecordBatchIterator::new(vec![Ok(rb)], schema))
 }
--- a/rust/lancedb/examples/full_text_search.rs
+++ b/rust/lancedb/examples/full_text_search.rs
@@ -3,13 +3,12 @@

 use std::sync::Arc;

-use arrow_array::{Int32Array, RecordBatch, RecordBatchIterator, StringArray};
+use arrow_array::{Int32Array, RecordBatch, RecordBatchIterator, RecordBatchReader, StringArray};
 use arrow_schema::{DataType, Field, Schema};

 use futures::TryStreamExt;
 use lance_index::scalar::FullTextSearchQuery;
 use lancedb::connection::Connection;
-
 use lancedb::index::scalar::FtsIndexBuilder;
 use lancedb::index::Index;
 use lancedb::query::{ExecutableQuery, QueryBase};
@@ -30,7 +29,7 @@ async fn main() -> Result<()> {
    Ok(())
 }

-fn create_some_records() -> Result<Box<dyn arrow_array::RecordBatchReader + Send>> {
+fn create_some_records() -> Result<Box<dyn RecordBatchReader + Send>> {
    const TOTAL: usize = 1000;

    let schema = Arc::new(Schema::new(vec![
@@ -67,7 +66,7 @@ fn create_some_records() -> Result<Box<dyn arrow_array::RecordBatchReader + Send
 }

 async fn create_table(db: &Connection) -> Result<Table> {
-    let initial_data = create_some_records()?;
+    let initial_data: Box<dyn RecordBatchReader + Send> = create_some_records()?;
    let tbl = db.create_table("my_table", initial_data).execute().await?;
    Ok(tbl)
 }
--- a/rust/lancedb/examples/hybrid_search.rs
+++ b/rust/lancedb/examples/hybrid_search.rs
@@ -1,13 +1,14 @@
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright The LanceDB Authors

-use arrow_array::{RecordBatch, StringArray};
+use arrow_array::{RecordBatch, RecordBatchIterator, StringArray};
 use arrow_schema::{DataType, Field, Schema};
 use futures::TryStreamExt;
 use lance_index::scalar::FullTextSearchQuery;
 use lancedb::index::scalar::FtsIndexBuilder;
 use lancedb::index::Index;
 use lancedb::{
+    arrow::IntoArrow,
    connect,
    embeddings::{
        sentence_transformers::SentenceTransformersEmbeddings, EmbeddingDefinition,
@@ -69,7 +70,7 @@ async fn main() -> Result<()> {
    Ok(())
 }

-fn make_data() -> RecordBatch {
+fn make_data() -> impl IntoArrow {
    let schema = Schema::new(vec![Field::new("facts", DataType::Utf8, false)]);

    let facts = StringArray::from_iter_values(vec![
@@ -100,7 +101,8 @@ fn make_data() -> RecordBatch {
        "The first chatbot was ELIZA, created in the 1960s.",
    ]);
    let schema = Arc::new(schema);
-    RecordBatch::try_new(schema.clone(), vec![Arc::new(facts)]).unwrap()
+    let rb = RecordBatch::try_new(schema.clone(), vec![Arc::new(facts)]).unwrap();
+    Box::new(RecordBatchIterator::new(vec![Ok(rb)], schema))
 }

 async fn create_index(table: &Table) -> Result<()> {
--- a/rust/lancedb/examples/ivf_pq.rs
+++ b/rust/lancedb/examples/ivf_pq.rs
@@ -8,12 +8,13 @@
 use std::sync::Arc;

 use arrow_array::types::Float32Type;
-use arrow_array::{FixedSizeListArray, Int32Array, RecordBatch, RecordBatchIterator};
+use arrow_array::{
+    FixedSizeListArray, Int32Array, RecordBatch, RecordBatchIterator, RecordBatchReader,
+};
 use arrow_schema::{DataType, Field, Schema};

 use futures::TryStreamExt;
 use lancedb::connection::Connection;
-
 use lancedb::index::vector::IvfPqIndexBuilder;
 use lancedb::index::Index;
 use lancedb::query::{ExecutableQuery, QueryBase};
@@ -33,7 +34,7 @@ async fn main() -> Result<()> {
    Ok(())
 }

-fn create_some_records() -> Result<Box<dyn arrow_array::RecordBatchReader + Send>> {
+fn create_some_records() -> Result<Box<dyn RecordBatchReader + Send>> {
    const TOTAL: usize = 1000;
    const DIM: usize = 128;

@@ -72,9 +73,9 @@ fn create_some_records() -> Result<Box<dyn arrow_array::RecordBatchReader + Send
 }

 async fn create_table(db: &Connection) -> Result<Table> {
-    let initial_data = create_some_records()?;
+    let initial_data: Box<dyn RecordBatchReader + Send> = create_some_records()?;
    let tbl = db
-        .create_table("my_table", initial_data)
+        .create_table("my_table", Box::new(initial_data))
        .execute()
        .await
        .unwrap();
--- a/rust/lancedb/examples/openai.rs
+++ b/rust/lancedb/examples/openai.rs
@@ -5,9 +5,11 @@

 use std::{iter::once, sync::Arc};

-use arrow_array::{RecordBatch, StringArray};
+use arrow_array::{Float64Array, Int32Array, RecordBatch, RecordBatchIterator, StringArray};
+use arrow_schema::{DataType, Field, Schema};
 use futures::StreamExt;
 use lancedb::{
+    arrow::IntoArrow,
    connect,
    embeddings::{openai::OpenAIEmbeddingFunction, EmbeddingDefinition, EmbeddingFunction},
    query::{ExecutableQuery, QueryBase},
@@ -62,20 +64,26 @@ async fn main() -> Result<()> {
 }
 // --8<-- [end:openai_embeddings]

-fn make_data() -> RecordBatch {
-    arrow_array::record_batch!(
-        ("id", Int32, [1, 2, 3, 4]),
-        (
-            "text",
-            Utf8,
-            [
-                "Black T-Shirt",
-                "Leather Jacket",
-                "Winter Parka",
-                "Hooded Sweatshirt"
-            ]
-        ),
-        ("price", Float64, [10.0, 50.0, 100.0, 30.0])
+fn make_data() -> impl IntoArrow {
+    let schema = Schema::new(vec![
+        Field::new("id", DataType::Int32, true),
+        Field::new("text", DataType::Utf8, false),
+        Field::new("price", DataType::Float64, false),
+    ]);
+
+    let id = Int32Array::from(vec![1, 2, 3, 4]);
+    let text = StringArray::from_iter_values(vec![
+        "Black T-Shirt",
+        "Leather Jacket",
+        "Winter Parka",
+        "Hooded Sweatshirt",
+    ]);
+    let price = Float64Array::from(vec![10.0, 50.0, 100.0, 30.0]);
+    let schema = Arc::new(schema);
+    let rb = RecordBatch::try_new(
+        schema.clone(),
+        vec![Arc::new(id), Arc::new(text), Arc::new(price)],
    )
-    .unwrap()
+    .unwrap();
+    Box::new(RecordBatchIterator::new(vec![Ok(rb)], schema))
 }
--- a/rust/lancedb/examples/sentence_transformers.rs
+++ b/rust/lancedb/examples/sentence_transformers.rs
@@ -3,10 +3,11 @@

 use std::{iter::once, sync::Arc};

-use arrow_array::{RecordBatch, StringArray};
+use arrow_array::{RecordBatch, RecordBatchIterator, StringArray};
 use arrow_schema::{DataType, Field, Schema};
 use futures::StreamExt;
 use lancedb::{
+    arrow::IntoArrow,
    connect,
    embeddings::{
        sentence_transformers::SentenceTransformersEmbeddings, EmbeddingDefinition,
@@ -58,7 +59,7 @@ async fn main() -> Result<()> {
    Ok(())
 }

-fn make_data() -> RecordBatch {
+fn make_data() -> impl IntoArrow {
    let schema = Schema::new(vec![Field::new("facts", DataType::Utf8, false)]);

    let facts = StringArray::from_iter_values(vec![
@@ -89,5 +90,6 @@ fn make_data() -> RecordBatch {
        "The first chatbot was ELIZA, created in the 1960s.",
    ]);
    let schema = Arc::new(schema);
-    RecordBatch::try_new(schema.clone(), vec![Arc::new(facts)]).unwrap()
+    let rb = RecordBatch::try_new(schema.clone(), vec![Arc::new(facts)]).unwrap();
+    Box::new(RecordBatchIterator::new(vec![Ok(rb)], schema))
 }
--- a/rust/lancedb/examples/simple.rs
+++ b/rust/lancedb/examples/simple.rs
@@ -8,9 +8,11 @@
 use std::sync::Arc;

 use arrow_array::types::Float32Type;
-use arrow_array::{FixedSizeListArray, Int32Array, RecordBatch};
+use arrow_array::{FixedSizeListArray, Int32Array, RecordBatch, RecordBatchIterator};
 use arrow_schema::{DataType, Field, Schema};
 use futures::TryStreamExt;
+
+use lancedb::arrow::IntoArrow;
 use lancedb::connection::Connection;
 use lancedb::index::Index;
 use lancedb::query::{ExecutableQuery, QueryBase};
@@ -57,7 +59,7 @@ async fn open_with_existing_tbl() -> Result<()> {
    Ok(())
 }

-fn create_some_records() -> Result<RecordBatch> {
+fn create_some_records() -> Result<impl IntoArrow> {
    const TOTAL: usize = 1000;
    const DIM: usize = 128;

@@ -74,18 +76,25 @@ fn create_some_records() -> Result<RecordBatch> {
    ]));

    // Create a RecordBatch stream.
-    Ok(RecordBatch::try_new(
-        schema.clone(),
-        vec![
-            Arc::new(Int32Array::from_iter_values(0..TOTAL as i32)),
-            Arc::new(
-                FixedSizeListArray::from_iter_primitive::<Float32Type, _, _>(
-                    (0..TOTAL).map(|_| Some(vec![Some(1.0); DIM])),
-                    DIM as i32,
+    let batches = RecordBatchIterator::new(
+        vec![RecordBatch::try_new(
+            schema.clone(),
+            vec![
+                Arc::new(Int32Array::from_iter_values(0..TOTAL as i32)),
+                Arc::new(
+                    FixedSizeListArray::from_iter_primitive::<Float32Type, _, _>(
+                        (0..TOTAL).map(|_| Some(vec![Some(1.0); DIM])),
+                        DIM as i32,
+                    ),
                ),
-            ),
-        ],
-    )?)
+            ],
+        )
+        .unwrap()]
+        .into_iter()
+        .map(Ok),
+        schema.clone(),
+    );
+    Ok(Box::new(batches))
 }

 async fn create_table(db: &Connection) -> Result<LanceDbTable> {
--- a/rust/lancedb/src/connection.rs
+++ b/rust/lancedb/src/connection.rs
@@ -6,8 +6,8 @@
 use std::collections::HashMap;
 use std::sync::Arc;

-use arrow_array::RecordBatch;
-use arrow_schema::SchemaRef;
+use arrow_array::RecordBatchReader;
+use arrow_schema::{Field, SchemaRef};
 use lance::dataset::ReadParams;
 use lance_namespace::models::{
    CreateNamespaceRequest, CreateNamespaceResponse, DescribeNamespaceRequest,
@@ -17,20 +17,24 @@ use lance_namespace::models::{
 #[cfg(feature = "aws")]
 use object_store::aws::AwsCredential;

-use crate::connection::create_table::CreateTableBuilder;
-use crate::data::scannable::Scannable;
-use crate::database::listing::ListingDatabase;
-use crate::database::{
-    CloneTableRequest, Database, DatabaseOptions, OpenTableRequest, ReadConsistency,
-    TableNamesRequest,
+use crate::arrow::{IntoArrow, IntoArrowStream, SendableRecordBatchStream};
+use crate::database::listing::{
+    ListingDatabase, OPT_NEW_TABLE_STORAGE_VERSION, OPT_NEW_TABLE_V2_MANIFEST_PATHS,
+};
+use crate::database::{
+    CloneTableRequest, CreateTableData, CreateTableMode, CreateTableRequest, Database,
+    DatabaseOptions, OpenTableRequest, ReadConsistency, TableNamesRequest,
+};
+use crate::embeddings::{
+    EmbeddingDefinition, EmbeddingFunction, EmbeddingRegistry, MemoryRegistry, WithEmbeddings,
 };
-use crate::embeddings::{EmbeddingRegistry, MemoryRegistry};
 use crate::error::{Error, Result};
 #[cfg(feature = "remote")]
 use crate::remote::{
    client::ClientConfig,
    db::{OPT_REMOTE_API_KEY, OPT_REMOTE_HOST_OVERRIDE, OPT_REMOTE_REGION},
 };
+use crate::table::{TableDefinition, WriteOptions};
 use crate::Table;
 use lance::io::ObjectStoreParams;
 pub use lance_encoding::version::LanceFileVersion;
@@ -38,8 +42,6 @@ pub use lance_encoding::version::LanceFileVersion;
 use lance_io::object_store::StorageOptions;
 use lance_io::object_store::{StorageOptionsAccessor, StorageOptionsProvider};

-mod create_table;
-
 fn merge_storage_options(
    store_params: &mut ObjectStoreParams,
    pairs: impl IntoIterator<Item = (String, String)>,
@@ -114,6 +116,337 @@ impl TableNamesBuilder {
    }
 }

+pub struct NoData {}
+
+impl IntoArrow for NoData {
+    fn into_arrow(self) -> Result<Box<dyn arrow_array::RecordBatchReader + Send>> {
+        unreachable!("NoData should never be converted to Arrow")
+    }
+}
+
+// Stores the value given from the initial CreateTableBuilder::new call
+// and defers errors until `execute` is called
+enum CreateTableBuilderInitialData {
+    None,
+    Iterator(Result<Box<dyn RecordBatchReader + Send>>),
+    Stream(Result<SendableRecordBatchStream>),
+}
+
+/// A builder for configuring a [`Connection::create_table`] operation
+pub struct CreateTableBuilder<const HAS_DATA: bool> {
+    parent: Arc<dyn Database>,
+    embeddings: Vec<(EmbeddingDefinition, Arc<dyn EmbeddingFunction>)>,
+    embedding_registry: Arc<dyn EmbeddingRegistry>,
+    request: CreateTableRequest,
+    // This is a bit clumsy but we defer errors until `execute` is called
+    // to maintain backwards compatibility
+    data: CreateTableBuilderInitialData,
+}
+
+// Builder methods that only apply when we have initial data
+impl CreateTableBuilder<true> {
+    fn new<T: IntoArrow>(
+        parent: Arc<dyn Database>,
+        name: String,
+        data: T,
+        embedding_registry: Arc<dyn EmbeddingRegistry>,
+    ) -> Self {
+        let dummy_schema = Arc::new(arrow_schema::Schema::new(Vec::<Field>::default()));
+        Self {
+            parent,
+            request: CreateTableRequest::new(
+                name,
+                CreateTableData::Empty(TableDefinition::new_from_schema(dummy_schema)),
+            ),
+            embeddings: Vec::new(),
+            embedding_registry,
+            data: CreateTableBuilderInitialData::Iterator(data.into_arrow()),
+        }
+    }
+
+    fn new_streaming<T: IntoArrowStream>(
+        parent: Arc<dyn Database>,
+        name: String,
+        data: T,
+        embedding_registry: Arc<dyn EmbeddingRegistry>,
+    ) -> Self {
+        let dummy_schema = Arc::new(arrow_schema::Schema::new(Vec::<Field>::default()));
+        Self {
+            parent,
+            request: CreateTableRequest::new(
+                name,
+                CreateTableData::Empty(TableDefinition::new_from_schema(dummy_schema)),
+            ),
+            embeddings: Vec::new(),
+            embedding_registry,
+            data: CreateTableBuilderInitialData::Stream(data.into_arrow()),
+        }
+    }
+
+    /// Execute the create table operation
+    pub async fn execute(self) -> Result<Table> {
+        let embedding_registry = self.embedding_registry.clone();
+        let parent = self.parent.clone();
+        let request = self.into_request()?;
+        Ok(Table::new_with_embedding_registry(
+            parent.create_table(request).await?,
+            parent,
+            embedding_registry,
+        ))
+    }
+
+    fn into_request(self) -> Result<CreateTableRequest> {
+        if self.embeddings.is_empty() {
+            match self.data {
+                CreateTableBuilderInitialData::Iterator(maybe_iter) => {
+                    let data = maybe_iter?;
+                    Ok(CreateTableRequest {
+                        data: CreateTableData::Data(data),
+                        ..self.request
+                    })
+                }
+                CreateTableBuilderInitialData::None => {
+                    unreachable!("No data provided for CreateTableBuilder<true>")
+                }
+                CreateTableBuilderInitialData::Stream(maybe_stream) => {
+                    let data = maybe_stream?;
+                    Ok(CreateTableRequest {
+                        data: CreateTableData::StreamingData(data),
+                        ..self.request
+                    })
+                }
+            }
+        } else {
+            let CreateTableBuilderInitialData::Iterator(maybe_iter) = self.data else {
+                return Err(Error::NotSupported { message: "Creating a table with embeddings is currently not support when the input is streaming".to_string() });
+            };
+            let data = maybe_iter?;
+            let data = Box::new(WithEmbeddings::new(data, self.embeddings));
+            Ok(CreateTableRequest {
+                data: CreateTableData::Data(data),
+                ..self.request
+            })
+        }
+    }
+}
+
+// Builder methods that only apply when we do not have initial data
+impl CreateTableBuilder<false> {
+    fn new(
+        parent: Arc<dyn Database>,
+        name: String,
+        schema: SchemaRef,
+        embedding_registry: Arc<dyn EmbeddingRegistry>,
+    ) -> Self {
+        let table_definition = TableDefinition::new_from_schema(schema);
+        Self {
+            parent,
+            request: CreateTableRequest::new(name, CreateTableData::Empty(table_definition)),
+            data: CreateTableBuilderInitialData::None,
+            embeddings: Vec::default(),
+            embedding_registry,
+        }
+    }
+
+    /// Execute the create table operation
+    pub async fn execute(self) -> Result<Table> {
+        let parent = self.parent.clone();
+        let embedding_registry = self.embedding_registry.clone();
+        let request = self.into_request()?;
+        Ok(Table::new_with_embedding_registry(
+            parent.create_table(request).await?,
+            parent,
+            embedding_registry,
+        ))
+    }
+
+    fn into_request(self) -> Result<CreateTableRequest> {
+        if self.embeddings.is_empty() {
+            return Ok(self.request);
+        }
+
+        let CreateTableData::Empty(table_def) = self.request.data else {
+            unreachable!("CreateTableBuilder<false> should always have Empty data")
+        };
+
+        let schema = table_def.schema.clone();
+        let empty_batch = arrow_array::RecordBatch::new_empty(schema.clone());
+
+        let reader = Box::new(std::iter::once(Ok(empty_batch)).collect::<Vec<_>>());
+        let reader = arrow_array::RecordBatchIterator::new(reader.into_iter(), schema);
+        let with_embeddings = WithEmbeddings::new(reader, self.embeddings);
+        let table_definition = with_embeddings.table_definition()?;
+
+        Ok(CreateTableRequest {
+            data: CreateTableData::Empty(table_definition),
+            ..self.request
+        })
+    }
+}
+
+impl<const HAS_DATA: bool> CreateTableBuilder<HAS_DATA> {
+    /// Set the mode for creating the table
+    ///
+    /// This controls what happens if a table with the given name already exists
+    pub fn mode(mut self, mode: CreateTableMode) -> Self {
+        self.request.mode = mode;
+        self
+    }
+
+    /// Apply the given write options when writing the initial data
+    pub fn write_options(mut self, write_options: WriteOptions) -> Self {
+        self.request.write_options = write_options;
+        self
+    }
+
+    /// Set an option for the storage layer.
+    ///
+    /// Options already set on the connection will be inherited by the table,
+    /// but can be overridden here.
+    ///
+    /// See available options at <https://lancedb.com/docs/storage/>
+    pub fn storage_option(mut self, key: impl Into<String>, value: impl Into<String>) -> Self {
+        let store_params = self
+            .request
+            .write_options
+            .lance_write_params
+            .get_or_insert(Default::default())
+            .store_params
+            .get_or_insert(Default::default());
+        merge_storage_options(store_params, [(key.into(), value.into())]);
+        self
+    }
+
+    /// Set multiple options for the storage layer.
+    ///
+    /// Options already set on the connection will be inherited by the table,
+    /// but can be overridden here.
+    ///
+    /// See available options at <https://lancedb.com/docs/storage/>
+    pub fn storage_options(
+        mut self,
+        pairs: impl IntoIterator<Item = (impl Into<String>, impl Into<String>)>,
+    ) -> Self {
+        let store_params = self
+            .request
+            .write_options
+            .lance_write_params
+            .get_or_insert(Default::default())
+            .store_params
+            .get_or_insert(Default::default());
+        let updates = pairs
+            .into_iter()
+            .map(|(key, value)| (key.into(), value.into()));
+        merge_storage_options(store_params, updates);
+        self
+    }
+
+    /// Add an embedding definition to the table.
+    ///
+    /// The `embedding_name` must match the name of an embedding function that
+    /// was previously registered with the connection's [`EmbeddingRegistry`].
+    pub fn add_embedding(mut self, definition: EmbeddingDefinition) -> Result<Self> {
+        // Early verification of the embedding name
+        let embedding_func = self
+            .embedding_registry
+            .get(&definition.embedding_name)
+            .ok_or_else(|| Error::EmbeddingFunctionNotFound {
+                name: definition.embedding_name.clone(),
+                reason: "No embedding function found in the connection's embedding_registry"
+                    .to_string(),
+            })?;
+
+        self.embeddings.push((definition, embedding_func));
+        Ok(self)
+    }
+
+    /// Set whether to use V2 manifest paths for the table. (default: false)
+    ///
+    /// These paths provide more efficient opening of tables with many
+    /// versions on object stores.
+    ///
+    /// <div class="warning">Turning this on will make the dataset unreadable
+    /// for older versions of LanceDB (prior to 0.10.0).</div>
+    ///
+    /// To migrate an existing dataset, instead use the
+    /// [[NativeTable::migrate_manifest_paths_v2]].
+    ///
+    /// This has no effect in LanceDB Cloud.
+    #[deprecated(since = "0.15.1", note = "Use `database_options` instead")]
+    pub fn enable_v2_manifest_paths(mut self, use_v2_manifest_paths: bool) -> Self {
+        let store_params = self
+            .request
+            .write_options
+            .lance_write_params
+            .get_or_insert_with(Default::default)
+            .store_params
+            .get_or_insert_with(Default::default);
+        let value = if use_v2_manifest_paths {
+            "true".to_string()
+        } else {
+            "false".to_string()
+        };
+        merge_storage_options(
+            store_params,
+            [(OPT_NEW_TABLE_V2_MANIFEST_PATHS.to_string(), value)],
+        );
+        self
+    }
+
+    /// Set the data storage version.
+    ///
+    /// The default is `LanceFileVersion::Stable`.
+    #[deprecated(since = "0.15.1", note = "Use `database_options` instead")]
+    pub fn data_storage_version(mut self, data_storage_version: LanceFileVersion) -> Self {
+        let store_params = self
+            .request
+            .write_options
+            .lance_write_params
+            .get_or_insert_with(Default::default)
+            .store_params
+            .get_or_insert_with(Default::default);
+        merge_storage_options(
+            store_params,
+            [(
+                OPT_NEW_TABLE_STORAGE_VERSION.to_string(),
+                data_storage_version.to_string(),
+            )],
+        );
+        self
+    }
+
+    /// Set the namespace for the table
+    pub fn namespace(mut self, namespace: Vec<String>) -> Self {
+        self.request.namespace = namespace;
+        self
+    }
+
+    /// Set a custom location for the table.
+    ///
+    /// If not set, the database will derive a location from its URI and the table name.
+    /// This is useful when integrating with namespace systems that manage table locations.
+    pub fn location(mut self, location: impl Into<String>) -> Self {
+        self.request.location = Some(location.into());
+        self
+    }
+
+    /// Set a storage options provider for automatic credential refresh.
+    ///
+    /// This allows tables to automatically refresh cloud storage credentials
+    /// when they expire, enabling long-running operations on remote storage.
+    pub fn storage_options_provider(mut self, provider: Arc<dyn StorageOptionsProvider>) -> Self {
+        let store_params = self
+            .request
+            .write_options
+            .lance_write_params
+            .get_or_insert(Default::default())
+            .store_params
+            .get_or_insert(Default::default());
+        set_storage_options_provider(store_params, provider);
+        self
+    }
+}
+
 #[derive(Clone, Debug)]
 pub struct OpenTableBuilder {
    parent: Arc<dyn Database>,
@@ -351,17 +684,35 @@ impl Connection {
    ///
    /// * `name` - The name of the table
    /// * `initial_data` - The initial data to write to the table
-    pub fn create_table<T: Scannable + 'static>(
+    pub fn create_table<T: IntoArrow>(
        &self,
        name: impl Into<String>,
        initial_data: T,
-    ) -> CreateTableBuilder {
-        let initial_data = Box::new(initial_data);
-        CreateTableBuilder::new(
+    ) -> CreateTableBuilder<true> {
+        CreateTableBuilder::<true>::new(
            self.internal.clone(),
-            self.embedding_registry.clone(),
            name.into(),
            initial_data,
+            self.embedding_registry.clone(),
+        )
+    }
+
+    /// Create a new table from a stream of data
+    ///
+    /// # Parameters
+    ///
+    /// * `name` - The name of the table
+    /// * `initial_data` - The initial data to write to the table
+    pub fn create_table_streaming<T: IntoArrowStream>(
+        &self,
+        name: impl Into<String>,
+        initial_data: T,
+    ) -> CreateTableBuilder<true> {
+        CreateTableBuilder::<true>::new_streaming(
+            self.internal.clone(),
+            name.into(),
+            initial_data,
+            self.embedding_registry.clone(),
        )
    }

@@ -375,9 +726,13 @@ impl Connection {
        &self,
        name: impl Into<String>,
        schema: SchemaRef,
-    ) -> CreateTableBuilder {
-        let empty_batch = RecordBatch::new_empty(schema);
-        self.create_table(name, empty_batch)
+    ) -> CreateTableBuilder<false> {
+        CreateTableBuilder::<false>::new(
+            self.internal.clone(),
+            name.into(),
+            schema,
+            self.embedding_registry.clone(),
+        )
    }

    /// Open an existing table in the database
@@ -994,11 +1349,20 @@ mod test_utils {

 #[cfg(test)]
 mod tests {
+    use crate::database::listing::{ListingDatabaseOptions, NewTableConfig};
+    use crate::query::QueryBase;
+    use crate::query::{ExecutableQuery, QueryExecutionOptions};
+    use crate::test_utils::connection::new_test_connection;
+    use arrow::compute::concat_batches;
+    use arrow_array::RecordBatchReader;
    use arrow_schema::{DataType, Field, Schema};
+    use datafusion_physical_plan::stream::RecordBatchStreamAdapter;
+    use futures::{stream, TryStreamExt};
+    use lance_core::error::{ArrowResult, DataFusionResult};
    use lance_testing::datagen::{BatchGenerator, IncrementingInt32};
    use tempfile::tempdir;

-    use crate::test_utils::connection::new_test_connection;
+    use crate::arrow::SimpleRecordBatchStream;

    use super::*;

@@ -1114,6 +1478,139 @@ mod tests {
        assert_eq!(tables, vec!["table1".to_owned()]);
    }

+    fn make_data() -> Box<dyn RecordBatchReader + Send + 'static> {
+        let id = Box::new(IncrementingInt32::new().named("id".to_string()));
+        Box::new(BatchGenerator::new().col(id).batches(10, 2000))
+    }
+
+    #[tokio::test]
+    async fn test_create_table_v2() {
+        let tmp_dir = tempdir().unwrap();
+        let uri = tmp_dir.path().to_str().unwrap();
+        let db = connect(uri)
+            .database_options(&ListingDatabaseOptions {
+                new_table_config: NewTableConfig {
+                    data_storage_version: Some(LanceFileVersion::Legacy),
+                    ..Default::default()
+                },
+                ..Default::default()
+            })
+            .execute()
+            .await
+            .unwrap();
+
+        let tbl = db
+            .create_table("v1_test", make_data())
+            .execute()
+            .await
+            .unwrap();
+
+        // In v1 the row group size will trump max_batch_length
+        let batches = tbl
+            .query()
+            .limit(20000)
+            .execute_with_options(QueryExecutionOptions {
+                max_batch_length: 50000,
+                ..Default::default()
+            })
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+        assert_eq!(batches.len(), 20);
+
+        let db = connect(uri)
+            .database_options(&ListingDatabaseOptions {
+                new_table_config: NewTableConfig {
+                    data_storage_version: Some(LanceFileVersion::Stable),
+                    ..Default::default()
+                },
+                ..Default::default()
+            })
+            .execute()
+            .await
+            .unwrap();
+
+        let tbl = db
+            .create_table("v2_test", make_data())
+            .execute()
+            .await
+            .unwrap();
+
+        // In v2 the page size is much bigger than 50k so we should get a single batch
+        let batches = tbl
+            .query()
+            .execute_with_options(QueryExecutionOptions {
+                max_batch_length: 50000,
+                ..Default::default()
+            })
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+
+        assert_eq!(batches.len(), 1);
+    }
+
+    #[tokio::test]
+    async fn test_create_table_streaming() {
+        let tmp_dir = tempdir().unwrap();
+
+        let uri = tmp_dir.path().to_str().unwrap();
+        let db = connect(uri).execute().await.unwrap();
+
+        let batches = make_data().collect::<ArrowResult<Vec<_>>>().unwrap();
+
+        let schema = batches.first().unwrap().schema();
+        let one_batch = concat_batches(&schema, batches.iter()).unwrap();
+
+        let ldb_stream = stream::iter(batches.clone().into_iter().map(Result::Ok));
+        let ldb_stream: SendableRecordBatchStream =
+            Box::pin(SimpleRecordBatchStream::new(ldb_stream, schema.clone()));
+
+        let tbl1 = db
+            .create_table_streaming("one", ldb_stream)
+            .execute()
+            .await
+            .unwrap();
+
+        let df_stream = stream::iter(batches.into_iter().map(DataFusionResult::Ok));
+        let df_stream: datafusion_physical_plan::SendableRecordBatchStream =
+            Box::pin(RecordBatchStreamAdapter::new(schema.clone(), df_stream));
+
+        let tbl2 = db
+            .create_table_streaming("two", df_stream)
+            .execute()
+            .await
+            .unwrap();
+
+        let tbl1_data = tbl1
+            .query()
+            .execute()
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+
+        let tbl1_data = concat_batches(&schema, tbl1_data.iter()).unwrap();
+        assert_eq!(tbl1_data, one_batch);
+
+        let tbl2_data = tbl2
+            .query()
+            .execute()
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+
+        let tbl2_data = concat_batches(&schema, tbl2_data.iter()).unwrap();
+        assert_eq!(tbl2_data, one_batch);
+    }
+
    #[tokio::test]
    async fn drop_table() {
        let tc = new_test_connection().await.unwrap();
@@ -1143,6 +1640,41 @@ mod tests {
        assert_eq!(tables.len(), 0);
    }

+    #[tokio::test]
+    async fn test_create_table_already_exists() {
+        let tmp_dir = tempdir().unwrap();
+        let uri = tmp_dir.path().to_str().unwrap();
+        let db = connect(uri).execute().await.unwrap();
+        let schema = Arc::new(Schema::new(vec![Field::new("x", DataType::Int32, false)]));
+        db.create_empty_table("test", schema.clone())
+            .execute()
+            .await
+            .unwrap();
+        // TODO: None of the open table options are "inspectable" right now but once one is we
+        // should assert we are passing these options in correctly
+        db.create_empty_table("test", schema)
+            .mode(CreateTableMode::exist_ok(|mut req| {
+                req.index_cache_size = Some(16);
+                req
+            }))
+            .execute()
+            .await
+            .unwrap();
+        let other_schema = Arc::new(Schema::new(vec![Field::new("y", DataType::Int32, false)]));
+        assert!(db
+            .create_empty_table("test", other_schema.clone())
+            .execute()
+            .await
+            .is_err());
+        let overwritten = db
+            .create_empty_table("test", other_schema.clone())
+            .mode(CreateTableMode::Overwrite)
+            .execute()
+            .await
+            .unwrap();
+        assert_eq!(other_schema, overwritten.schema().await.unwrap());
+    }
+
    #[tokio::test]
    async fn test_clone_table() {
        let tmp_dir = tempdir().unwrap();
@@ -1153,8 +1685,7 @@ mod tests {
        let mut batch_gen = BatchGenerator::new()
            .col(Box::new(IncrementingInt32::new().named("id")))
            .col(Box::new(IncrementingInt32::new().named("value")));
-        let reader: Box<dyn arrow_array::RecordBatchReader + Send> =
-            Box::new(batch_gen.batches(5, 100));
+        let reader = batch_gen.batches(5, 100);

        let source_table = db
            .create_table("source_table", reader)
@@ -1189,4 +1720,128 @@ mod tests {
        let cloned_count = cloned_table.count_rows(None).await.unwrap();
        assert_eq!(source_count, cloned_count);
    }
+
+    #[tokio::test]
+    async fn test_create_empty_table_with_embeddings() {
+        use crate::embeddings::{EmbeddingDefinition, EmbeddingFunction};
+        use arrow_array::{
+            Array, FixedSizeListArray, Float32Array, RecordBatch, RecordBatchIterator, StringArray,
+        };
+        use std::borrow::Cow;
+
+        #[derive(Debug, Clone)]
+        struct MockEmbedding {
+            dim: usize,
+        }
+
+        impl EmbeddingFunction for MockEmbedding {
+            fn name(&self) -> &str {
+                "test_embedding"
+            }
+
+            fn source_type(&self) -> Result<Cow<'_, DataType>> {
+                Ok(Cow::Owned(DataType::Utf8))
+            }
+
+            fn dest_type(&self) -> Result<Cow<'_, DataType>> {
+                Ok(Cow::Owned(DataType::new_fixed_size_list(
+                    DataType::Float32,
+                    self.dim as i32,
+                    true,
+                )))
+            }
+
+            fn compute_source_embeddings(&self, source: Arc<dyn Array>) -> Result<Arc<dyn Array>> {
+                let len = source.len();
+                let values = vec![1.0f32; len * self.dim];
+                let values = Arc::new(Float32Array::from(values));
+                let field = Arc::new(Field::new("item", DataType::Float32, true));
+                Ok(Arc::new(FixedSizeListArray::new(
+                    field,
+                    self.dim as i32,
+                    values,
+                    None,
+                )))
+            }
+
+            fn compute_query_embeddings(&self, _input: Arc<dyn Array>) -> Result<Arc<dyn Array>> {
+                unimplemented!()
+            }
+        }
+
+        let tmp_dir = tempdir().unwrap();
+        let uri = tmp_dir.path().to_str().unwrap();
+        let db = connect(uri).execute().await.unwrap();
+
+        let embed_func = Arc::new(MockEmbedding { dim: 128 });
+        db.embedding_registry()
+            .register("test_embedding", embed_func.clone())
+            .unwrap();
+
+        let schema = Arc::new(Schema::new(vec![Field::new("name", DataType::Utf8, true)]));
+        let ed = EmbeddingDefinition {
+            source_column: "name".to_owned(),
+            dest_column: Some("name_embedding".to_owned()),
+            embedding_name: "test_embedding".to_owned(),
+        };
+
+        let table = db
+            .create_empty_table("test", schema)
+            .mode(CreateTableMode::Overwrite)
+            .add_embedding(ed)
+            .unwrap()
+            .execute()
+            .await
+            .unwrap();
+
+        let table_schema = table.schema().await.unwrap();
+        assert!(table_schema.column_with_name("name").is_some());
+        assert!(table_schema.column_with_name("name_embedding").is_some());
+
+        let embedding_field = table_schema.field_with_name("name_embedding").unwrap();
+        assert_eq!(
+            embedding_field.data_type(),
+            &DataType::new_fixed_size_list(DataType::Float32, 128, true)
+        );
+
+        let input_schema = Arc::new(Schema::new(vec![Field::new("name", DataType::Utf8, true)]));
+        let input_batch = RecordBatch::try_new(
+            input_schema.clone(),
+            vec![Arc::new(StringArray::from(vec![
+                Some("Alice"),
+                Some("Bob"),
+                Some("Charlie"),
+            ]))],
+        )
+        .unwrap();
+
+        let input_reader = Box::new(RecordBatchIterator::new(
+            vec![Ok(input_batch)].into_iter(),
+            input_schema,
+        ));
+
+        table.add(input_reader).execute().await.unwrap();
+
+        let results = table
+            .query()
+            .execute()
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+
+        assert_eq!(results.len(), 1);
+        let batch = &results[0];
+        assert_eq!(batch.num_rows(), 3);
+        assert!(batch.column_by_name("name_embedding").is_some());
+
+        let embedding_col = batch
+            .column_by_name("name_embedding")
+            .unwrap()
+            .as_any()
+            .downcast_ref::<FixedSizeListArray>()
+            .unwrap();
+        assert_eq!(embedding_col.len(), 3);
+    }
 }
--- a/rust/lancedb/src/connection/create_table.rs
+++ b/rust/lancedb/src/connection/create_table.rs
@@ -1,612 +0,0 @@
-// SPDX-License-Identifier: Apache-2.0
-// SPDX-FileCopyrightText: Copyright The LanceDB Authors
-
-use std::sync::Arc;
-
-use lance_io::object_store::StorageOptionsProvider;
-
-use crate::{
-    connection::{merge_storage_options, set_storage_options_provider},
-    data::scannable::{Scannable, WithEmbeddingsScannable},
-    database::{CreateTableMode, CreateTableRequest, Database},
-    embeddings::{EmbeddingDefinition, EmbeddingFunction, EmbeddingRegistry},
-    table::WriteOptions,
-    Error, Result, Table,
-};
-
-pub struct CreateTableBuilder {
-    parent: Arc<dyn Database>,
-    embeddings: Vec<(EmbeddingDefinition, Arc<dyn EmbeddingFunction>)>,
-    embedding_registry: Arc<dyn EmbeddingRegistry>,
-    request: CreateTableRequest,
-}
-
-impl CreateTableBuilder {
-    pub(super) fn new(
-        parent: Arc<dyn Database>,
-        embedding_registry: Arc<dyn EmbeddingRegistry>,
-        name: String,
-        data: Box<dyn Scannable>,
-    ) -> Self {
-        Self {
-            parent,
-            embeddings: Vec::new(),
-            embedding_registry,
-            request: CreateTableRequest::new(name, data),
-        }
-    }
-
-    /// Set the mode for creating the table
-    ///
-    /// This controls what happens if a table with the given name already exists
-    pub fn mode(mut self, mode: CreateTableMode) -> Self {
-        self.request.mode = mode;
-        self
-    }
-
-    /// Apply the given write options when writing the initial data
-    pub fn write_options(mut self, write_options: WriteOptions) -> Self {
-        self.request.write_options = write_options;
-        self
-    }
-
-    /// Set an option for the storage layer.
-    ///
-    /// Options already set on the connection will be inherited by the table,
-    /// but can be overridden here.
-    ///
-    /// See available options at <https://lancedb.com/docs/storage/>
-    pub fn storage_option(mut self, key: impl Into<String>, value: impl Into<String>) -> Self {
-        let store_params = self
-            .request
-            .write_options
-            .lance_write_params
-            .get_or_insert(Default::default())
-            .store_params
-            .get_or_insert(Default::default());
-        merge_storage_options(store_params, [(key.into(), value.into())]);
-        self
-    }
-
-    /// Set multiple options for the storage layer.
-    ///
-    /// Options already set on the connection will be inherited by the table,
-    /// but can be overridden here.
-    ///
-    /// See available options at <https://lancedb.com/docs/storage/>
-    pub fn storage_options(
-        mut self,
-        pairs: impl IntoIterator<Item = (impl Into<String>, impl Into<String>)>,
-    ) -> Self {
-        let store_params = self
-            .request
-            .write_options
-            .lance_write_params
-            .get_or_insert(Default::default())
-            .store_params
-            .get_or_insert(Default::default());
-        let updates = pairs
-            .into_iter()
-            .map(|(key, value)| (key.into(), value.into()));
-        merge_storage_options(store_params, updates);
-        self
-    }
-
-    /// Add an embedding definition to the table.
-    ///
-    /// The `embedding_name` must match the name of an embedding function that
-    /// was previously registered with the connection's [`EmbeddingRegistry`].
-    pub fn add_embedding(mut self, definition: EmbeddingDefinition) -> Result<Self> {
-        // Early verification of the embedding name
-        let embedding_func = self
-            .embedding_registry
-            .get(&definition.embedding_name)
-            .ok_or_else(|| Error::EmbeddingFunctionNotFound {
-                name: definition.embedding_name.clone(),
-                reason: "No embedding function found in the connection's embedding_registry"
-                    .to_string(),
-            })?;
-
-        self.embeddings.push((definition, embedding_func));
-        Ok(self)
-    }
-
-    /// Set the namespace for the table
-    pub fn namespace(mut self, namespace: Vec<String>) -> Self {
-        self.request.namespace = namespace;
-        self
-    }
-
-    /// Set a custom location for the table.
-    ///
-    /// If not set, the database will derive a location from its URI and the table name.
-    /// This is useful when integrating with namespace systems that manage table locations.
-    pub fn location(mut self, location: impl Into<String>) -> Self {
-        self.request.location = Some(location.into());
-        self
-    }
-
-    /// Set a storage options provider for automatic credential refresh.
-    ///
-    /// This allows tables to automatically refresh cloud storage credentials
-    /// when they expire, enabling long-running operations on remote storage.
-    pub fn storage_options_provider(mut self, provider: Arc<dyn StorageOptionsProvider>) -> Self {
-        let store_params = self
-            .request
-            .write_options
-            .lance_write_params
-            .get_or_insert(Default::default())
-            .store_params
-            .get_or_insert(Default::default());
-        set_storage_options_provider(store_params, provider);
-        self
-    }
-
-    /// Execute the create table operation
-    pub async fn execute(mut self) -> Result<Table> {
-        let embedding_registry = self.embedding_registry.clone();
-        let parent = self.parent.clone();
-
-        // If embeddings were configured via add_embedding(), wrap the data
-        if !self.embeddings.is_empty() {
-            let wrapped_data: Box<dyn Scannable> = Box::new(WithEmbeddingsScannable::try_new(
-                self.request.data,
-                self.embeddings,
-            )?);
-            self.request.data = wrapped_data;
-        }
-
-        Ok(Table::new_with_embedding_registry(
-            parent.create_table(self.request).await?,
-            parent,
-            embedding_registry,
-        ))
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use arrow_array::{
-        record_batch, Array, FixedSizeListArray, Float32Array, RecordBatch, RecordBatchIterator,
-    };
-    use arrow_schema::{ArrowError, DataType, Field, Schema};
-    use futures::TryStreamExt;
-    use lance_file::version::LanceFileVersion;
-    use tempfile::tempdir;
-
-    use crate::{
-        arrow::{SendableRecordBatchStream, SimpleRecordBatchStream},
-        connect,
-        database::listing::{ListingDatabaseOptions, NewTableConfig},
-        embeddings::{EmbeddingDefinition, EmbeddingFunction, MemoryRegistry},
-        query::{ExecutableQuery, QueryBase, Select},
-        test_utils::embeddings::MockEmbed,
-    };
-    use std::borrow::Cow;
-
-    use super::*;
-
-    #[tokio::test]
-    async fn create_empty_table() {
-        let db = connect("memory://").execute().await.unwrap();
-        let schema = Arc::new(Schema::new(vec![
-            Field::new("id", DataType::Int64, false),
-            Field::new("value", DataType::Float64, false),
-        ]));
-        db.create_empty_table("name", schema.clone())
-            .execute()
-            .await
-            .unwrap();
-        let table = db.open_table("name").execute().await.unwrap();
-        assert_eq!(table.schema().await.unwrap(), schema);
-        assert_eq!(table.count_rows(None).await.unwrap(), 0);
-    }
-
-    async fn test_create_table_with_data<T>(data: T)
-    where
-        T: Scannable + 'static,
-    {
-        let db = connect("memory://").execute().await.unwrap();
-        let schema = data.schema();
-        db.create_table("data_table", data).execute().await.unwrap();
-        let table = db.open_table("data_table").execute().await.unwrap();
-        assert_eq!(table.count_rows(None).await.unwrap(), 3);
-        assert_eq!(table.schema().await.unwrap(), schema);
-    }
-
-    #[tokio::test]
-    async fn create_table_with_batch() {
-        let batch = record_batch!(("id", Int64, [1, 2, 3])).unwrap();
-        test_create_table_with_data(batch).await;
-    }
-
-    #[tokio::test]
-    async fn test_create_table_with_vec_batch() {
-        let data = vec![
-            record_batch!(("id", Int64, [1, 2])).unwrap(),
-            record_batch!(("id", Int64, [3])).unwrap(),
-        ];
-        test_create_table_with_data(data).await;
-    }
-
-    #[tokio::test]
-    async fn test_create_table_with_record_batch_reader() {
-        let data = vec![
-            record_batch!(("id", Int64, [1, 2])).unwrap(),
-            record_batch!(("id", Int64, [3])).unwrap(),
-        ];
-        let schema = data[0].schema();
-        let reader: Box<dyn arrow_array::RecordBatchReader + Send> = Box::new(
-            RecordBatchIterator::new(data.into_iter().map(Ok), schema.clone()),
-        );
-        test_create_table_with_data(reader).await;
-    }
-
-    #[tokio::test]
-    async fn test_create_table_with_stream() {
-        let data = vec![
-            record_batch!(("id", Int64, [1, 2])).unwrap(),
-            record_batch!(("id", Int64, [3])).unwrap(),
-        ];
-        let schema = data[0].schema();
-        let inner = futures::stream::iter(data.into_iter().map(Ok));
-        let stream: SendableRecordBatchStream = Box::pin(SimpleRecordBatchStream {
-            schema,
-            stream: inner,
-        });
-        test_create_table_with_data(stream).await;
-    }
-
-    #[derive(Debug)]
-    struct MyError;
-
-    impl std::fmt::Display for MyError {
-        fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-            write!(f, "MyError occurred")
-        }
-    }
-
-    impl std::error::Error for MyError {}
-
-    #[tokio::test]
-    async fn test_create_preserves_reader_error() {
-        let first_batch = record_batch!(("id", Int64, [1, 2])).unwrap();
-        let schema = first_batch.schema();
-        let iterator = vec![
-            Ok(first_batch),
-            Err(ArrowError::ExternalError(Box::new(MyError))),
-        ];
-        let reader: Box<dyn arrow_array::RecordBatchReader + Send> = Box::new(
-            RecordBatchIterator::new(iterator.into_iter(), schema.clone()),
-        );
-
-        let db = connect("memory://").execute().await.unwrap();
-        let result = db.create_table("failing_table", reader).execute().await;
-
-        assert!(result.is_err());
-        // TODO: when we upgrade to Lance 2.0.0, this should pass
-        // assert!(matches!(result, Err(Error::External { source})
-        //     if source.downcast_ref::<MyError>().is_some()
-        // ));
-    }
-
-    #[tokio::test]
-    async fn test_create_preserves_stream_error() {
-        let first_batch = record_batch!(("id", Int64, [1, 2])).unwrap();
-        let schema = first_batch.schema();
-        let iterator = vec![
-            Ok(first_batch),
-            Err(Error::External {
-                source: Box::new(MyError),
-            }),
-        ];
-        let stream = futures::stream::iter(iterator);
-        let stream: SendableRecordBatchStream = Box::pin(SimpleRecordBatchStream {
-            schema: schema.clone(),
-            stream,
-        });
-
-        let db = connect("memory://").execute().await.unwrap();
-        let result = db
-            .create_table("failing_stream_table", stream)
-            .execute()
-            .await;
-
-        assert!(result.is_err());
-        // TODO: when we upgrade to Lance 2.0.0, this should pass
-        // assert!(matches!(result, Err(Error::External { source})
-        //     if source.downcast_ref::<MyError>().is_some()
-        // ));
-    }
-
-    #[tokio::test]
-    #[allow(deprecated)]
-    async fn test_create_table_with_storage_options() {
-        let batch = record_batch!(("id", Int64, [1, 2, 3])).unwrap();
-        let db = connect("memory://").execute().await.unwrap();
-
-        let table = db
-            .create_table("options_table", batch)
-            .storage_option("timeout", "30s")
-            .storage_options([("retry_count", "3")])
-            .execute()
-            .await
-            .unwrap();
-
-        let final_options = table.storage_options().await.unwrap();
-        assert_eq!(final_options.get("timeout"), Some(&"30s".to_string()));
-        assert_eq!(final_options.get("retry_count"), Some(&"3".to_string()));
-    }
-
-    #[tokio::test]
-    async fn test_create_table_unregistered_embedding() {
-        let db = connect("memory://").execute().await.unwrap();
-        let batch = record_batch!(("text", Utf8, ["hello", "world"])).unwrap();
-
-        // Try to add an embedding that doesn't exist in the registry
-        let result = db
-            .create_table("embed_table", batch)
-            .add_embedding(EmbeddingDefinition::new(
-                "text",
-                "nonexistent_embedding_function",
-                None::<&str>,
-            ));
-
-        match result {
-            Err(Error::EmbeddingFunctionNotFound { name, .. }) => {
-                assert_eq!(name, "nonexistent_embedding_function");
-            }
-            Err(other) => panic!("Expected EmbeddingFunctionNotFound error, got: {:?}", other),
-            Ok(_) => panic!("Expected error, but got Ok"),
-        }
-    }
-
-    #[tokio::test]
-    async fn test_create_table_already_exists() {
-        let tmp_dir = tempdir().unwrap();
-        let uri = tmp_dir.path().to_str().unwrap();
-        let db = connect(uri).execute().await.unwrap();
-        let schema = Arc::new(Schema::new(vec![Field::new("x", DataType::Int32, false)]));
-        db.create_empty_table("test", schema.clone())
-            .execute()
-            .await
-            .unwrap();
-        db.create_empty_table("test", schema)
-            .mode(CreateTableMode::exist_ok(|mut req| {
-                req.index_cache_size = Some(16);
-                req
-            }))
-            .execute()
-            .await
-            .unwrap();
-        let other_schema = Arc::new(Schema::new(vec![Field::new("y", DataType::Int32, false)]));
-        assert!(db
-            .create_empty_table("test", other_schema.clone())
-            .execute()
-            .await
-            .is_err()); // TODO: assert what this error is
-        let overwritten = db
-            .create_empty_table("test", other_schema.clone())
-            .mode(CreateTableMode::Overwrite)
-            .execute()
-            .await
-            .unwrap();
-        assert_eq!(other_schema, overwritten.schema().await.unwrap());
-    }
-
-    #[tokio::test]
-    #[rstest::rstest]
-    #[case(LanceFileVersion::Legacy)]
-    #[case(LanceFileVersion::Stable)]
-    async fn test_create_table_with_storage_version(
-        #[case] data_storage_version: LanceFileVersion,
-    ) {
-        let db = connect("memory://")
-            .database_options(&ListingDatabaseOptions {
-                new_table_config: NewTableConfig {
-                    data_storage_version: Some(data_storage_version),
-                    ..Default::default()
-                },
-                ..Default::default()
-            })
-            .execute()
-            .await
-            .unwrap();
-
-        let batch = record_batch!(("id", Int64, [1, 2, 3])).unwrap();
-        let table = db
-            .create_table("legacy_table", batch)
-            .execute()
-            .await
-            .unwrap();
-
-        let native_table = table.as_native().unwrap();
-        let storage_format = native_table
-            .manifest()
-            .await
-            .unwrap()
-            .data_storage_format
-            .lance_file_version()
-            .unwrap();
-        // Compare resolved versions since Stable/Next are aliases that resolve at storage time
-        assert_eq!(storage_format.resolve(), data_storage_version.resolve());
-    }
-
-    #[tokio::test]
-    async fn test_create_table_with_embedding() {
-        // Register the mock embedding function
-        let registry = Arc::new(MemoryRegistry::new());
-        let mock_embedding: Arc<dyn EmbeddingFunction> = Arc::new(MockEmbed::new("mock", 4));
-        registry.register("mock", mock_embedding).unwrap();
-
-        // Connect with the custom registry
-        let conn = connect("memory://")
-            .embedding_registry(registry)
-            .execute()
-            .await
-            .unwrap();
-
-        // Create data without the embedding column
-        let batch = record_batch!(("text", Utf8, ["hello", "world", "test"])).unwrap();
-
-        // Create table with add_embedding - embeddings should be computed automatically
-        let table = conn
-            .create_table("embed_test", batch)
-            .add_embedding(EmbeddingDefinition::new(
-                "text",
-                "mock",
-                Some("text_embedding"),
-            ))
-            .unwrap()
-            .execute()
-            .await
-            .unwrap();
-
-        // Verify row count
-        assert_eq!(table.count_rows(None).await.unwrap(), 3);
-
-        // Verify the schema includes the embedding column
-        let result_schema = table.schema().await.unwrap();
-        assert_eq!(result_schema.fields().len(), 2);
-        assert_eq!(result_schema.field(0).name(), "text");
-        assert_eq!(result_schema.field(1).name(), "text_embedding");
-
-        // Verify the embedding column has the correct type
-        assert!(matches!(
-            result_schema.field(1).data_type(),
-            DataType::FixedSizeList(_, 4)
-        ));
-
-        // Query to verify the embeddings were computed
-        let results: Vec<RecordBatch> = table
-            .query()
-            .select(Select::columns(&["text", "text_embedding"]))
-            .execute()
-            .await
-            .unwrap()
-            .try_collect()
-            .await
-            .unwrap();
-
-        let total_rows: usize = results.iter().map(|b| b.num_rows()).sum();
-        assert_eq!(total_rows, 3);
-
-        // Check that all rows have embedding values (not null)
-        for batch in &results {
-            let embedding_col = batch.column(1);
-            assert_eq!(embedding_col.null_count(), 0);
-            assert_eq!(embedding_col.len(), batch.num_rows());
-        }
-
-        // Verify the schema metadata contains the column definitions
-        assert!(
-            result_schema
-                .metadata
-                .contains_key("lancedb::column_definitions"),
-            "Schema metadata should contain column definitions"
-        );
-    }
-
-    #[tokio::test]
-    async fn test_create_empty_table_with_embeddings() {
-        #[derive(Debug, Clone)]
-        struct MockEmbedding {
-            dim: usize,
-        }
-
-        impl EmbeddingFunction for MockEmbedding {
-            fn name(&self) -> &str {
-                "test_embedding"
-            }
-
-            fn source_type(&self) -> Result<Cow<'_, DataType>> {
-                Ok(Cow::Owned(DataType::Utf8))
-            }
-
-            fn dest_type(&self) -> Result<Cow<'_, DataType>> {
-                Ok(Cow::Owned(DataType::new_fixed_size_list(
-                    DataType::Float32,
-                    self.dim as i32,
-                    true,
-                )))
-            }
-
-            fn compute_source_embeddings(&self, source: Arc<dyn Array>) -> Result<Arc<dyn Array>> {
-                let len = source.len();
-                let values = vec![1.0f32; len * self.dim];
-                let values = Arc::new(Float32Array::from(values));
-                let field = Arc::new(Field::new("item", DataType::Float32, true));
-                Ok(Arc::new(FixedSizeListArray::new(
-                    field,
-                    self.dim as i32,
-                    values,
-                    None,
-                )))
-            }
-
-            fn compute_query_embeddings(&self, _input: Arc<dyn Array>) -> Result<Arc<dyn Array>> {
-                unimplemented!()
-            }
-        }
-
-        let tmp_dir = tempdir().unwrap();
-        let uri = tmp_dir.path().to_str().unwrap();
-        let db = connect(uri).execute().await.unwrap();
-
-        let embed_func = Arc::new(MockEmbedding { dim: 128 });
-        db.embedding_registry()
-            .register("test_embedding", embed_func.clone())
-            .unwrap();
-
-        let schema = Arc::new(Schema::new(vec![Field::new("name", DataType::Utf8, true)]));
-        let ed = EmbeddingDefinition {
-            source_column: "name".to_owned(),
-            dest_column: Some("name_embedding".to_owned()),
-            embedding_name: "test_embedding".to_owned(),
-        };
-
-        let table = db
-            .create_empty_table("test", schema)
-            .mode(CreateTableMode::Overwrite)
-            .add_embedding(ed)
-            .unwrap()
-            .execute()
-            .await
-            .unwrap();
-
-        let table_schema = table.schema().await.unwrap();
-        assert!(table_schema.column_with_name("name").is_some());
-        assert!(table_schema.column_with_name("name_embedding").is_some());
-
-        let embedding_field = table_schema.field_with_name("name_embedding").unwrap();
-        assert_eq!(
-            embedding_field.data_type(),
-            &DataType::new_fixed_size_list(DataType::Float32, 128, true)
-        );
-
-        let input_batch = record_batch!(("name", Utf8, ["Alice", "Bob", "Charlie"])).unwrap();
-        table.add(input_batch).execute().await.unwrap();
-
-        let results = table
-            .query()
-            .execute()
-            .await
-            .unwrap()
-            .try_collect::<Vec<_>>()
-            .await
-            .unwrap();
-
-        assert_eq!(results.len(), 1);
-        let batch = &results[0];
-        assert_eq!(batch.num_rows(), 3);
-        assert!(batch.column_by_name("name_embedding").is_some());
-
-        let embedding_col = batch
-            .column_by_name("name_embedding")
-            .unwrap()
-            .as_any()
-            .downcast_ref::<FixedSizeListArray>()
-            .unwrap();
-        assert_eq!(embedding_col.len(), 3);
-    }
-}
--- a/rust/lancedb/src/data.rs
+++ b/rust/lancedb/src/data.rs
@@ -5,4 +5,3 @@

 pub mod inspect;
 pub mod sanitize;
-pub mod scannable;
--- a/rust/lancedb/src/data/scannable.rs
+++ b/rust/lancedb/src/data/scannable.rs
@@ -1,580 +0,0 @@
-// SPDX-License-Identifier: Apache-2.0
-// SPDX-FileCopyrightText: Copyright The LanceDB Authors
-
-//! Data source abstraction for LanceDB.
-//!
-//! This module provides a [`Scannable`] trait that allows input data sources to express
-//! capabilities (row count, rescannability) so the insert pipeline can make
-//! better decisions about write parallelism and retry strategies.
-
-use std::sync::Arc;
-
-use arrow_array::{RecordBatch, RecordBatchIterator, RecordBatchReader};
-use arrow_schema::{ArrowError, SchemaRef};
-use async_trait::async_trait;
-use futures::stream::once;
-use futures::StreamExt;
-use lance_datafusion::utils::StreamingWriteSource;
-
-use crate::arrow::{
-    SendableRecordBatchStream, SendableRecordBatchStreamExt, SimpleRecordBatchStream,
-};
-use crate::embeddings::{
-    compute_embeddings_for_batch, compute_output_schema, EmbeddingDefinition, EmbeddingFunction,
-    EmbeddingRegistry,
-};
-use crate::table::{ColumnDefinition, ColumnKind, TableDefinition};
-use crate::{Error, Result};
-
-pub trait Scannable: Send {
-    /// Returns the schema of the data.
-    fn schema(&self) -> SchemaRef;
-
-    /// Read data as a stream of record batches.
-    ///
-    /// For rescannable sources (in-memory data like RecordBatch, Vec<RecordBatch>),
-    /// this can be called multiple times and returns cloned data each time.
-    ///
-    /// For non-rescannable sources (streams, readers), this can only be called once.
-    /// Calling it a second time returns a stream whose first item is an error.
-    fn scan_as_stream(&mut self) -> SendableRecordBatchStream;
-
-    /// Optional hint about the number of rows.
-    ///
-    /// When available, this allows the pipeline to estimate total data size
-    /// and choose appropriate partitioning.
-    fn num_rows(&self) -> Option<usize> {
-        None
-    }
-
-    /// Whether the source can be re-read from the beginning.
-    ///
-    /// `true` for in-memory data (Tables, DataFrames) and disk-based sources (Datasets).
-    /// `false` for streaming sources (DuckDB results, network streams).
-    ///
-    /// When true, the pipeline can retry failed writes by rescanning.
-    fn rescannable(&self) -> bool {
-        false
-    }
-}
-
-impl std::fmt::Debug for dyn Scannable {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        f.debug_struct("Scannable")
-            .field("schema", &self.schema())
-            .field("num_rows", &self.num_rows())
-            .field("rescannable", &self.rescannable())
-            .finish()
-    }
-}
-
-impl Scannable for RecordBatch {
-    fn schema(&self) -> SchemaRef {
-        Self::schema(self)
-    }
-
-    fn scan_as_stream(&mut self) -> SendableRecordBatchStream {
-        let batch = self.clone();
-        let schema = batch.schema();
-        Box::pin(SimpleRecordBatchStream {
-            schema,
-            stream: once(async move { Ok(batch) }),
-        })
-    }
-
-    fn num_rows(&self) -> Option<usize> {
-        Some(Self::num_rows(self))
-    }
-
-    fn rescannable(&self) -> bool {
-        true
-    }
-}
-
-impl Scannable for Vec<RecordBatch> {
-    fn schema(&self) -> SchemaRef {
-        if self.is_empty() {
-            Arc::new(arrow_schema::Schema::empty())
-        } else {
-            self[0].schema()
-        }
-    }
-
-    fn scan_as_stream(&mut self) -> SendableRecordBatchStream {
-        if self.is_empty() {
-            let schema = Scannable::schema(self);
-            return Box::pin(SimpleRecordBatchStream {
-                schema,
-                stream: once(async {
-                    Err(Error::InvalidInput {
-                        message: "Cannot scan an empty Vec<RecordBatch>".to_string(),
-                    })
-                }),
-            });
-        }
-        let schema = Scannable::schema(self);
-        let batches = self.clone();
-        let stream = futures::stream::iter(batches.into_iter().map(Ok));
-        Box::pin(SimpleRecordBatchStream { schema, stream })
-    }
-
-    fn num_rows(&self) -> Option<usize> {
-        Some(self.iter().map(|b| b.num_rows()).sum())
-    }
-
-    fn rescannable(&self) -> bool {
-        true
-    }
-}
-
-impl Scannable for Box<dyn RecordBatchReader + Send> {
-    fn schema(&self) -> SchemaRef {
-        RecordBatchReader::schema(self.as_ref())
-    }
-
-    fn scan_as_stream(&mut self) -> SendableRecordBatchStream {
-        let schema = Scannable::schema(self);
-
-        // Swap self with a reader that errors on iteration, so a second call
-        // produces a clear error instead of silently returning empty data.
-        let err_reader: Box<dyn RecordBatchReader + Send> = Box::new(RecordBatchIterator::new(
-            vec![Err(ArrowError::InvalidArgumentError(
-                "Reader has already been consumed".into(),
-            ))],
-            schema.clone(),
-        ));
-        let reader = std::mem::replace(self, err_reader);
-
-        // Bridge the blocking RecordBatchReader to an async stream via a channel.
-        let (tx, rx) = tokio::sync::mpsc::channel::<crate::Result<RecordBatch>>(2);
-        tokio::task::spawn_blocking(move || {
-            for batch_result in reader {
-                let result = batch_result.map_err(Into::into);
-                if tx.blocking_send(result).is_err() {
-                    break;
-                }
-            }
-        });
-
-        let stream = futures::stream::unfold(rx, |mut rx| async move {
-            rx.recv().await.map(|batch| (batch, rx))
-        })
-        .fuse();
-
-        Box::pin(SimpleRecordBatchStream { schema, stream })
-    }
-}
-
-impl Scannable for SendableRecordBatchStream {
-    fn schema(&self) -> SchemaRef {
-        self.as_ref().schema()
-    }
-
-    fn scan_as_stream(&mut self) -> SendableRecordBatchStream {
-        let schema = Scannable::schema(self);
-
-        // Swap self with an error stream so a second call produces a clear error.
-        let error_stream = Box::pin(SimpleRecordBatchStream {
-            schema: schema.clone(),
-            stream: once(async {
-                Err(Error::InvalidInput {
-                    message: "Stream has already been consumed".to_string(),
-                })
-            }),
-        });
-        std::mem::replace(self, error_stream)
-    }
-}
-
-#[async_trait]
-impl StreamingWriteSource for Box<dyn Scannable> {
-    fn arrow_schema(&self) -> SchemaRef {
-        self.schema()
-    }
-
-    fn into_stream(mut self) -> datafusion_physical_plan::SendableRecordBatchStream {
-        self.scan_as_stream().into_df_stream()
-    }
-}
-
-/// A scannable that applies embeddings to the stream.
-pub struct WithEmbeddingsScannable {
-    inner: Box<dyn Scannable>,
-    embeddings: Vec<(EmbeddingDefinition, Arc<dyn EmbeddingFunction>)>,
-    output_schema: SchemaRef,
-}
-
-impl WithEmbeddingsScannable {
-    /// Create a new WithEmbeddingsScannable.
-    ///
-    /// The embeddings are applied to the inner scannable's data as new columns.
-    pub fn try_new(
-        inner: Box<dyn Scannable>,
-        embeddings: Vec<(EmbeddingDefinition, Arc<dyn EmbeddingFunction>)>,
-    ) -> Result<Self> {
-        let output_schema = compute_output_schema(&inner.schema(), &embeddings)?;
-
-        // Build column definitions: Physical for base columns, Embedding for new ones
-        let base_col_count = inner.schema().fields().len();
-        let column_definitions: Vec<ColumnDefinition> = (0..base_col_count)
-            .map(|_| ColumnDefinition {
-                kind: ColumnKind::Physical,
-            })
-            .chain(embeddings.iter().map(|(ed, _)| ColumnDefinition {
-                kind: ColumnKind::Embedding(ed.clone()),
-            }))
-            .collect();
-
-        let table_definition = TableDefinition::new(output_schema, column_definitions);
-        let output_schema = table_definition.into_rich_schema();
-
-        Ok(Self {
-            inner,
-            embeddings,
-            output_schema,
-        })
-    }
-}
-
-impl Scannable for WithEmbeddingsScannable {
-    fn schema(&self) -> SchemaRef {
-        self.output_schema.clone()
-    }
-
-    fn scan_as_stream(&mut self) -> SendableRecordBatchStream {
-        let inner_stream = self.inner.scan_as_stream();
-        let embeddings = self.embeddings.clone();
-        let output_schema = self.output_schema.clone();
-
-        let mapped_stream = inner_stream.then(move |batch_result| {
-            let embeddings = embeddings.clone();
-            async move {
-                let batch = batch_result?;
-                let result = tokio::task::spawn_blocking(move || {
-                    compute_embeddings_for_batch(batch, &embeddings)
-                })
-                .await
-                .map_err(|e| Error::Runtime {
-                    message: format!("Task panicked during embedding computation: {}", e),
-                })??;
-                Ok(result)
-            }
-        });
-
-        Box::pin(SimpleRecordBatchStream {
-            schema: output_schema,
-            stream: mapped_stream,
-        })
-    }
-
-    fn num_rows(&self) -> Option<usize> {
-        self.inner.num_rows()
-    }
-
-    fn rescannable(&self) -> bool {
-        self.inner.rescannable()
-    }
-}
-
-pub fn scannable_with_embeddings(
-    inner: Box<dyn Scannable>,
-    table_definition: &TableDefinition,
-    registry: Option<&Arc<dyn EmbeddingRegistry>>,
-) -> Result<Box<dyn Scannable>> {
-    if let Some(registry) = registry {
-        let mut embeddings = Vec::with_capacity(table_definition.column_definitions.len());
-        for cd in table_definition.column_definitions.iter() {
-            if let ColumnKind::Embedding(embedding_def) = &cd.kind {
-                match registry.get(&embedding_def.embedding_name) {
-                    Some(func) => {
-                        embeddings.push((embedding_def.clone(), func));
-                    }
-                    None => {
-                        return Err(Error::EmbeddingFunctionNotFound {
-                            name: embedding_def.embedding_name.clone(),
-                            reason: format!(
-                                "Table was defined with an embedding column `{}` but no embedding function was found with that name within the registry.",
-                                embedding_def.embedding_name
-                            ),
-                        });
-                    }
-                }
-            }
-        }
-
-        if !embeddings.is_empty() {
-            return Ok(Box::new(WithEmbeddingsScannable::try_new(
-                inner, embeddings,
-            )?));
-        }
-    }
-
-    Ok(inner)
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use arrow_array::record_batch;
-    use futures::TryStreamExt;
-
-    #[tokio::test]
-    async fn test_record_batch_rescannable() {
-        let mut batch = record_batch!(("id", Int64, [0, 1, 2])).unwrap();
-
-        let stream1 = batch.scan_as_stream();
-        let batches1: Vec<RecordBatch> = stream1.try_collect().await.unwrap();
-        assert_eq!(batches1.len(), 1);
-        assert_eq!(batches1[0], batch);
-
-        assert!(batch.rescannable());
-        let stream2 = batch.scan_as_stream();
-        let batches2: Vec<RecordBatch> = stream2.try_collect().await.unwrap();
-        assert_eq!(batches2.len(), 1);
-        assert_eq!(batches2[0], batch);
-    }
-
-    #[tokio::test]
-    async fn test_vec_batch_rescannable() {
-        let mut batches = vec![
-            record_batch!(("id", Int64, [0, 1])).unwrap(),
-            record_batch!(("id", Int64, [2, 3, 4])).unwrap(),
-        ];
-
-        let stream1 = batches.scan_as_stream();
-        let result1: Vec<RecordBatch> = stream1.try_collect().await.unwrap();
-        assert_eq!(result1.len(), 2);
-        assert_eq!(result1[0], batches[0]);
-        assert_eq!(result1[1], batches[1]);
-
-        assert!(batches.rescannable());
-        let stream2 = batches.scan_as_stream();
-        let result2: Vec<RecordBatch> = stream2.try_collect().await.unwrap();
-        assert_eq!(result2.len(), 2);
-        assert_eq!(result2[0], batches[0]);
-        assert_eq!(result2[1], batches[1]);
-    }
-
-    #[tokio::test]
-    async fn test_vec_batch_empty_errors() {
-        let mut empty: Vec<RecordBatch> = vec![];
-        let mut stream = empty.scan_as_stream();
-        let result = stream.next().await;
-        assert!(result.is_some());
-        assert!(result.unwrap().is_err());
-    }
-
-    #[tokio::test]
-    async fn test_reader_not_rescannable() {
-        let batch = record_batch!(("id", Int64, [0, 1, 2])).unwrap();
-        let schema = batch.schema();
-        let mut reader: Box<dyn arrow_array::RecordBatchReader + Send> = Box::new(
-            RecordBatchIterator::new(vec![Ok(batch.clone())], schema.clone()),
-        );
-
-        let stream1 = reader.scan_as_stream();
-        let result1: Vec<RecordBatch> = stream1.try_collect().await.unwrap();
-        assert_eq!(result1.len(), 1);
-        assert_eq!(result1[0], batch);
-
-        assert!(!reader.rescannable());
-        // Second call returns a stream whose first item is an error
-        let mut stream2 = reader.scan_as_stream();
-        let result2 = stream2.next().await;
-        assert!(result2.is_some());
-        assert!(result2.unwrap().is_err());
-    }
-
-    #[tokio::test]
-    async fn test_stream_not_rescannable() {
-        let batch = record_batch!(("id", Int64, [0, 1, 2])).unwrap();
-        let schema = batch.schema();
-        let inner_stream = futures::stream::iter(vec![Ok(batch.clone())]);
-        let mut stream: SendableRecordBatchStream = Box::pin(SimpleRecordBatchStream {
-            schema: schema.clone(),
-            stream: inner_stream,
-        });
-
-        let stream1 = stream.scan_as_stream();
-        let result1: Vec<RecordBatch> = stream1.try_collect().await.unwrap();
-        assert_eq!(result1.len(), 1);
-        assert_eq!(result1[0], batch);
-
-        assert!(!stream.rescannable());
-        // Second call returns a stream whose first item is an error
-        let mut stream2 = stream.scan_as_stream();
-        let result2 = stream2.next().await;
-        assert!(result2.is_some());
-        assert!(result2.unwrap().is_err());
-    }
-
-    mod embedding_tests {
-        use super::*;
-        use crate::embeddings::MemoryRegistry;
-        use crate::table::{ColumnDefinition, ColumnKind};
-        use crate::test_utils::embeddings::MockEmbed;
-        use arrow_array::Array as _;
-        use arrow_array::{ArrayRef, StringArray};
-        use arrow_schema::{DataType, Field, Schema};
-
-        #[tokio::test]
-        async fn test_with_embeddings_scannable() {
-            let schema = Arc::new(Schema::new(vec![Field::new("text", DataType::Utf8, false)]));
-            let text_array = StringArray::from(vec!["hello", "world", "test"]);
-            let batch =
-                RecordBatch::try_new(schema.clone(), vec![Arc::new(text_array) as ArrayRef])
-                    .unwrap();
-
-            let mock_embedding: Arc<dyn EmbeddingFunction> = Arc::new(MockEmbed::new("mock", 4));
-            let embedding_def = EmbeddingDefinition::new("text", "mock", Some("text_embedding"));
-
-            let mut scannable = WithEmbeddingsScannable::try_new(
-                Box::new(batch.clone()),
-                vec![(embedding_def, mock_embedding)],
-            )
-            .unwrap();
-
-            // Check that schema has the embedding column
-            let output_schema = scannable.schema();
-            assert_eq!(output_schema.fields().len(), 2);
-            assert_eq!(output_schema.field(0).name(), "text");
-            assert_eq!(output_schema.field(1).name(), "text_embedding");
-
-            // Check num_rows and rescannable are preserved
-            assert_eq!(scannable.num_rows(), Some(3));
-            assert!(scannable.rescannable());
-
-            // Read the data
-            let stream = scannable.scan_as_stream();
-            let results: Vec<RecordBatch> = stream.try_collect().await.unwrap();
-            assert_eq!(results.len(), 1);
-
-            let result_batch = &results[0];
-            assert_eq!(result_batch.num_rows(), 3);
-            assert_eq!(result_batch.num_columns(), 2);
-
-            // Verify the embedding column is present and has the right shape
-            let embedding_col = result_batch.column(1);
-            assert_eq!(embedding_col.len(), 3);
-        }
-
-        #[tokio::test]
-        async fn test_maybe_embedded_scannable_no_embeddings() {
-            let batch = record_batch!(("id", Int64, [1, 2, 3])).unwrap();
-
-            // Create a table definition with no embedding columns
-            let table_def = TableDefinition::new_from_schema(batch.schema());
-
-            // Even with a registry, if there are no embedding columns, it's a passthrough
-            let registry: Arc<dyn EmbeddingRegistry> = Arc::new(MemoryRegistry::new());
-            let mut scannable =
-                scannable_with_embeddings(Box::new(batch.clone()), &table_def, Some(&registry))
-                    .unwrap();
-
-            // Check that data passes through unchanged
-            let stream = scannable.scan_as_stream();
-            let results: Vec<RecordBatch> = stream.try_collect().await.unwrap();
-            assert_eq!(results.len(), 1);
-            assert_eq!(results[0], batch);
-        }
-
-        #[tokio::test]
-        async fn test_maybe_embedded_scannable_with_embeddings() {
-            let schema = Arc::new(Schema::new(vec![Field::new("text", DataType::Utf8, false)]));
-            let text_array = StringArray::from(vec!["hello", "world"]);
-            let batch =
-                RecordBatch::try_new(schema.clone(), vec![Arc::new(text_array) as ArrayRef])
-                    .unwrap();
-
-            // Create a table definition with an embedding column
-            let embedding_def = EmbeddingDefinition::new("text", "mock", Some("text_embedding"));
-            let embedding_schema = Arc::new(Schema::new(vec![
-                Field::new("text", DataType::Utf8, false),
-                Field::new(
-                    "text_embedding",
-                    DataType::FixedSizeList(
-                        Arc::new(Field::new("item", DataType::Float32, true)),
-                        4,
-                    ),
-                    false,
-                ),
-            ]));
-            let table_def = TableDefinition::new(
-                embedding_schema,
-                vec![
-                    ColumnDefinition {
-                        kind: ColumnKind::Physical,
-                    },
-                    ColumnDefinition {
-                        kind: ColumnKind::Embedding(embedding_def.clone()),
-                    },
-                ],
-            );
-
-            // Register the mock embedding function
-            let registry: Arc<dyn EmbeddingRegistry> = Arc::new(MemoryRegistry::new());
-            let mock_embedding: Arc<dyn EmbeddingFunction> = Arc::new(MockEmbed::new("mock", 4));
-            registry.register("mock", mock_embedding).unwrap();
-
-            let mut scannable =
-                scannable_with_embeddings(Box::new(batch), &table_def, Some(&registry)).unwrap();
-
-            // Read and verify the data has embeddings
-            let stream = scannable.scan_as_stream();
-            let results: Vec<RecordBatch> = stream.try_collect().await.unwrap();
-            assert_eq!(results.len(), 1);
-
-            let result_batch = &results[0];
-            assert_eq!(result_batch.num_columns(), 2);
-            assert_eq!(result_batch.schema().field(1).name(), "text_embedding");
-        }
-
-        #[tokio::test]
-        async fn test_maybe_embedded_scannable_missing_function() {
-            let schema = Arc::new(Schema::new(vec![Field::new("text", DataType::Utf8, false)]));
-            let text_array = StringArray::from(vec!["hello"]);
-            let batch =
-                RecordBatch::try_new(schema.clone(), vec![Arc::new(text_array) as ArrayRef])
-                    .unwrap();
-
-            // Create a table definition with an embedding column
-            let embedding_def =
-                EmbeddingDefinition::new("text", "nonexistent", Some("text_embedding"));
-            let embedding_schema = Arc::new(Schema::new(vec![
-                Field::new("text", DataType::Utf8, false),
-                Field::new(
-                    "text_embedding",
-                    DataType::FixedSizeList(
-                        Arc::new(Field::new("item", DataType::Float32, true)),
-                        4,
-                    ),
-                    false,
-                ),
-            ]));
-            let table_def = TableDefinition::new(
-                embedding_schema,
-                vec![
-                    ColumnDefinition {
-                        kind: ColumnKind::Physical,
-                    },
-                    ColumnDefinition {
-                        kind: ColumnKind::Embedding(embedding_def),
-                    },
-                ],
-            );
-
-            // Registry has no embedding functions registered
-            let registry: Arc<dyn EmbeddingRegistry> = Arc::new(MemoryRegistry::new());
-
-            let result = scannable_with_embeddings(Box::new(batch), &table_def, Some(&registry));
-
-            // Should fail because the embedding function is not found
-            assert!(result.is_err());
-            let err = result.err().unwrap();
-            assert!(
-                matches!(err, Error::EmbeddingFunctionNotFound { .. }),
-                "Expected EmbeddingFunctionNotFound"
-            );
-        }
-    }
-}
--- a/rust/lancedb/src/database.rs
+++ b/rust/lancedb/src/database.rs
@@ -18,7 +18,12 @@ use std::collections::HashMap;
 use std::sync::Arc;
 use std::time::Duration;

+use arrow_array::RecordBatchReader;
+use async_trait::async_trait;
+use datafusion_physical_plan::stream::RecordBatchStreamAdapter;
+use futures::stream;
 use lance::dataset::ReadParams;
+use lance_datafusion::utils::StreamingWriteSource;
 use lance_namespace::models::{
    CreateNamespaceRequest, CreateNamespaceResponse, DescribeNamespaceRequest,
    DescribeNamespaceResponse, DropNamespaceRequest, DropNamespaceResponse, ListNamespacesRequest,
@@ -26,9 +31,9 @@ use lance_namespace::models::{
 };
 use lance_namespace::LanceNamespace;

-use crate::data::scannable::Scannable;
+use crate::arrow::{SendableRecordBatchStream, SendableRecordBatchStreamExt};
 use crate::error::Result;
-use crate::table::{BaseTable, WriteOptions};
+use crate::table::{BaseTable, TableDefinition, WriteOptions};

 pub mod listing;
 pub mod namespace;
@@ -110,14 +115,51 @@ impl Default for CreateTableMode {
    }
 }

+/// The data to start a table or a schema to create an empty table
+pub enum CreateTableData {
+    /// Creates a table using an iterator of data, the schema will be obtained from the data
+    Data(Box<dyn RecordBatchReader + Send>),
+    /// Creates a table using a stream of data, the schema will be obtained from the data
+    StreamingData(SendableRecordBatchStream),
+    /// Creates an empty table, the definition / schema must be provided separately
+    Empty(TableDefinition),
+}
+
+impl CreateTableData {
+    pub fn schema(&self) -> Arc<arrow_schema::Schema> {
+        match self {
+            Self::Data(reader) => reader.schema(),
+            Self::StreamingData(stream) => stream.schema(),
+            Self::Empty(definition) => definition.schema.clone(),
+        }
+    }
+}
+
+#[async_trait]
+impl StreamingWriteSource for CreateTableData {
+    fn arrow_schema(&self) -> Arc<arrow_schema::Schema> {
+        self.schema()
+    }
+    fn into_stream(self) -> datafusion_physical_plan::SendableRecordBatchStream {
+        match self {
+            Self::Data(reader) => reader.into_stream(),
+            Self::StreamingData(stream) => stream.into_df_stream(),
+            Self::Empty(table_definition) => {
+                let schema = table_definition.schema.clone();
+                Box::pin(RecordBatchStreamAdapter::new(schema, stream::empty()))
+            }
+        }
+    }
+}
+
 /// A request to create a table
 pub struct CreateTableRequest {
    /// The name of the new table
    pub name: String,
    /// The namespace to create the table in. Empty list represents root namespace.
    pub namespace: Vec<String>,
-    /// Initial data to write to the table, can be empty.
-    pub data: Box<dyn Scannable>,
+    /// Initial data to write to the table, can be None to create an empty table
+    pub data: CreateTableData,
    /// The mode to use when creating the table
    pub mode: CreateTableMode,
    /// Options to use when writing data (only used if `data` is not None)
@@ -131,7 +173,7 @@ pub struct CreateTableRequest {
 }

 impl CreateTableRequest {
-    pub fn new(name: String, data: Box<dyn Scannable>) -> Self {
+    pub fn new(name: String, data: CreateTableData) -> Self {
        Self {
            name,
            namespace: vec![],
--- a/rust/lancedb/src/database/listing.rs
+++ b/rust/lancedb/src/database/listing.rs
@@ -922,7 +922,7 @@ impl Database for ListingDatabase {
            .with_read_params(read_params.clone())
            .load()
            .await
-            .map_err(|e| -> Error { e.into() })?;
+            .map_err(|e| Error::Lance { source: e })?;

        let version_ref = match (request.source_version, request.source_tag) {
            (Some(v), None) => Ok(Ref::Version(None, Some(v))),
@@ -937,7 +937,7 @@ impl Database for ListingDatabase {
        source_dataset
            .shallow_clone(&target_uri, version_ref, Some(storage_params))
            .await
-            .map_err(|e| -> Error { e.into() })?;
+            .map_err(|e| Error::Lance { source: e })?;

        let cloned_table = NativeTable::open_with_params(
            &target_uri,
@@ -1098,10 +1098,8 @@ impl Database for ListingDatabase {
 mod tests {
    use super::*;
    use crate::connection::ConnectRequest;
-    use crate::data::scannable::Scannable;
-    use crate::database::{CreateTableMode, CreateTableRequest};
-    use crate::table::WriteOptions;
-    use crate::Table;
+    use crate::database::{CreateTableData, CreateTableMode, CreateTableRequest, WriteOptions};
+    use crate::table::{Table, TableDefinition};
    use arrow_array::{Int32Array, RecordBatch, StringArray};
    use arrow_schema::{DataType, Field, Schema};
    use std::path::PathBuf;
@@ -1141,7 +1139,7 @@ mod tests {
            .create_table(CreateTableRequest {
                name: "source_table".to_string(),
                namespace: vec![],
-                data: Box::new(RecordBatch::new_empty(schema.clone())) as Box<dyn Scannable>,
+                data: CreateTableData::Empty(TableDefinition::new_from_schema(schema.clone())),
                mode: CreateTableMode::Create,
                write_options: Default::default(),
                location: None,
@@ -1198,11 +1196,16 @@ mod tests {
        )
        .unwrap();

+        let reader = Box::new(arrow_array::RecordBatchIterator::new(
+            vec![Ok(batch)],
+            schema.clone(),
+        ));
+
        let source_table = db
            .create_table(CreateTableRequest {
                name: "source_with_data".to_string(),
                namespace: vec![],
-                data: Box::new(batch) as Box<dyn Scannable>,
+                data: CreateTableData::Data(reader),
                mode: CreateTableMode::Create,
                write_options: Default::default(),
                location: None,
@@ -1261,7 +1264,7 @@ mod tests {
        db.create_table(CreateTableRequest {
            name: "source".to_string(),
            namespace: vec![],
-            data: Box::new(RecordBatch::new_empty(schema)) as Box<dyn Scannable>,
+            data: CreateTableData::Empty(TableDefinition::new_from_schema(schema)),
            mode: CreateTableMode::Create,
            write_options: Default::default(),
            location: None,
@@ -1297,7 +1300,7 @@ mod tests {
        db.create_table(CreateTableRequest {
            name: "source".to_string(),
            namespace: vec![],
-            data: Box::new(RecordBatch::new_empty(schema)) as Box<dyn Scannable>,
+            data: CreateTableData::Empty(TableDefinition::new_from_schema(schema)),
            mode: CreateTableMode::Create,
            write_options: Default::default(),
            location: None,
@@ -1337,7 +1340,7 @@ mod tests {
        db.create_table(CreateTableRequest {
            name: "source".to_string(),
            namespace: vec![],
-            data: Box::new(RecordBatch::new_empty(schema)) as Box<dyn Scannable>,
+            data: CreateTableData::Empty(TableDefinition::new_from_schema(schema)),
            mode: CreateTableMode::Create,
            write_options: Default::default(),
            location: None,
@@ -1377,7 +1380,7 @@ mod tests {
        db.create_table(CreateTableRequest {
            name: "source".to_string(),
            namespace: vec![],
-            data: Box::new(RecordBatch::new_empty(schema)) as Box<dyn Scannable>,
+            data: CreateTableData::Empty(TableDefinition::new_from_schema(schema)),
            mode: CreateTableMode::Create,
            write_options: Default::default(),
            location: None,
@@ -1432,7 +1435,7 @@ mod tests {
        db.create_table(CreateTableRequest {
            name: "source".to_string(),
            namespace: vec![],
-            data: Box::new(RecordBatch::new_empty(schema)) as Box<dyn Scannable>,
+            data: CreateTableData::Empty(TableDefinition::new_from_schema(schema)),
            mode: CreateTableMode::Create,
            write_options: Default::default(),
            location: None,
@@ -1481,11 +1484,16 @@ mod tests {
        )
        .unwrap();

+        let reader = Box::new(arrow_array::RecordBatchIterator::new(
+            vec![Ok(batch1)],
+            schema.clone(),
+        ));
+
        let source_table = db
            .create_table(CreateTableRequest {
                name: "versioned_source".to_string(),
                namespace: vec![],
-                data: Box::new(batch1) as Box<dyn Scannable>,
+                data: CreateTableData::Data(reader),
                mode: CreateTableMode::Create,
                write_options: Default::default(),
                location: None,
@@ -1509,7 +1517,14 @@ mod tests {

        let db = Arc::new(db);
        let source_table_obj = Table::new(source_table.clone(), db.clone());
-        source_table_obj.add(batch2).execute().await.unwrap();
+        source_table_obj
+            .add(Box::new(arrow_array::RecordBatchIterator::new(
+                vec![Ok(batch2)],
+                schema.clone(),
+            )))
+            .execute()
+            .await
+            .unwrap();

        // Verify source table now has 4 rows
        assert_eq!(source_table.count_rows(None).await.unwrap(), 4);
@@ -1555,11 +1570,16 @@ mod tests {
        )
        .unwrap();

+        let reader = Box::new(arrow_array::RecordBatchIterator::new(
+            vec![Ok(batch1)],
+            schema.clone(),
+        ));
+
        let source_table = db
            .create_table(CreateTableRequest {
                name: "tagged_source".to_string(),
                namespace: vec![],
-                data: Box::new(batch1),
+                data: CreateTableData::Data(reader),
                mode: CreateTableMode::Create,
                write_options: Default::default(),
                location: None,
@@ -1587,7 +1607,14 @@ mod tests {
        .unwrap();

        let source_table_obj = Table::new(source_table.clone(), db.clone());
-        source_table_obj.add(batch2).execute().await.unwrap();
+        source_table_obj
+            .add(Box::new(arrow_array::RecordBatchIterator::new(
+                vec![Ok(batch2)],
+                schema.clone(),
+            )))
+            .execute()
+            .await
+            .unwrap();

        // Source table should have 4 rows
        assert_eq!(source_table.count_rows(None).await.unwrap(), 4);
@@ -1630,11 +1657,16 @@ mod tests {
        )
        .unwrap();

+        let reader = Box::new(arrow_array::RecordBatchIterator::new(
+            vec![Ok(batch1)],
+            schema.clone(),
+        ));
+
        let source_table = db
            .create_table(CreateTableRequest {
                name: "independent_source".to_string(),
                namespace: vec![],
-                data: Box::new(batch1),
+                data: CreateTableData::Data(reader),
                mode: CreateTableMode::Create,
                write_options: Default::default(),
                location: None,
@@ -1674,7 +1706,14 @@ mod tests {

        let db = Arc::new(db);
        let cloned_table_obj = Table::new(cloned_table.clone(), db.clone());
-        cloned_table_obj.add(batch_clone).execute().await.unwrap();
+        cloned_table_obj
+            .add(Box::new(arrow_array::RecordBatchIterator::new(
+                vec![Ok(batch_clone)],
+                schema.clone(),
+            )))
+            .execute()
+            .await
+            .unwrap();

        // Add different data to the source table
        let batch_source = RecordBatch::try_new(
@@ -1687,7 +1726,14 @@ mod tests {
        .unwrap();

        let source_table_obj = Table::new(source_table.clone(), db);
-        source_table_obj.add(batch_source).execute().await.unwrap();
+        source_table_obj
+            .add(Box::new(arrow_array::RecordBatchIterator::new(
+                vec![Ok(batch_source)],
+                schema.clone(),
+            )))
+            .execute()
+            .await
+            .unwrap();

        // Verify they have evolved independently
        assert_eq!(source_table.count_rows(None).await.unwrap(), 4); // 2 + 2
@@ -1705,11 +1751,16 @@ mod tests {
            RecordBatch::try_new(schema.clone(), vec![Arc::new(Int32Array::from(vec![1, 2]))])
                .unwrap();

+        let reader = Box::new(arrow_array::RecordBatchIterator::new(
+            vec![Ok(batch1)],
+            schema.clone(),
+        ));
+
        let source_table = db
            .create_table(CreateTableRequest {
                name: "latest_version_source".to_string(),
                namespace: vec![],
-                data: Box::new(batch1),
+                data: CreateTableData::Data(reader),
                mode: CreateTableMode::Create,
                write_options: Default::default(),
                location: None,
@@ -1728,7 +1779,14 @@ mod tests {
            .unwrap();

            let source_table_obj = Table::new(source_table.clone(), db.clone());
-            source_table_obj.add(batch).execute().await.unwrap();
+            source_table_obj
+                .add(Box::new(arrow_array::RecordBatchIterator::new(
+                    vec![Ok(batch)],
+                    schema.clone(),
+                )))
+                .execute()
+                .await
+                .unwrap();
        }

        // Source should have 8 rows total (2 + 2 + 2 + 2)
@@ -1791,11 +1849,16 @@ mod tests {
        )
        .unwrap();

+        let reader = Box::new(arrow_array::RecordBatchIterator::new(
+            vec![Ok(batch)],
+            schema.clone(),
+        ));
+
        let table = db
            .create_table(CreateTableRequest {
                name: "test_stable".to_string(),
                namespace: vec![],
-                data: Box::new(batch),
+                data: CreateTableData::Data(reader),
                mode: CreateTableMode::Create,
                write_options: Default::default(),
                location: None,
@@ -1824,6 +1887,11 @@ mod tests {
        )
        .unwrap();

+        let reader = Box::new(arrow_array::RecordBatchIterator::new(
+            vec![Ok(batch)],
+            schema.clone(),
+        ));
+
        let mut storage_options = HashMap::new();
        storage_options.insert(
            OPT_NEW_TABLE_ENABLE_STABLE_ROW_IDS.to_string(),
@@ -1846,7 +1914,7 @@ mod tests {
            .create_table(CreateTableRequest {
                name: "test_stable_table_level".to_string(),
                namespace: vec![],
-                data: Box::new(batch),
+                data: CreateTableData::Data(reader),
                mode: CreateTableMode::Create,
                write_options,
                location: None,
@@ -1895,6 +1963,11 @@ mod tests {
        )
        .unwrap();

+        let reader = Box::new(arrow_array::RecordBatchIterator::new(
+            vec![Ok(batch)],
+            schema.clone(),
+        ));
+
        let mut storage_options = HashMap::new();
        storage_options.insert(
            OPT_NEW_TABLE_ENABLE_STABLE_ROW_IDS.to_string(),
@@ -1917,7 +1990,7 @@ mod tests {
            .create_table(CreateTableRequest {
                name: "test_override".to_string(),
                namespace: vec![],
-                data: Box::new(batch),
+                data: CreateTableData::Data(reader),
                mode: CreateTableMode::Create,
                write_options,
                location: None,
@@ -2035,7 +2108,7 @@ mod tests {
        db.create_table(CreateTableRequest {
            name: "table1".to_string(),
            namespace: vec![],
-            data: Box::new(RecordBatch::new_empty(schema.clone())) as Box<dyn Scannable>,
+            data: CreateTableData::Empty(TableDefinition::new_from_schema(schema.clone())),
            mode: CreateTableMode::Create,
            write_options: Default::default(),
            location: None,
@@ -2047,7 +2120,7 @@ mod tests {
        db.create_table(CreateTableRequest {
            name: "table2".to_string(),
            namespace: vec![],
-            data: Box::new(RecordBatch::new_empty(schema)) as Box<dyn Scannable>,
+            data: CreateTableData::Empty(TableDefinition::new_from_schema(schema)),
            mode: CreateTableMode::Create,
            write_options: Default::default(),
            location: None,
--- a/rust/lancedb/src/database/namespace.rs
+++ b/rust/lancedb/src/database/namespace.rs
@@ -354,13 +354,15 @@ mod tests {
    use super::*;
    use crate::connect_namespace;
    use crate::query::ExecutableQuery;
-    use arrow_array::{Int32Array, RecordBatch, StringArray};
+    use arrow_array::{Int32Array, RecordBatch, RecordBatchIterator, StringArray};
    use arrow_schema::{DataType, Field, Schema};
    use futures::TryStreamExt;
    use tempfile::tempdir;

    /// Helper function to create test data
-    fn create_test_data() -> RecordBatch {
+    fn create_test_data() -> RecordBatchIterator<
+        std::vec::IntoIter<std::result::Result<RecordBatch, arrow_schema::ArrowError>>,
+    > {
        let schema = Arc::new(Schema::new(vec![
            Field::new("id", DataType::Int32, false),
            Field::new("name", DataType::Utf8, false),
@@ -369,7 +371,12 @@ mod tests {
        let id_array = Int32Array::from(vec![1, 2, 3, 4, 5]);
        let name_array = StringArray::from(vec!["Alice", "Bob", "Charlie", "David", "Eve"]);

-        RecordBatch::try_new(schema, vec![Arc::new(id_array), Arc::new(name_array)]).unwrap()
+        let batch = RecordBatch::try_new(
+            schema.clone(),
+            vec![Arc::new(id_array), Arc::new(name_array)],
+        )
+        .unwrap();
+        RecordBatchIterator::new(vec![std::result::Result::Ok(batch)].into_iter(), schema)
    }

    #[tokio::test]
@@ -611,7 +618,13 @@ mod tests {

        // Test: Overwrite the table
        let table2 = conn
-            .create_table("overwrite_test", test_data2)
+            .create_table(
+                "overwrite_test",
+                RecordBatchIterator::new(
+                    vec![std::result::Result::Ok(test_data2)].into_iter(),
+                    schema,
+                ),
+            )
            .namespace(vec!["test_ns".into()])
            .mode(CreateTableMode::Overwrite)
            .execute()
--- a/rust/lancedb/src/dataloader/permutation/builder.rs
+++ b/rust/lancedb/src/dataloader/permutation/builder.rs
@@ -13,7 +13,7 @@ use lance_datafusion::exec::SessionContextExt;
 use crate::{
    arrow::{SendableRecordBatchStream, SendableRecordBatchStreamExt, SimpleRecordBatchStream},
    connect,
-    database::{CreateTableRequest, Database},
+    database::{CreateTableData, CreateTableRequest, Database},
    dataloader::permutation::{
        shuffle::{Shuffler, ShufflerConfig},
        split::{SplitStrategy, Splitter, SPLIT_ID_COLUMN},
@@ -313,8 +313,10 @@ impl PermutationBuilder {
            }
        };

-        let create_table_request =
-            CreateTableRequest::new(name.to_string(), Box::new(streaming_data));
+        let create_table_request = CreateTableRequest::new(
+            name.to_string(),
+            CreateTableData::StreamingData(streaming_data),
+        );

        let table = database.create_table(create_table_request).await?;

@@ -345,7 +347,7 @@ mod tests {
            .col("col_b", lance_datagen::array::step::<Int32Type>())
            .into_ldb_stream(RowCount::from(100), BatchCount::from(10));
        let data_table = db
-            .create_table("base_tbl", initial_data)
+            .create_table_streaming("base_tbl", initial_data)
            .execute()
            .await
            .unwrap();
@@ -385,7 +387,7 @@ mod tests {
            .col("some_value", lance_datagen::array::step::<Int32Type>())
            .into_ldb_stream(RowCount::from(100), BatchCount::from(10));
        let data_table = db
-            .create_table("mytbl", initial_data)
+            .create_table_streaming("mytbl", initial_data)
            .execute()
            .await
            .unwrap();
--- a/rust/lancedb/src/embeddings.rs
+++ b/rust/lancedb/src/embeddings.rs
@@ -18,7 +18,7 @@ use std::{
 };

 use arrow_array::{Array, RecordBatch, RecordBatchReader};
-use arrow_schema::{DataType, Field, SchemaBuilder, SchemaRef};
+use arrow_schema::{DataType, Field, SchemaBuilder};
 // use async_trait::async_trait;
 use serde::{Deserialize, Serialize};

@@ -190,112 +190,6 @@ impl<R: RecordBatchReader> WithEmbeddings<R> {
    }
 }

-/// Compute embedding arrays for a batch.
-///
-/// When multiple embedding functions are defined, they are computed in parallel using
-/// scoped threads. For a single embedding function, computation is done inline.
-fn compute_embedding_arrays(
-    batch: &RecordBatch,
-    embeddings: &[(EmbeddingDefinition, Arc<dyn EmbeddingFunction>)],
-) -> Result<Vec<Arc<dyn Array>>> {
-    if embeddings.len() == 1 {
-        let (fld, func) = &embeddings[0];
-        let src_column =
-            batch
-                .column_by_name(&fld.source_column)
-                .ok_or_else(|| Error::InvalidInput {
-                    message: format!("Source column '{}' not found", fld.source_column),
-                })?;
-        return Ok(vec![func.compute_source_embeddings(src_column.clone())?]);
-    }
-
-    // Parallel path: multiple embeddings
-    std::thread::scope(|s| {
-        let handles: Vec<_> = embeddings
-            .iter()
-            .map(|(fld, func)| {
-                let src_column = batch.column_by_name(&fld.source_column).ok_or_else(|| {
-                    Error::InvalidInput {
-                        message: format!("Source column '{}' not found", fld.source_column),
-                    }
-                })?;
-
-                let handle = s.spawn(move || func.compute_source_embeddings(src_column.clone()));
-
-                Ok(handle)
-            })
-            .collect::<Result<_>>()?;
-
-        handles
-            .into_iter()
-            .map(|h| {
-                h.join().map_err(|e| Error::Runtime {
-                    message: format!("Thread panicked during embedding computation: {:?}", e),
-                })?
-            })
-            .collect()
-    })
-}
-
-/// Compute the output schema when embeddings are applied to a base schema.
-///
-/// This returns the schema with embedding columns appended.
-pub fn compute_output_schema(
-    base_schema: &SchemaRef,
-    embeddings: &[(EmbeddingDefinition, Arc<dyn EmbeddingFunction>)],
-) -> Result<SchemaRef> {
-    let mut sb: SchemaBuilder = base_schema.as_ref().into();
-
-    for (ed, func) in embeddings {
-        let src_field = base_schema
-            .field_with_name(&ed.source_column)
-            .map_err(|_| Error::InvalidInput {
-                message: format!("Source column '{}' not found in schema", ed.source_column),
-            })?;
-
-        let field_name = ed
-            .dest_column
-            .clone()
-            .unwrap_or_else(|| format!("{}_embedding", &ed.source_column));
-
-        sb.push(Field::new(
-            field_name,
-            func.dest_type()?.into_owned(),
-            src_field.is_nullable(),
-        ));
-    }
-
-    Ok(Arc::new(sb.finish()))
-}
-
-/// Compute embeddings for a batch and append as new columns.
-///
-/// This function computes embeddings using the provided embedding functions and
-/// appends them as new columns to the batch.
-pub fn compute_embeddings_for_batch(
-    batch: RecordBatch,
-    embeddings: &[(EmbeddingDefinition, Arc<dyn EmbeddingFunction>)],
-) -> Result<RecordBatch> {
-    let embedding_arrays = compute_embedding_arrays(&batch, embeddings)?;
-
-    let mut result = batch;
-    for ((fld, _), embedding) in embeddings.iter().zip(embedding_arrays.iter()) {
-        let dst_field_name = fld
-            .dest_column
-            .clone()
-            .unwrap_or_else(|| format!("{}_embedding", &fld.source_column));
-
-        let dst_field = Field::new(
-            dst_field_name,
-            embedding.data_type().clone(),
-            embedding.nulls().is_some(),
-        );
-
-        result = result.try_with_column(dst_field, embedding.clone())?;
-    }
-    Ok(result)
-}
-
 impl<R: RecordBatchReader> WithEmbeddings<R> {
    fn dest_fields(&self) -> Result<Vec<Field>> {
        let schema = self.inner.schema();
@@ -346,6 +240,48 @@ impl<R: RecordBatchReader> WithEmbeddings<R> {
            column_definitions,
        })
    }
+
+    fn compute_embeddings_parallel(&self, batch: &RecordBatch) -> Result<Vec<Arc<dyn Array>>> {
+        if self.embeddings.len() == 1 {
+            let (fld, func) = &self.embeddings[0];
+            let src_column =
+                batch
+                    .column_by_name(&fld.source_column)
+                    .ok_or_else(|| Error::InvalidInput {
+                        message: format!("Source column '{}' not found", fld.source_column),
+                    })?;
+            return Ok(vec![func.compute_source_embeddings(src_column.clone())?]);
+        }
+
+        // Parallel path: multiple embeddings
+        std::thread::scope(|s| {
+            let handles: Vec<_> = self
+                .embeddings
+                .iter()
+                .map(|(fld, func)| {
+                    let src_column = batch.column_by_name(&fld.source_column).ok_or_else(|| {
+                        Error::InvalidInput {
+                            message: format!("Source column '{}' not found", fld.source_column),
+                        }
+                    })?;
+
+                    let handle =
+                        s.spawn(move || func.compute_source_embeddings(src_column.clone()));
+
+                    Ok(handle)
+                })
+                .collect::<Result<_>>()?;
+
+            handles
+                .into_iter()
+                .map(|h| {
+                    h.join().map_err(|e| Error::Runtime {
+                        message: format!("Thread panicked during embedding computation: {:?}", e),
+                    })?
+                })
+                .collect()
+        })
+    }
 }

 impl<R: RecordBatchReader> Iterator for MaybeEmbedded<R> {
@@ -373,13 +309,37 @@ impl<R: RecordBatchReader> Iterator for WithEmbeddings<R> {
    fn next(&mut self) -> Option<Self::Item> {
        let batch = self.inner.next()?;
        match batch {
-            Ok(batch) => match compute_embeddings_for_batch(batch, &self.embeddings) {
-                Ok(batch_with_embeddings) => Some(Ok(batch_with_embeddings)),
-                Err(e) => Some(Err(arrow_schema::ArrowError::ComputeError(format!(
-                    "Error computing embedding: {}",
-                    e
-                )))),
-            },
+            Ok(batch) => {
+                let embeddings = match self.compute_embeddings_parallel(&batch) {
+                    Ok(emb) => emb,
+                    Err(e) => {
+                        return Some(Err(arrow_schema::ArrowError::ComputeError(format!(
+                            "Error computing embedding: {}",
+                            e
+                        ))))
+                    }
+                };
+
+                let mut batch = batch;
+                for ((fld, _), embedding) in self.embeddings.iter().zip(embeddings.iter()) {
+                    let dst_field_name = fld
+                        .dest_column
+                        .clone()
+                        .unwrap_or_else(|| format!("{}_embedding", &fld.source_column));
+
+                    let dst_field = Field::new(
+                        dst_field_name,
+                        embedding.data_type().clone(),
+                        embedding.nulls().is_some(),
+                    );
+
+                    match batch.try_with_column(dst_field.clone(), embedding.clone()) {
+                        Ok(b) => batch = b,
+                        Err(e) => return Some(Err(e)),
+                    };
+                }
+                Some(Ok(batch))
+            }
            Err(e) => Some(Err(e)),
        }
    }
--- a/rust/lancedb/src/error.rs
+++ b/rust/lancedb/src/error.rs
@@ -6,7 +6,7 @@ use std::sync::PoisonError;
 use arrow_schema::ArrowError;
 use snafu::Snafu;

-pub(crate) type BoxError = Box<dyn std::error::Error + Send + Sync>;
+type BoxError = Box<dyn std::error::Error + Send + Sync>;

 #[derive(Debug, Snafu)]
 #[snafu(visibility(pub(crate)))]
@@ -80,9 +80,6 @@ pub enum Error {
    Arrow { source: ArrowError },
    #[snafu(display("LanceDBError: not supported: {message}"))]
    NotSupported { message: String },
-    /// External error pass through from user code.
-    #[snafu(transparent)]
-    External { source: BoxError },
    #[snafu(whatever, display("{message}"))]
    Other {
        message: String,
@@ -95,26 +92,15 @@ pub type Result<T> = std::result::Result<T, Error>;

 impl From<ArrowError> for Error {
    fn from(source: ArrowError) -> Self {
-        match source {
-            ArrowError::ExternalError(source) => match source.downcast::<Self>() {
-                Ok(e) => *e,
-                Err(source) => Self::External { source },
-            },
-            _ => Self::Arrow { source },
-        }
+        Self::Arrow { source }
    }
 }

 impl From<lance::Error> for Error {
    fn from(source: lance::Error) -> Self {
-        // Try to unwrap external errors that were wrapped by lance
-        match source {
-            lance::Error::Wrapped { error, .. } => match error.downcast::<Self>() {
-                Ok(e) => *e,
-                Err(source) => Self::External { source },
-            },
-            _ => Self::Lance { source },
-        }
+        // TODO: Once Lance is changed to preserve ObjectStore, DataFusion, and Arrow errors, we can
+        // pass those variants through here as well.
+        Self::Lance { source }
    }
 }

--- a/rust/lancedb/src/io/object_store.rs
+++ b/rust/lancedb/src/io/object_store.rs
@@ -218,9 +218,8 @@ mod test {
        datagen = datagen.col(Box::<IncrementingInt32>::default());
        datagen = datagen.col(Box::new(RandomVector::default().named("vector".into())));

-        let data: Box<dyn arrow_array::RecordBatchReader + Send> = Box::new(datagen.batch(100));
        let res = db
-            .create_table("test", data)
+            .create_table("test", Box::new(datagen.batch(100)))
            .write_options(WriteOptions {
                lance_write_params: Some(param),
            })
--- a/rust/lancedb/src/ipc.rs
+++ b/rust/lancedb/src/ipc.rs
@@ -12,10 +12,10 @@ use arrow_schema::Schema;
 use crate::{Error, Result};

 /// Convert a Arrow IPC file to a batch reader
-pub fn ipc_file_to_batches(buf: Vec<u8>) -> Result<Box<dyn RecordBatchReader + Send>> {
+pub fn ipc_file_to_batches(buf: Vec<u8>) -> Result<impl RecordBatchReader> {
    let buf_reader = Cursor::new(buf);
    let reader = FileReader::try_new(buf_reader, None)?;
-    Ok(Box::new(reader))
+    Ok(reader)
 }

 /// Convert record batches to Arrow IPC file
--- a/rust/lancedb/src/lib.rs
+++ b/rust/lancedb/src/lib.rs
@@ -39,6 +39,7 @@
 //! #### Connect to a database.
 //!
 //! ```rust
+//! # use arrow_schema::{Field, Schema};
 //! # tokio::runtime::Runtime::new().unwrap().block_on(async {
 //! let db = lancedb::connect("data/sample-lancedb").execute().await.unwrap();
 //! # });
@@ -73,10 +74,7 @@
 //!
 //! #### Create a table
 //!
-//! To create a Table, you need to provide an [`arrow_array::RecordBatch`]. The
-//! schema of the `RecordBatch` determines the schema of the table.
-//!
-//! Vector columns should be represented as `FixedSizeList<Float16/Float32>` data type.
+//! To create a Table, you need to provide a [`arrow_schema::Schema`] and a [`arrow_array::RecordBatch`] stream.
 //!
 //! ```rust
 //! # use std::sync::Arc;
@@ -87,29 +85,34 @@
 //! # tokio::runtime::Runtime::new().unwrap().block_on(async {
 //! # let tmpdir = tempfile::tempdir().unwrap();
 //! # let db = lancedb::connect(tmpdir.path().to_str().unwrap()).execute().await.unwrap();
-//! let ndims = 128;
 //! let schema = Arc::new(Schema::new(vec![
 //!     Field::new("id", DataType::Int32, false),
 //!     Field::new(
 //!         "vector",
-//!         DataType::FixedSizeList(Arc::new(Field::new("item", DataType::Float32, true)), ndims),
+//!         DataType::FixedSizeList(Arc::new(Field::new("item", DataType::Float32, true)), 128),
 //!         true,
 //!     ),
 //! ]));
-//! let data = RecordBatch::try_new(
+//! // Create a RecordBatch stream.
+//! let batches = RecordBatchIterator::new(
+//!     vec![RecordBatch::try_new(
 //!         schema.clone(),
 //!         vec![
 //!             Arc::new(Int32Array::from_iter_values(0..256)),
 //!             Arc::new(
 //!                 FixedSizeListArray::from_iter_primitive::<Float32Type, _, _>(
-//!                     (0..256).map(|_| Some(vec![Some(1.0); ndims as usize])),
-//!                     ndims,
+//!                     (0..256).map(|_| Some(vec![Some(1.0); 128])),
+//!                     128,
 //!                 ),
 //!             ),
 //!         ],
 //!     )
-//!     .unwrap();
-//! db.create_table("my_table", data)
+//!     .unwrap()]
+//!     .into_iter()
+//!     .map(Ok),
+//!     schema.clone(),
+//! );
+//! db.create_table("my_table", Box::new(batches))
 //!     .execute()
 //!     .await
 //!     .unwrap();
@@ -148,18 +151,42 @@
 //! #### Open table and search
 //!
 //! ```rust
+//! # use std::sync::Arc;
 //! # use futures::TryStreamExt;
+//! # use arrow_schema::{DataType, Schema, Field};
+//! # use arrow_array::{RecordBatch, RecordBatchIterator};
+//! # use arrow_array::{FixedSizeListArray, Float32Array, Int32Array, types::Float32Type};
 //! # use lancedb::query::{ExecutableQuery, QueryBase};
-//! # async fn example(table: &lancedb::Table) -> lancedb::Result<()> {
+//! # tokio::runtime::Runtime::new().unwrap().block_on(async {
+//! # let tmpdir = tempfile::tempdir().unwrap();
+//! # let db = lancedb::connect(tmpdir.path().to_str().unwrap()).execute().await.unwrap();
+//! # let schema = Arc::new(Schema::new(vec![
+//! #  Field::new("id", DataType::Int32, false),
+//! #  Field::new("vector", DataType::FixedSizeList(
+//! #    Arc::new(Field::new("item", DataType::Float32, true)), 128), true),
+//! # ]));
+//! # let batches = RecordBatchIterator::new(vec![
+//! #    RecordBatch::try_new(schema.clone(),
+//! #       vec![
+//! #           Arc::new(Int32Array::from_iter_values(0..10)),
+//! #           Arc::new(FixedSizeListArray::from_iter_primitive::<Float32Type, _, _>(
+//! #               (0..10).map(|_| Some(vec![Some(1.0); 128])), 128)),
+//! #       ]).unwrap()
+//! #   ].into_iter().map(Ok),
+//! #    schema.clone());
+//! # db.create_table("my_table", Box::new(batches)).execute().await.unwrap();
+//! # let table = db.open_table("my_table").execute().await.unwrap();
 //! let results = table
 //!     .query()
-//!     .nearest_to(&[1.0; 128])?
+//!     .nearest_to(&[1.0; 128])
+//!     .unwrap()
 //!     .execute()
-//!     .await?
+//!     .await
+//!     .unwrap()
 //!     .try_collect::<Vec<_>>()
-//!     .await?;
-//! #   Ok(())
-//! # }
+//!     .await
+//!     .unwrap();
+//! # });
 //! ```

 pub mod arrow;
--- a/rust/lancedb/src/query.rs
+++ b/rust/lancedb/src/query.rs
@@ -1381,7 +1381,7 @@ mod tests {
    use arrow::{array::downcast_array, compute::concat_batches, datatypes::Int32Type};
    use arrow_array::{
        cast::AsArray, types::Float32Type, FixedSizeListArray, Float32Array, Int32Array,
-        RecordBatch, StringArray,
+        RecordBatch, RecordBatchIterator, RecordBatchReader, StringArray,
    };
    use arrow_schema::{DataType, Field as ArrowField, Schema as ArrowSchema};
    use futures::{StreamExt, TryStreamExt};
@@ -1402,7 +1402,7 @@ mod tests {
        let batches = make_test_batches();
        let conn = connect(uri).execute().await.unwrap();
        let table = conn
-            .create_table("my_table", batches)
+            .create_table("my_table", Box::new(batches))
            .execute()
            .await
            .unwrap();
@@ -1463,7 +1463,7 @@ mod tests {
        let batches = make_non_empty_batches();
        let conn = connect(uri).execute().await.unwrap();
        let table = conn
-            .create_table("my_table", batches)
+            .create_table("my_table", Box::new(batches))
            .execute()
            .await
            .unwrap();
@@ -1525,7 +1525,7 @@ mod tests {
        let batches = make_non_empty_batches();
        let conn = connect(uri).execute().await.unwrap();
        let table = conn
-            .create_table("my_table", batches)
+            .create_table("my_table", Box::new(batches))
            .execute()
            .await
            .unwrap();
@@ -1578,7 +1578,7 @@ mod tests {
        let batches = make_non_empty_batches();
        let conn = connect(uri).execute().await.unwrap();
        let table = conn
-            .create_table("my_table", batches)
+            .create_table("my_table", Box::new(batches))
            .execute()
            .await
            .unwrap();
@@ -1599,13 +1599,13 @@ mod tests {
        assert!(result.is_err());
    }

-    fn make_non_empty_batches() -> Box<dyn arrow_array::RecordBatchReader + Send> {
+    fn make_non_empty_batches() -> impl RecordBatchReader + Send + 'static {
        let vec = Box::new(RandomVector::new().named("vector".to_string()));
        let id = Box::new(IncrementingInt32::new().named("id".to_string()));
-        Box::new(BatchGenerator::new().col(vec).col(id).batch(512))
+        BatchGenerator::new().col(vec).col(id).batch(512)
    }

-    fn make_test_batches() -> RecordBatch {
+    fn make_test_batches() -> impl RecordBatchReader + Send + 'static {
        let dim: usize = 128;
        let schema = Arc::new(ArrowSchema::new(vec![
            ArrowField::new("key", DataType::Int32, false),
@@ -1619,7 +1619,12 @@ mod tests {
            ),
            ArrowField::new("uri", DataType::Utf8, true),
        ]));
-        RecordBatch::new_empty(schema)
+        RecordBatchIterator::new(
+            vec![RecordBatch::new_empty(schema.clone())]
+                .into_iter()
+                .map(Ok),
+            schema,
+        )
    }

    async fn make_test_table(tmp_dir: &tempfile::TempDir) -> Table {
@@ -1628,7 +1633,7 @@ mod tests {

        let batches = make_non_empty_batches();
        let conn = connect(uri).execute().await.unwrap();
-        conn.create_table("my_table", batches)
+        conn.create_table("my_table", Box::new(batches))
            .execute()
            .await
            .unwrap()
@@ -1857,8 +1862,10 @@ mod tests {

        let record_batch =
            RecordBatch::try_new(schema.clone(), vec![Arc::new(text), Arc::new(vector)]).unwrap();
+        let record_batch_iter =
+            RecordBatchIterator::new(vec![record_batch].into_iter().map(Ok), schema.clone());
        let table = conn
-            .create_table("my_table", record_batch)
+            .create_table("my_table", record_batch_iter)
            .execute()
            .await
            .unwrap();
@@ -1942,8 +1949,10 @@ mod tests {
            ],
        )
        .unwrap();
+        let record_batch_iter =
+            RecordBatchIterator::new(vec![record_batch].into_iter().map(Ok), schema.clone());
        let table = conn
-            .create_table("my_table", record_batch)
+            .create_table("my_table", record_batch_iter)
            .mode(CreateTableMode::Overwrite)
            .execute()
            .await
@@ -2053,6 +2062,8 @@ mod tests {
    async fn test_pagination_with_fts() {
        let db = connect("memory://test").execute().await.unwrap();
        let data = fts_test_data(400);
+        let schema = data.schema();
+        let data = RecordBatchIterator::new(vec![Ok(data)], schema);
        let table = db.create_table("test_table", data).execute().await.unwrap();

        table
--- a/rust/lancedb/src/remote/client.rs
+++ b/rust/lancedb/src/remote/client.rs
@@ -491,7 +491,7 @@ impl<S: HttpSend> RestfulLanceDbClient<S> {
    }

    /// Apply dynamic headers from the header provider if configured
-    pub(crate) async fn apply_dynamic_headers(&self, mut request: Request) -> Result<Request> {
+    async fn apply_dynamic_headers(&self, mut request: Request) -> Result<Request> {
        if let Some(ref provider) = self.header_provider {
            let headers = provider.get_headers().await?;
            let request_headers = request.headers_mut();
@@ -555,9 +555,7 @@ impl<S: HttpSend> RestfulLanceDbClient<S> {
            message: "Attempted to retry a request that cannot be cloned".to_string(),
        })?;
        let (_, r) = tmp_req.build_split();
-        let mut r = r.map_err(|e| Error::Runtime {
-            message: format!("Failed to build request: {}", e),
-        })?;
+        let mut r = r.unwrap();
        let request_id = self.extract_request_id(&mut r);
        let mut retry_counter = RetryCounter::new(retry_config, request_id.clone());

@@ -573,9 +571,7 @@ impl<S: HttpSend> RestfulLanceDbClient<S> {
            }

            let (c, request) = req_builder.build_split();
-            let mut request = request.map_err(|e| Error::Runtime {
-                message: format!("Failed to build request: {}", e),
-            })?;
+            let mut request = request.unwrap();
            self.set_request_id(&mut request, &request_id.clone());

            // Apply dynamic headers before each retry attempt
@@ -625,7 +621,7 @@ impl<S: HttpSend> RestfulLanceDbClient<S> {
        }
    }

-    pub(crate) fn log_request(&self, request: &Request, request_id: &String) {
+    fn log_request(&self, request: &Request, request_id: &String) {
        if log::log_enabled!(log::Level::Debug) {
            let content_type = request
                .headers()
--- a/rust/lancedb/src/remote/db.rs
+++ b/rust/lancedb/src/remote/db.rs
@@ -4,11 +4,13 @@
 use std::collections::HashMap;
 use std::sync::Arc;

+use arrow_array::RecordBatchIterator;
 use async_trait::async_trait;
 use http::StatusCode;
 use lance_io::object_store::StorageOptions;
 use moka::future::Cache;
 use reqwest::header::CONTENT_TYPE;
+use tokio::task::spawn_blocking;

 use lance_namespace::models::{
    CreateNamespaceRequest, CreateNamespaceResponse, DescribeNamespaceRequest,
@@ -17,17 +19,16 @@ use lance_namespace::models::{
 };

 use crate::database::{
-    CloneTableRequest, CreateTableMode, CreateTableRequest, Database, DatabaseOptions,
-    OpenTableRequest, ReadConsistency, TableNamesRequest,
+    CloneTableRequest, CreateTableData, CreateTableMode, CreateTableRequest, Database,
+    DatabaseOptions, OpenTableRequest, ReadConsistency, TableNamesRequest,
 };
 use crate::error::Result;
-use crate::remote::util::stream_as_body;
 use crate::table::BaseTable;
 use crate::Error;

 use super::client::{ClientConfig, HttpSend, RequestResultExt, RestfulLanceDbClient, Sender};
 use super::table::RemoteTable;
-use super::util::parse_server_version;
+use super::util::{batches_to_ipc_bytes, parse_server_version};
 use super::ARROW_STREAM_CONTENT_TYPE;

 // Request structure for the remote clone table API
@@ -435,8 +436,26 @@ impl<S: HttpSend> Database for RemoteDatabase<S> {
        Ok(response)
    }

-    async fn create_table(&self, mut request: CreateTableRequest) -> Result<Arc<dyn BaseTable>> {
-        let body = stream_as_body(request.data.scan_as_stream())?;
+    async fn create_table(&self, request: CreateTableRequest) -> Result<Arc<dyn BaseTable>> {
+        let data = match request.data {
+            CreateTableData::Data(data) => data,
+            CreateTableData::StreamingData(_) => {
+                return Err(Error::NotSupported {
+                    message: "Creating a remote table from a streaming source".to_string(),
+                })
+            }
+            CreateTableData::Empty(table_definition) => {
+                let schema = table_definition.schema.clone();
+                Box::new(RecordBatchIterator::new(vec![], schema))
+            }
+        };
+
+        // TODO: https://github.com/lancedb/lancedb/issues/1026
+        // We should accept data from an async source.  In the meantime, spawn this as blocking
+        // to make sure we don't block the tokio runtime if the source is slow.
+        let data_buffer = spawn_blocking(move || batches_to_ipc_bytes(data))
+            .await
+            .unwrap()?;

        let identifier =
            build_table_identifier(&request.name, &request.namespace, &self.client.id_delimiter);
@@ -444,7 +463,7 @@ impl<S: HttpSend> Database for RemoteDatabase<S> {
            .client
            .post(&format!("/v1/table/{}/create/", identifier))
            .query(&[("mode", Into::<&str>::into(&request.mode))])
-            .body(body)
+            .body(data_buffer)
            .header(CONTENT_TYPE, ARROW_STREAM_CONTENT_TYPE);

        let (request_id, rsp) = self.client.send(req).await?;
@@ -794,7 +813,7 @@ mod tests {
    use std::collections::HashMap;
    use std::sync::{Arc, OnceLock};

-    use arrow_array::{Int32Array, RecordBatch};
+    use arrow_array::{Int32Array, RecordBatch, RecordBatchIterator};
    use arrow_schema::{DataType, Field, Schema};

    use crate::connection::ConnectBuilder;
@@ -974,7 +993,8 @@ mod tests {
            vec![Arc::new(Int32Array::from(vec![1, 2, 3]))],
        )
        .unwrap();
-        let table = conn.create_table("table1", data).execute().await.unwrap();
+        let reader = RecordBatchIterator::new([Ok(data.clone())], data.schema());
+        let table = conn.create_table("table1", reader).execute().await.unwrap();
        assert_eq!(table.name(), "table1");
    }

@@ -991,7 +1011,8 @@ mod tests {
            vec![Arc::new(Int32Array::from(vec![1, 2, 3]))],
        )
        .unwrap();
-        let result = conn.create_table("table1", data).execute().await;
+        let reader = RecordBatchIterator::new([Ok(data.clone())], data.schema());
+        let result = conn.create_table("table1", reader).execute().await;
        assert!(result.is_err());
        assert!(
            matches!(result, Err(crate::Error::TableAlreadyExists { name }) if name == "table1")
@@ -1024,7 +1045,8 @@ mod tests {
                vec![Arc::new(Int32Array::from(vec![1, 2, 3]))],
            )
            .unwrap();
-            let mut builder = conn.create_table("table1", data.clone());
+            let reader = RecordBatchIterator::new([Ok(data.clone())], data.schema());
+            let mut builder = conn.create_table("table1", reader);
            if let Some(mode) = mode {
                builder = builder.mode(mode);
            }
@@ -1049,8 +1071,9 @@ mod tests {
        .unwrap();

        let called: Arc<OnceLock<bool>> = Arc::new(OnceLock::new());
+        let reader = RecordBatchIterator::new([Ok(data.clone())], data.schema());
        let called_in_cb = called.clone();
-        conn.create_table("table1", data)
+        conn.create_table("table1", reader)
            .mode(CreateTableMode::ExistOk(Box::new(move |b| {
                called_in_cb.clone().set(true).unwrap();
                b
@@ -1239,8 +1262,9 @@ mod tests {
            vec![Arc::new(Int32Array::from(vec![1, 2, 3]))],
        )
        .unwrap();
+        let reader = RecordBatchIterator::new([Ok(data.clone())], data.schema());
        let table = conn
-            .create_table("table1", data)
+            .create_table("table1", reader)
            .namespace(vec!["ns1".to_string()])
            .execute()
            .await
@@ -1706,8 +1730,10 @@ mod tests {
                vec![Arc::new(Int32Array::from(vec![1, 2, 3]))],
            )
            .unwrap();
+            let reader = RecordBatchIterator::new([Ok(data.clone())], schema.clone());
+
            let table = conn
-                .create_table("test_table", data)
+                .create_table("test_table", reader)
                .namespace(namespace.clone())
                .execute()
                .await;
@@ -1780,7 +1806,9 @@ mod tests {
                let data =
                    RecordBatch::try_new(schema.clone(), vec![Arc::new(Int32Array::from(vec![i]))])
                        .unwrap();
-                conn.create_table(format!("table{}", i), data)
+                let reader = RecordBatchIterator::new([Ok(data.clone())], schema.clone());
+
+                conn.create_table(format!("table{}", i), reader)
                    .namespace(namespace.clone())
                    .execute()
                    .await
--- a/rust/lancedb/src/remote/table.rs
+++ b/rust/lancedb/src/remote/table.rs
--- a/rust/lancedb/src/remote/util.rs
+++ b/rust/lancedb/src/remote/util.rs
@@ -1,50 +1,29 @@
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright The LanceDB Authors

-use arrow_ipc::CompressionType;
-use futures::{Stream, StreamExt};
+use std::io::Cursor;
+
+use arrow_array::RecordBatchReader;
 use reqwest::Response;

-use crate::{arrow::SendableRecordBatchStream, Result};
+use crate::Result;

 use super::db::ServerVersion;

-pub fn stream_as_ipc(
-    data: SendableRecordBatchStream,
-) -> Result<impl Stream<Item = Result<bytes::Bytes>>> {
-    let options = arrow_ipc::writer::IpcWriteOptions::default()
-        .try_with_compression(Some(CompressionType::LZ4_FRAME))?;
+pub fn batches_to_ipc_bytes(batches: impl RecordBatchReader) -> Result<Vec<u8>> {
    const WRITE_BUF_SIZE: usize = 4096;
    let buf = Vec::with_capacity(WRITE_BUF_SIZE);
-    let writer =
-        arrow_ipc::writer::StreamWriter::try_new_with_options(buf, &data.schema(), options)?;
-    let stream = futures::stream::try_unfold(
-        (data, writer, false),
-        move |(mut data, mut writer, finished)| async move {
-            if finished {
-                return Ok(None);
-            }
-            match data.next().await {
-                Some(Ok(batch)) => {
-                    writer.write(&batch)?;
-                    let buffer = std::mem::take(writer.get_mut());
-                    Ok(Some((bytes::Bytes::from(buffer), (data, writer, false))))
-                }
-                Some(Err(e)) => Err(e),
-                None => {
-                    writer.finish()?;
-                    let buffer = std::mem::take(writer.get_mut());
-                    Ok(Some((bytes::Bytes::from(buffer), (data, writer, true))))
-                }
-            }
-        },
-    );
-    Ok(stream)
-}
+    let mut buf = Cursor::new(buf);
+    {
+        let mut writer = arrow_ipc::writer::StreamWriter::try_new(&mut buf, &batches.schema())?;

-pub fn stream_as_body(data: SendableRecordBatchStream) -> Result<reqwest::Body> {
-    let stream = stream_as_ipc(data)?;
-    Ok(reqwest::Body::wrap_stream(stream))
+        for batch in batches {
+            let batch = batch?;
+            writer.write(&batch)?;
+        }
+        writer.finish()?;
+    }
+    Ok(buf.into_inner())
 }

 pub fn parse_server_version(req_id: &str, rsp: &Response) -> Result<ServerVersion> {
--- a/rust/lancedb/src/table.rs
+++ b/rust/lancedb/src/table.rs
@@ -5,7 +5,7 @@

 use arrow::array::{AsArray, FixedSizeListBuilder, Float32Builder};
 use arrow::datatypes::{Float32Type, UInt8Type};
-use arrow_array::{RecordBatch, RecordBatchReader};
+use arrow_array::{RecordBatchIterator, RecordBatchReader};
 use arrow_schema::{DataType, Field, Schema, SchemaRef};
 use async_trait::async_trait;
 use datafusion_expr::Expr;
@@ -50,9 +50,10 @@ use std::format;
 use std::path::Path;
 use std::sync::Arc;

-use crate::data::scannable::{scannable_with_embeddings, Scannable};
+use crate::arrow::IntoArrow;
+use crate::connection::NoData;
 use crate::database::Database;
-use crate::embeddings::{EmbeddingDefinition, EmbeddingRegistry, MemoryRegistry};
+use crate::embeddings::{EmbeddingDefinition, EmbeddingRegistry, MaybeEmbedded, MemoryRegistry};
 use crate::error::{Error, Result};
 use crate::index::vector::VectorIndex;
 use crate::index::IndexStatistics;
@@ -71,7 +72,6 @@ use crate::utils::{
 use self::dataset::DatasetConsistencyWrapper;
 use self::merge::MergeInsertBuilder;

-mod add_data;
 pub mod datafusion;
 pub(crate) mod dataset;
 pub mod delete;
@@ -80,8 +80,6 @@ pub mod optimize;
 pub mod schema_evolution;
 pub mod update;

-pub use add_data::{AddDataBuilder, AddDataMode, AddResult};
-
 use crate::index::waiter::wait_for_index;
 pub use chrono::Duration;
 pub use delete::DeleteResult;
@@ -198,6 +196,60 @@ pub struct WriteOptions {
    pub lance_write_params: Option<WriteParams>,
 }

+#[derive(Debug, Clone, Default)]
+pub enum AddDataMode {
+    /// Rows will be appended to the table (the default)
+    #[default]
+    Append,
+    /// The existing table will be overwritten with the new data
+    Overwrite,
+}
+
+/// A builder for configuring a [`crate::connection::Connection::create_table`] or [`Table::add`]
+/// operation
+pub struct AddDataBuilder<T: IntoArrow> {
+    parent: Arc<dyn BaseTable>,
+    pub(crate) data: T,
+    pub(crate) mode: AddDataMode,
+    pub(crate) write_options: WriteOptions,
+    embedding_registry: Option<Arc<dyn EmbeddingRegistry>>,
+}
+
+impl<T: IntoArrow> std::fmt::Debug for AddDataBuilder<T> {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("AddDataBuilder")
+            .field("parent", &self.parent)
+            .field("mode", &self.mode)
+            .field("write_options", &self.write_options)
+            .finish()
+    }
+}
+
+impl<T: IntoArrow> AddDataBuilder<T> {
+    pub fn mode(mut self, mode: AddDataMode) -> Self {
+        self.mode = mode;
+        self
+    }
+
+    pub fn write_options(mut self, options: WriteOptions) -> Self {
+        self.write_options = options;
+        self
+    }
+
+    pub async fn execute(self) -> Result<AddResult> {
+        let parent = self.parent.clone();
+        let data = self.data.into_arrow()?;
+        let without_data = AddDataBuilder::<NoData> {
+            data: NoData {},
+            mode: self.mode,
+            parent: self.parent,
+            write_options: self.write_options,
+            embedding_registry: self.embedding_registry,
+        };
+        parent.add(without_data, data).await
+    }
+}
+
 /// Filters that can be used to limit the rows returned by a query
 pub enum Filter {
    /// A SQL filter string
@@ -231,6 +283,15 @@ pub trait Tags: Send + Sync {
    async fn update(&mut self, tag: &str, version: u64) -> Result<()>;
 }

+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Default)]
+pub struct AddResult {
+    // The commit version associated with the operation.
+    // A version of `0` indicates compatibility with legacy servers that do not return
+    /// a commit version.
+    #[serde(default)]
+    pub version: u64,
+}
+
 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Default)]
 pub struct MergeResult {
    // The commit version associated with the operation.
@@ -303,7 +364,11 @@ pub trait BaseTable: std::fmt::Display + std::fmt::Debug + Send + Sync {
    ) -> Result<String>;

    /// Add new records to the table.
-    async fn add(&self, add: AddDataBuilder) -> Result<AddResult>;
+    async fn add(
+        &self,
+        add: AddDataBuilder<NoData>,
+        data: Box<dyn arrow_array::RecordBatchReader + Send>,
+    ) -> Result<AddResult>;
    /// Delete rows from the table.
    async fn delete(&self, predicate: &str) -> Result<DeleteResult>;
    /// Update rows in the table.
@@ -448,30 +513,6 @@ mod test_utils {
                embedding_registry: Arc::new(MemoryRegistry::new()),
            }
        }
-
-        pub fn new_with_handler_and_config<T>(
-            name: impl Into<String>,
-            handler: impl Fn(reqwest::Request) -> http::Response<T> + Clone + Send + Sync + 'static,
-            config: crate::remote::ClientConfig,
-        ) -> Self
-        where
-            T: Into<reqwest::Body>,
-        {
-            let inner = Arc::new(crate::remote::table::RemoteTable::new_mock_with_config(
-                name.into(),
-                handler.clone(),
-                config.clone(),
-            ));
-            let database = Arc::new(crate::remote::db::RemoteDatabase::new_mock_with_config(
-                handler, config,
-            ));
-            Self {
-                inner,
-                database: Some(database),
-                // Registry is unused.
-                embedding_registry: Arc::new(MemoryRegistry::new()),
-            }
-        }
    }
 }

@@ -572,14 +613,16 @@ impl Table {
    ///
    /// # Arguments
    ///
-    /// * `data` data to be added to the Table
+    /// * `batches` data to be added to the Table
    /// * `options` options to control how data is added
-    pub fn add<T: Scannable + 'static>(&self, data: T) -> AddDataBuilder {
-        AddDataBuilder::new(
-            self.inner.clone(),
-            Box::new(data),
-            Some(self.embedding_registry.clone()),
-        )
+    pub fn add<T: IntoArrow>(&self, batches: T) -> AddDataBuilder<T> {
+        AddDataBuilder {
+            parent: self.inner.clone(),
+            data: batches,
+            mode: AddDataMode::Append,
+            write_options: WriteOptions::default(),
+            embedding_registry: Some(self.embedding_registry.clone()),
+        }
    }

    /// Update existing records in the Table
@@ -618,26 +661,31 @@ impl Table {
    ///     .execute()
    ///     .await
    ///     .unwrap();
-    /// let schema = Arc::new(Schema::new(vec![
-    ///     Field::new("id", DataType::Int32, false),
-    ///     Field::new("vector", DataType::FixedSizeList(
-    ///         Arc::new(Field::new("item", DataType::Float32, true)), 128), true),
-    /// ]));
-    /// let data = RecordBatch::try_new(
-    ///     schema.clone(),
-    ///     vec![
-    ///         Arc::new(Int32Array::from_iter_values(0..10)),
-    ///         Arc::new(
-    ///             FixedSizeListArray::from_iter_primitive::<Float32Type, _, _>(
-    ///                 (0..10).map(|_| Some(vec![Some(1.0); 128])),
-    ///                 128,
+    /// # let schema = Arc::new(Schema::new(vec![
+    /// #  Field::new("id", DataType::Int32, false),
+    /// #  Field::new("vector", DataType::FixedSizeList(
+    /// #    Arc::new(Field::new("item", DataType::Float32, true)), 128), true),
+    /// # ]));
+    /// let batches = RecordBatchIterator::new(
+    ///     vec![RecordBatch::try_new(
+    ///         schema.clone(),
+    ///         vec![
+    ///             Arc::new(Int32Array::from_iter_values(0..10)),
+    ///             Arc::new(
+    ///                 FixedSizeListArray::from_iter_primitive::<Float32Type, _, _>(
+    ///                     (0..10).map(|_| Some(vec![Some(1.0); 128])),
+    ///                     128,
+    ///                 ),
    ///             ),
-    ///         ),
-    ///     ],
-    /// )
-    /// .unwrap();
+    ///         ],
+    ///     )
+    ///     .unwrap()]
+    ///     .into_iter()
+    ///     .map(Ok),
+    ///     schema.clone(),
+    /// );
    /// let tbl = db
-    ///     .create_table("delete_test", data)
+    ///     .create_table("delete_test", Box::new(batches))
    ///     .execute()
    ///     .await
    ///     .unwrap();
@@ -1397,7 +1445,7 @@ impl NativeTable {
                    name: name.to_string(),
                    source: Box::new(e),
                },
-                e => e.into(),
+                source => Error::Lance { source },
            })?;

        let dataset = DatasetConsistencyWrapper::new_latest(dataset, read_consistency_interval);
@@ -1481,7 +1529,7 @@ impl NativeTable {
                lance::Error::Namespace { source, .. } => Error::Runtime {
                    message: format!("Failed to get table info from namespace: {:?}", source),
                },
-                e => e.into(),
+                source => Error::Lance { source },
            })?;

        let dataset = builder
@@ -1493,7 +1541,7 @@ impl NativeTable {
                    name: name.to_string(),
                    source: Box::new(e),
                },
-                e => e.into(),
+                source => Error::Lance { source },
            })?;

        let uri = dataset.uri().to_string();
@@ -1587,7 +1635,7 @@ impl NativeTable {
                lance::Error::DatasetAlreadyExists { .. } => Error::TableAlreadyExists {
                    name: name.to_string(),
                },
-                e => e.into(),
+                source => Error::Lance { source },
            })?;

        let id = Self::build_id(&namespace, name);
@@ -1614,12 +1662,12 @@ impl NativeTable {
        read_consistency_interval: Option<std::time::Duration>,
        namespace_client: Option<Arc<dyn LanceNamespace>>,
    ) -> Result<Self> {
-        let data: Box<dyn Scannable> = Box::new(RecordBatch::new_empty(schema));
+        let batches = RecordBatchIterator::new(vec![], schema);
        Self::create(
            uri,
            name,
            namespace,
-            data,
+            batches,
            write_store_wrapper,
            params,
            read_consistency_interval,
@@ -1708,7 +1756,7 @@ impl NativeTable {
                lance::Error::DatasetAlreadyExists { .. } => Error::TableAlreadyExists {
                    name: name.to_string(),
                },
-                e => e.into(),
+                source => Error::Lance { source },
            })?;

        let id = Self::build_id(&namespace, name);
@@ -2490,7 +2538,17 @@ impl BaseTable for NativeTable {
        }
    }

-    async fn add(&self, add: AddDataBuilder) -> Result<AddResult> {
+    async fn add(
+        &self,
+        add: AddDataBuilder<NoData>,
+        data: Box<dyn RecordBatchReader + Send>,
+    ) -> Result<AddResult> {
+        let data = Box::new(MaybeEmbedded::try_new(
+            data,
+            self.table_definition().await?,
+            add.embedding_registry,
+        )?) as Box<dyn RecordBatchReader + Send>;
+
        let lance_params = add.write_options.lance_write_params.unwrap_or(WriteParams {
            mode: match add.mode {
                AddDataMode::Append => WriteMode::Append,
@@ -2499,11 +2557,6 @@ impl BaseTable for NativeTable {
            ..Default::default()
        });

-        // Apply embeddings if configured
-        let table_def = self.table_definition().await?;
-        let data =
-            scannable_with_embeddings(add.data, &table_def, add.embedding_registry.as_ref())?;
-
        let dataset = {
            // Limited scope for the mutable borrow of self.dataset avoids deadlock.
            let ds = self.dataset.get_mut().await?;
@@ -3110,7 +3163,7 @@ mod tests {
    use arrow_array::{BinaryArray, LargeBinaryArray};
    use arrow_data::ArrayDataBuilder;
    use arrow_schema::{DataType, Field, Schema};
-    use futures::TryStreamExt;
+    use lance::dataset::WriteMode;
    use lance::io::{ObjectStoreParams, WrappingObjectStore};
    use lance::Dataset;
    use rand::Rng;
@@ -3121,17 +3174,14 @@ mod tests {
    use crate::connection::ConnectBuilder;
    use crate::index::scalar::{BTreeIndexBuilder, BitmapIndexBuilder};
    use crate::index::vector::{IvfHnswPqIndexBuilder, IvfHnswSqIndexBuilder};
-    use crate::query::{ExecutableQuery, QueryBase};
-    use crate::test_utils::connection::new_test_connection;

    #[tokio::test]
    async fn test_open() {
        let tmp_dir = tempdir().unwrap();
        let dataset_path = tmp_dir.path().join("test.lance");

-        let batch = make_test_batches();
-        let reader = RecordBatchIterator::new(vec![Ok(batch.clone())], batch.schema());
-        Dataset::write(reader, dataset_path.to_str().unwrap(), None)
+        let batches = make_test_batches();
+        Dataset::write(batches, dataset_path.to_str().unwrap(), None)
            .await
            .unwrap();

@@ -3164,12 +3214,9 @@ mod tests {
        let tmp_dir = tempdir().unwrap();
        let uri = tmp_dir.path().to_str().unwrap();

-        let batch = make_test_batches();
-        let reader: Box<dyn RecordBatchReader + Send> = Box::new(RecordBatchIterator::new(
-            vec![Ok(batch.clone())],
-            batch.schema(),
-        ));
-        let table = NativeTable::create(uri, "test", vec![], reader, None, None, None, None)
+        let batches = make_test_batches();
+        let batches = Box::new(batches) as Box<dyn RecordBatchReader + Send>;
+        let table = NativeTable::create(uri, "test", vec![], batches, None, None, None, None)
            .await
            .unwrap();

@@ -3183,6 +3230,33 @@ mod tests {
        );
    }

+    #[tokio::test]
+    async fn test_add() {
+        let tmp_dir = tempdir().unwrap();
+        let uri = tmp_dir.path().to_str().unwrap();
+        let conn = connect(uri).execute().await.unwrap();
+
+        let batches = make_test_batches();
+        let schema = batches.schema().clone();
+        let table = conn.create_table("test", batches).execute().await.unwrap();
+        assert_eq!(table.count_rows(None).await.unwrap(), 10);
+
+        let new_batches = RecordBatchIterator::new(
+            vec![RecordBatch::try_new(
+                schema.clone(),
+                vec![Arc::new(Int32Array::from_iter_values(100..110))],
+            )
+            .unwrap()]
+            .into_iter()
+            .map(Ok),
+            schema.clone(),
+        );
+
+        table.add(new_batches).execute().await.unwrap();
+        assert_eq!(table.count_rows(None).await.unwrap(), 20);
+        assert_eq!(table.name(), "test");
+    }
+
    #[tokio::test]
    async fn test_merge_insert() {
        let tmp_dir = tempdir().unwrap();
@@ -3199,7 +3273,7 @@ mod tests {
        assert_eq!(table.count_rows(None).await.unwrap(), 10);

        // Create new data with i=5..15
-        let new_batches = merge_insert_test_batches(5, 1);
+        let new_batches = Box::new(merge_insert_test_batches(5, 1));

        // Perform a "insert if not exists"
        let mut merge_insert_builder = table.merge_insert(&["i"]);
@@ -3213,7 +3287,7 @@ mod tests {
        assert_eq!(result.num_attempts, 1);

        // Create new data with i=15..25 (no id matches)
-        let new_batches = merge_insert_test_batches(15, 2);
+        let new_batches = Box::new(merge_insert_test_batches(15, 2));
        // Perform a "bulk update" (should not affect anything)
        let mut merge_insert_builder = table.merge_insert(&["i"]);
        merge_insert_builder.when_matched_update_all(None);
@@ -3226,7 +3300,7 @@ mod tests {
        );

        // Conditional update that only replaces the age=0 data
-        let new_batches = merge_insert_test_batches(5, 3);
+        let new_batches = Box::new(merge_insert_test_batches(5, 3));
        let mut merge_insert_builder = table.merge_insert(&["i"]);
        merge_insert_builder.when_matched_update_all(Some("target.age = 0".to_string()));
        merge_insert_builder.execute(new_batches).await.unwrap();
@@ -3252,7 +3326,7 @@ mod tests {
        assert_eq!(table.count_rows(None).await.unwrap(), 10);

        // Test use_index=true (default behavior)
-        let new_batches = merge_insert_test_batches(5, 1);
+        let new_batches = Box::new(merge_insert_test_batches(5, 1));
        let mut merge_insert_builder = table.merge_insert(&["i"]);
        merge_insert_builder.when_not_matched_insert_all();
        merge_insert_builder.use_index(true);
@@ -3260,7 +3334,7 @@ mod tests {
        assert_eq!(table.count_rows(None).await.unwrap(), 15);

        // Test use_index=false (force table scan)
-        let new_batches = merge_insert_test_batches(15, 2);
+        let new_batches = Box::new(merge_insert_test_batches(15, 2));
        let mut merge_insert_builder = table.merge_insert(&["i"]);
        merge_insert_builder.when_not_matched_insert_all();
        merge_insert_builder.use_index(false);
@@ -3268,6 +3342,59 @@ mod tests {
        assert_eq!(table.count_rows(None).await.unwrap(), 25);
    }

+    #[tokio::test]
+    async fn test_add_overwrite() {
+        let tmp_dir = tempdir().unwrap();
+        let uri = tmp_dir.path().to_str().unwrap();
+        let conn = connect(uri).execute().await.unwrap();
+
+        let batches = make_test_batches();
+        let schema = batches.schema().clone();
+        let table = conn.create_table("test", batches).execute().await.unwrap();
+        assert_eq!(table.count_rows(None).await.unwrap(), 10);
+
+        let batches = vec![RecordBatch::try_new(
+            schema.clone(),
+            vec![Arc::new(Int32Array::from_iter_values(100..110))],
+        )
+        .unwrap()]
+        .into_iter()
+        .map(Ok);
+
+        let new_batches = RecordBatchIterator::new(batches.clone(), schema.clone());
+
+        // Can overwrite using AddDataOptions::mode
+        table
+            .add(new_batches)
+            .mode(AddDataMode::Overwrite)
+            .execute()
+            .await
+            .unwrap();
+        assert_eq!(table.count_rows(None).await.unwrap(), 10);
+        assert_eq!(table.name(), "test");
+
+        // Can overwrite using underlying WriteParams (which
+        // take precedence over AddDataOptions::mode)
+
+        let param: WriteParams = WriteParams {
+            mode: WriteMode::Overwrite,
+            ..Default::default()
+        };
+
+        let new_batches = RecordBatchIterator::new(batches.clone(), schema.clone());
+        table
+            .add(new_batches)
+            .write_options(WriteOptions {
+                lance_write_params: Some(param),
+            })
+            .mode(AddDataMode::Append)
+            .execute()
+            .await
+            .unwrap();
+        assert_eq!(table.count_rows(None).await.unwrap(), 10);
+        assert_eq!(table.name(), "test");
+    }
+
    #[derive(Default, Debug)]
    struct NoOpCacheWrapper {
        called: AtomicBool,
@@ -3323,25 +3450,35 @@ mod tests {
        assert!(wrapper.called());
    }

-    fn merge_insert_test_batches(offset: i32, age: i32) -> Box<dyn RecordBatchReader + Send> {
+    fn merge_insert_test_batches(
+        offset: i32,
+        age: i32,
+    ) -> impl RecordBatchReader + Send + Sync + 'static {
        let schema = Arc::new(Schema::new(vec![
            Field::new("i", DataType::Int32, false),
            Field::new("age", DataType::Int32, false),
        ]));
-        let batch = RecordBatch::try_new(
-            schema.clone(),
-            vec![
-                Arc::new(Int32Array::from_iter_values(offset..(offset + 10))),
-                Arc::new(Int32Array::from_iter_values(std::iter::repeat_n(age, 10))),
-            ],
+        RecordBatchIterator::new(
+            vec![RecordBatch::try_new(
+                schema.clone(),
+                vec![
+                    Arc::new(Int32Array::from_iter_values(offset..(offset + 10))),
+                    Arc::new(Int32Array::from_iter_values(std::iter::repeat_n(age, 10))),
+                ],
+            )],
+            schema,
        )
-        .unwrap();
-        Box::new(RecordBatchIterator::new(vec![Ok(batch)], schema))
    }

-    fn make_test_batches() -> RecordBatch {
+    fn make_test_batches() -> impl RecordBatchReader + Send + Sync + 'static {
        let schema = Arc::new(Schema::new(vec![Field::new("i", DataType::Int32, false)]));
-        RecordBatch::try_new(schema, vec![Arc::new(Int32Array::from_iter_values(0..10))]).unwrap()
+        RecordBatchIterator::new(
+            vec![RecordBatch::try_new(
+                schema.clone(),
+                vec![Arc::new(Int32Array::from_iter_values(0..10))],
+            )],
+            schema,
+        )
    }

    #[tokio::test]
@@ -3429,9 +3566,14 @@ mod tests {
        );

        let vectors = Arc::new(create_fixed_size_list(float_arr, dimension).unwrap());
-        let batch = RecordBatch::try_new(schema.clone(), vec![vectors.clone()]).unwrap();
+        let batches = RecordBatchIterator::new(
+            vec![RecordBatch::try_new(schema.clone(), vec![vectors.clone()]).unwrap()]
+                .into_iter()
+                .map(Ok),
+            schema,
+        );

-        let table = conn.create_table("test", batch).execute().await.unwrap();
+        let table = conn.create_table("test", batches).execute().await.unwrap();

        assert_eq!(table.index_stats("my_index").await.unwrap(), None);

@@ -3462,31 +3604,6 @@ mod tests {
        assert_eq!(table.list_indices().await.unwrap().len(), 0);
    }

-    #[tokio::test]
-    async fn test_dynamic_select() {
-        let tc = new_test_connection().await.unwrap();
-        let db = tc.connection;
-
-        let table = db
-            .create_table("test", some_sample_data())
-            .execute()
-            .await
-            .unwrap();
-
-        let query = table.query().select(Select::dynamic(&[("i_alias", "i")]));
-
-        let result = query.execute().await;
-        let batches = result
-            .expect("should have result")
-            .try_collect::<Vec<_>>()
-            .await
-            .unwrap();
-
-        for batch in batches {
-            assert!(batch.column_by_name("i_alias").is_some());
-        }
-    }
-
    #[tokio::test]
    async fn test_ivf_pq_uses_default_partition_size_for_num_partitions() {
        use arrow_array::{Float32Array, RecordBatch};
@@ -3513,9 +3630,14 @@ mod tests {
        let float_arr =
            Float32Array::from_iter_values((0..(num_rows * dimension)).map(|v| v as f32));
        let vectors = Arc::new(create_fixed_size_list(float_arr, dimension as i32).unwrap());
-        let batch = RecordBatch::try_new(schema.clone(), vec![vectors]).unwrap();
+        let batches = RecordBatchIterator::new(
+            vec![RecordBatch::try_new(schema.clone(), vec![vectors]).unwrap()]
+                .into_iter()
+                .map(Ok),
+            schema,
+        );

-        let table = conn.create_table("test", batch).execute().await.unwrap();
+        let table = conn.create_table("test", batches).execute().await.unwrap();
        let native_table = table.as_native().unwrap();
        let builder = IvfPqIndexBuilder::default();
        table
@@ -3585,9 +3707,14 @@ mod tests {
        );

        let vectors = Arc::new(create_fixed_size_list(float_arr, dimension).unwrap());
-        let batch = RecordBatch::try_new(schema.clone(), vec![vectors.clone()]).unwrap();
+        let batches = RecordBatchIterator::new(
+            vec![RecordBatch::try_new(schema.clone(), vec![vectors.clone()]).unwrap()]
+                .into_iter()
+                .map(Ok),
+            schema,
+        );

-        let table = conn.create_table("test", batch).execute().await.unwrap();
+        let table = conn.create_table("test", batches).execute().await.unwrap();

        let stats = table.index_stats("my_index").await.unwrap();
        assert!(stats.is_none());
@@ -3645,9 +3772,14 @@ mod tests {
        );

        let vectors = Arc::new(create_fixed_size_list(float_arr, dimension).unwrap());
-        let batch = RecordBatch::try_new(schema.clone(), vec![vectors.clone()]).unwrap();
+        let batches = RecordBatchIterator::new(
+            vec![RecordBatch::try_new(schema.clone(), vec![vectors.clone()]).unwrap()]
+                .into_iter()
+                .map(Ok),
+            schema,
+        );

-        let table = conn.create_table("test", batch).execute().await.unwrap();
+        let table = conn.create_table("test", batches).execute().await.unwrap();
        let stats = table.index_stats("my_index").await.unwrap();
        assert!(stats.is_none());

@@ -3690,7 +3822,7 @@ mod tests {
        Ok(FixedSizeListArray::from(data))
    }

-    fn some_sample_data() -> Box<dyn arrow_array::RecordBatchReader + Send> {
+    fn some_sample_data() -> Box<dyn RecordBatchReader + Send> {
        let batch = RecordBatch::try_new(
            Arc::new(Schema::new(vec![Field::new("i", DataType::Int32, false)])),
            vec![Arc::new(Int32Array::from(vec![1]))],
@@ -3714,7 +3846,10 @@ mod tests {
        .unwrap();
        let conn = ConnectBuilder::new(uri).execute().await.unwrap();
        let table = conn
-            .create_table("my_table", batch.clone())
+            .create_table(
+                "my_table",
+                RecordBatchIterator::new(vec![Ok(batch.clone())], batch.schema()),
+            )
            .execute()
            .await
            .unwrap();
@@ -3793,7 +3928,10 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_bitmap", batch.clone())
+            .create_table(
+                "test_bitmap",
+                RecordBatchIterator::new(vec![Ok(batch.clone())], batch.schema()),
+            )
            .execute()
            .await
            .unwrap();
@@ -3894,7 +4032,10 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_bitmap", batch.clone())
+            .create_table(
+                "test_bitmap",
+                RecordBatchIterator::new(vec![Ok(batch.clone())], batch.schema()),
+            )
            .execute()
            .await
            .unwrap();
@@ -3954,7 +4095,10 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_bitmap", batch.clone())
+            .create_table(
+                "test_bitmap",
+                RecordBatchIterator::new(vec![Ok(batch.clone())], batch.schema()),
+            )
            .execute()
            .await
            .unwrap();
@@ -3999,7 +4143,7 @@ mod tests {

            let conn1 = ConnectBuilder::new(uri).execute().await.unwrap();
            let table1 = conn1
-                .create_empty_table("my_table", RecordBatchReader::schema(&data))
+                .create_empty_table("my_table", data.schema())
                .execute()
                .await
                .unwrap();
@@ -4269,7 +4413,10 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_stats", batch.clone())
+            .create_table(
+                "test_stats",
+                RecordBatchIterator::new(vec![Ok(batch.clone())], batch.schema()),
+            )
            .execute()
            .await
            .unwrap();
@@ -4282,11 +4429,21 @@ mod tests {
                ],
            )
            .unwrap();
-            table.add(batch.clone()).execute().await.unwrap();
+            table
+                .add(RecordBatchIterator::new(
+                    vec![Ok(batch.clone())],
+                    batch.schema(),
+                ))
+                .execute()
+                .await
+                .unwrap();
        }

        let empty_table = conn
-            .create_table("test_stats_empty", RecordBatch::new_empty(batch.schema()))
+            .create_table(
+                "test_stats_empty",
+                RecordBatchIterator::new(vec![], batch.schema()),
+            )
            .execute()
            .await
            .unwrap();
@@ -4360,12 +4517,22 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_list_indices_skip_frag_reuse", batch.clone())
+            .create_table(
+                "test_list_indices_skip_frag_reuse",
+                RecordBatchIterator::new(vec![Ok(batch.clone())], batch.schema()),
+            )
            .execute()
            .await
            .unwrap();

-        table.add(batch.clone()).execute().await.unwrap();
+        table
+            .add(RecordBatchIterator::new(
+                vec![Ok(batch.clone())],
+                batch.schema(),
+            ))
+            .execute()
+            .await
+            .unwrap();

        table
            .create_index(&["id"], Index::Bitmap(BitmapIndexBuilder {}))
@@ -4395,9 +4562,8 @@ mod tests {
        let tmp_dir = tempdir().unwrap();
        let dataset_path = tmp_dir.path().join("test_ns_query.lance");

-        let batch = make_test_batches();
-        let reader = RecordBatchIterator::new(vec![Ok(batch.clone())], batch.schema());
-        Dataset::write(reader, dataset_path.to_str().unwrap(), None)
+        let batches = make_test_batches();
+        Dataset::write(batches, dataset_path.to_str().unwrap(), None)
            .await
            .unwrap();

@@ -4449,9 +4615,8 @@ mod tests {
        let tmp_dir = tempdir().unwrap();
        let dataset_path = tmp_dir.path().join("test_ns_plain.lance");

-        let batch = make_test_batches();
-        let reader = RecordBatchIterator::new(vec![Ok(batch.clone())], batch.schema());
-        Dataset::write(reader, dataset_path.to_str().unwrap(), None)
+        let batches = make_test_batches();
+        Dataset::write(batches, dataset_path.to_str().unwrap(), None)
            .await
            .unwrap();

--- a/rust/lancedb/src/table/add_data.rs
+++ b/rust/lancedb/src/table/add_data.rs
@@ -1,343 +0,0 @@
-// SPDX-License-Identifier: Apache-2.0
-// SPDX-FileCopyrightText: Copyright The LanceDB Authors
-
-use std::sync::Arc;
-
-use serde::{Deserialize, Serialize};
-
-use crate::data::scannable::Scannable;
-use crate::embeddings::EmbeddingRegistry;
-use crate::Result;
-
-use super::{BaseTable, WriteOptions};
-
-#[derive(Debug, Clone, Default)]
-pub enum AddDataMode {
-    /// Rows will be appended to the table (the default)
-    #[default]
-    Append,
-    /// The existing table will be overwritten with the new data
-    Overwrite,
-}
-
-#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Default)]
-pub struct AddResult {
-    // The commit version associated with the operation.
-    // A version of `0` indicates compatibility with legacy servers that do not return
-    /// a commit version.
-    #[serde(default)]
-    pub version: u64,
-}
-
-/// A builder for configuring a [`crate::table::Table::add`] operation
-pub struct AddDataBuilder {
-    pub(crate) parent: Arc<dyn BaseTable>,
-    pub(crate) data: Box<dyn Scannable>,
-    pub(crate) mode: AddDataMode,
-    pub(crate) write_options: WriteOptions,
-    pub(crate) embedding_registry: Option<Arc<dyn EmbeddingRegistry>>,
-}
-
-impl std::fmt::Debug for AddDataBuilder {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        f.debug_struct("AddDataBuilder")
-            .field("parent", &self.parent)
-            .field("mode", &self.mode)
-            .field("write_options", &self.write_options)
-            .finish()
-    }
-}
-
-impl AddDataBuilder {
-    pub(crate) fn new(
-        parent: Arc<dyn BaseTable>,
-        data: Box<dyn Scannable>,
-        embedding_registry: Option<Arc<dyn EmbeddingRegistry>>,
-    ) -> Self {
-        Self {
-            parent,
-            data,
-            mode: AddDataMode::Append,
-            write_options: WriteOptions::default(),
-            embedding_registry,
-        }
-    }
-
-    pub fn mode(mut self, mode: AddDataMode) -> Self {
-        self.mode = mode;
-        self
-    }
-
-    pub fn write_options(mut self, options: WriteOptions) -> Self {
-        self.write_options = options;
-        self
-    }
-
-    pub async fn execute(self) -> Result<AddResult> {
-        self.parent.clone().add(self).await
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use std::sync::Arc;
-
-    use arrow_array::{record_batch, RecordBatch, RecordBatchIterator};
-    use arrow_schema::{ArrowError, DataType, Field, Schema};
-    use futures::TryStreamExt;
-    use lance::dataset::{WriteMode, WriteParams};
-
-    use crate::arrow::{SendableRecordBatchStream, SimpleRecordBatchStream};
-    use crate::connect;
-    use crate::data::scannable::Scannable;
-    use crate::embeddings::{
-        EmbeddingDefinition, EmbeddingFunction, EmbeddingRegistry, MemoryRegistry,
-    };
-    use crate::query::{ExecutableQuery, QueryBase, Select};
-    use crate::table::{ColumnDefinition, ColumnKind, Table, TableDefinition, WriteOptions};
-    use crate::test_utils::embeddings::MockEmbed;
-    use crate::Error;
-
-    use super::AddDataMode;
-
-    async fn create_test_table() -> Table {
-        let conn = connect("memory://").execute().await.unwrap();
-        let batch = record_batch!(("id", Int64, [1, 2, 3])).unwrap();
-        conn.create_table("test", batch).execute().await.unwrap()
-    }
-
-    async fn test_add_with_data<T>(data: T)
-    where
-        T: Scannable + 'static,
-    {
-        let table = create_test_table().await;
-        let schema = data.schema();
-        table.add(data).execute().await.unwrap();
-        assert_eq!(table.count_rows(None).await.unwrap(), 5); // 3 initial + 2 added
-        assert_eq!(table.schema().await.unwrap(), schema);
-    }
-
-    #[tokio::test]
-    async fn test_add_with_batch() {
-        let batch = record_batch!(("id", Int64, [4, 5])).unwrap();
-        test_add_with_data(batch).await;
-    }
-
-    #[tokio::test]
-    async fn test_add_with_vec_batch() {
-        let data = vec![
-            record_batch!(("id", Int64, [4])).unwrap(),
-            record_batch!(("id", Int64, [5])).unwrap(),
-        ];
-        test_add_with_data(data).await;
-    }
-
-    #[tokio::test]
-    async fn test_add_with_record_batch_reader() {
-        let data = vec![
-            record_batch!(("id", Int64, [4])).unwrap(),
-            record_batch!(("id", Int64, [5])).unwrap(),
-        ];
-        let schema = data[0].schema();
-        let reader: Box<dyn arrow_array::RecordBatchReader + Send> = Box::new(
-            RecordBatchIterator::new(data.into_iter().map(Ok), schema.clone()),
-        );
-        test_add_with_data(reader).await;
-    }
-
-    #[tokio::test]
-    async fn test_add_with_stream() {
-        let data = vec![
-            record_batch!(("id", Int64, [4])).unwrap(),
-            record_batch!(("id", Int64, [5])).unwrap(),
-        ];
-        let schema = data[0].schema();
-        let inner = futures::stream::iter(data.into_iter().map(Ok));
-        let stream: SendableRecordBatchStream = Box::pin(SimpleRecordBatchStream {
-            schema,
-            stream: inner,
-        });
-        test_add_with_data(stream).await;
-    }
-
-    #[derive(Debug)]
-    struct MyError;
-
-    impl std::fmt::Display for MyError {
-        fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-            write!(f, "MyError occurred")
-        }
-    }
-
-    impl std::error::Error for MyError {}
-
-    #[tokio::test]
-    async fn test_add_preserves_reader_error() {
-        let table = create_test_table().await;
-        let first_batch = record_batch!(("id", Int64, [4])).unwrap();
-        let schema = first_batch.schema();
-        let iterator = vec![
-            Ok(first_batch),
-            Err(ArrowError::ExternalError(Box::new(MyError))),
-        ];
-        let reader: Box<dyn arrow_array::RecordBatchReader + Send> = Box::new(
-            RecordBatchIterator::new(iterator.into_iter(), schema.clone()),
-        );
-
-        let result = table.add(reader).execute().await;
-
-        assert!(result.is_err());
-    }
-
-    #[tokio::test]
-    async fn test_add_preserves_stream_error() {
-        let table = create_test_table().await;
-        let first_batch = record_batch!(("id", Int64, [4])).unwrap();
-        let schema = first_batch.schema();
-        let iterator = vec![
-            Ok(first_batch),
-            Err(Error::External {
-                source: Box::new(MyError),
-            }),
-        ];
-        let stream = futures::stream::iter(iterator);
-        let stream: SendableRecordBatchStream = Box::pin(SimpleRecordBatchStream {
-            schema: schema.clone(),
-            stream,
-        });
-
-        let result = table.add(stream).execute().await;
-
-        assert!(result.is_err());
-    }
-
-    #[tokio::test]
-    async fn test_add() {
-        let conn = connect("memory://").execute().await.unwrap();
-
-        let batch = record_batch!(("i", Int32, [0, 1, 2])).unwrap();
-        let table = conn
-            .create_table("test", batch.clone())
-            .execute()
-            .await
-            .unwrap();
-        assert_eq!(table.count_rows(None).await.unwrap(), 3);
-
-        let new_batch = record_batch!(("i", Int32, [3])).unwrap();
-        table.add(new_batch).execute().await.unwrap();
-
-        assert_eq!(table.count_rows(None).await.unwrap(), 4);
-        assert_eq!(table.schema().await.unwrap(), batch.schema());
-    }
-
-    #[tokio::test]
-    async fn test_add_overwrite() {
-        let conn = connect("memory://").execute().await.unwrap();
-
-        let batch = record_batch!(("i", Int32, [0, 1, 2])).unwrap();
-        let table = conn
-            .create_table("test", batch.clone())
-            .execute()
-            .await
-            .unwrap();
-        assert_eq!(table.count_rows(None).await.unwrap(), batch.num_rows());
-
-        let new_batch = record_batch!(("x", Float32, [0.0, 1.0])).unwrap();
-        let res = table
-            .add(new_batch.clone())
-            .mode(AddDataMode::Overwrite)
-            .execute()
-            .await
-            .unwrap();
-        assert_eq!(res.version, table.version().await.unwrap());
-        assert_eq!(table.count_rows(None).await.unwrap(), new_batch.num_rows());
-        assert_eq!(table.schema().await.unwrap(), new_batch.schema());
-
-        // Can overwrite using underlying WriteParams (which
-        // take precedence over AddDataMode)
-        let param: WriteParams = WriteParams {
-            mode: WriteMode::Overwrite,
-            ..Default::default()
-        };
-
-        table
-            .add(new_batch.clone())
-            .write_options(WriteOptions {
-                lance_write_params: Some(param),
-            })
-            .mode(AddDataMode::Append)
-            .execute()
-            .await
-            .unwrap();
-        assert_eq!(table.count_rows(None).await.unwrap(), new_batch.num_rows());
-    }
-
-    #[tokio::test]
-    async fn test_add_with_embeddings() {
-        let registry = Arc::new(MemoryRegistry::new());
-        let mock_embedding: Arc<dyn EmbeddingFunction> = Arc::new(MockEmbed::new("mock", 4));
-        registry.register("mock", mock_embedding).unwrap();
-
-        let conn = connect("memory://")
-            .embedding_registry(registry)
-            .execute()
-            .await
-            .unwrap();
-
-        let schema = Arc::new(Schema::new(vec![
-            Field::new("text", DataType::Utf8, false),
-            Field::new(
-                "text_embedding",
-                DataType::FixedSizeList(Arc::new(Field::new("item", DataType::Float32, true)), 4),
-                false,
-            ),
-        ]));
-
-        // Add embedding metadata to the schema
-        let embedding_def = EmbeddingDefinition::new("text", "mock", Some("text_embedding"));
-        let table_def = TableDefinition::new(
-            schema.clone(),
-            vec![
-                ColumnDefinition {
-                    kind: ColumnKind::Physical,
-                },
-                ColumnDefinition {
-                    kind: ColumnKind::Embedding(embedding_def),
-                },
-            ],
-        );
-        let rich_schema = table_def.into_rich_schema();
-
-        let table = conn
-            .create_empty_table("embed_test", rich_schema)
-            .execute()
-            .await
-            .unwrap();
-
-        // Now add new data WITHOUT the embedding column - it should be computed automatically
-        let new_batch = record_batch!(("text", Utf8, ["hello", "world"])).unwrap();
-        table.add(new_batch).execute().await.unwrap();
-
-        assert_eq!(table.count_rows(None).await.unwrap(), 2);
-
-        // Query to verify the embeddings were computed for the new rows
-        let results: Vec<RecordBatch> = table
-            .query()
-            .select(Select::columns(&["text", "text_embedding"]))
-            .execute()
-            .await
-            .unwrap()
-            .try_collect()
-            .await
-            .unwrap();
-
-        let total_rows: usize = results.iter().map(|b| b.num_rows()).sum();
-        assert_eq!(total_rows, 2);
-
-        // Check that all rows have embedding values (not null)
-        for batch in &results {
-            let embedding_col = batch.column(1);
-            assert_eq!(embedding_col.null_count(), 0);
-        }
-    }
-}
--- a/rust/lancedb/src/table/datafusion.rs
+++ b/rust/lancedb/src/table/datafusion.rs
@@ -287,7 +287,8 @@ pub mod tests {

    use arrow::array::AsArray;
    use arrow_array::{
-        BinaryArray, Float64Array, Int32Array, Int64Array, RecordBatch, StringArray, UInt32Array,
+        BinaryArray, Float64Array, Int32Array, Int64Array, RecordBatch, RecordBatchIterator,
+        RecordBatchReader, StringArray, UInt32Array,
    };
    use arrow_schema::{DataType, Field, Schema};
    use datafusion::{
@@ -307,7 +308,7 @@ pub mod tests {
        table::datafusion::BaseTableAdapter,
    };

-    fn make_test_batches() -> RecordBatch {
+    fn make_test_batches() -> impl RecordBatchReader + Send + Sync + 'static {
        let metadata = HashMap::from_iter(vec![("foo".to_string(), "bar".to_string())]);
        let schema = Arc::new(
            Schema::new(vec![
@@ -316,17 +317,19 @@ pub mod tests {
            ])
            .with_metadata(metadata),
        );
-        RecordBatch::try_new(
+        RecordBatchIterator::new(
+            vec![RecordBatch::try_new(
+                schema.clone(),
+                vec![
+                    Arc::new(Int32Array::from_iter_values(0..10)),
+                    Arc::new(UInt32Array::from_iter_values(0..10)),
+                ],
+            )],
            schema,
-            vec![
-                Arc::new(Int32Array::from_iter_values(0..10)),
-                Arc::new(UInt32Array::from_iter_values(0..10)),
-            ],
        )
-        .unwrap()
    }

-    fn make_tbl_two_test_batches() -> RecordBatch {
+    fn make_tbl_two_test_batches() -> impl RecordBatchReader + Send + Sync + 'static {
        let metadata = HashMap::from_iter(vec![("foo".to_string(), "bar".to_string())]);
        let schema = Arc::new(
            Schema::new(vec![
@@ -339,26 +342,28 @@ pub mod tests {
            ])
            .with_metadata(metadata),
        );
-        RecordBatch::try_new(
+        RecordBatchIterator::new(
+            vec![RecordBatch::try_new(
+                schema.clone(),
+                vec![
+                    Arc::new(Int64Array::from_iter_values(0..1000)),
+                    Arc::new(StringArray::from_iter_values(
+                        (0..1000).map(|i| i.to_string()),
+                    )),
+                    Arc::new(Float64Array::from_iter_values((0..1000).map(|i| i as f64))),
+                    Arc::new(StringArray::from_iter_values(
+                        (0..1000).map(|i| format!("{{\"i\":{}}}", i)),
+                    )),
+                    Arc::new(BinaryArray::from_iter_values(
+                        (0..1000).map(|i| (i as u32).to_be_bytes().to_vec()),
+                    )),
+                    Arc::new(StringArray::from_iter_values(
+                        (0..1000).map(|i| i.to_string()),
+                    )),
+                ],
+            )],
            schema,
-            vec![
-                Arc::new(Int64Array::from_iter_values(0..1000)),
-                Arc::new(StringArray::from_iter_values(
-                    (0..1000).map(|i| i.to_string()),
-                )),
-                Arc::new(Float64Array::from_iter_values((0..1000).map(|i| i as f64))),
-                Arc::new(StringArray::from_iter_values(
-                    (0..1000).map(|i| format!("{{\"i\":{}}}", i)),
-                )),
-                Arc::new(BinaryArray::from_iter_values(
-                    (0..1000).map(|i| (i as u32).to_be_bytes().to_vec()),
-                )),
-                Arc::new(StringArray::from_iter_values(
-                    (0..1000).map(|i| i.to_string()),
-                )),
-            ],
        )
-        .unwrap()
    }

    struct TestFixture {
--- a/rust/lancedb/src/table/datafusion/insert.rs
+++ b/rust/lancedb/src/table/datafusion/insert.rs
@@ -222,7 +222,7 @@ mod tests {
    use std::vec;

    use super::*;
-    use arrow_array::{record_batch, RecordBatchIterator};
+    use arrow_array::{record_batch, Int32Array, RecordBatchIterator};
    use datafusion::prelude::SessionContext;
    use datafusion_catalog::MemTable;
    use tempfile::tempdir;
@@ -238,8 +238,11 @@ mod tests {

        // Create initial table
        let batch = record_batch!(("id", Int32, [1, 2, 3])).unwrap();
+        let schema = batch.schema();
+        let reader = RecordBatchIterator::new(vec![Ok(batch)], schema);
+
        let table = db
-            .create_table("test_insert", batch)
+            .create_table("test_insert", Box::new(reader))
            .execute()
            .await
            .unwrap();
@@ -276,8 +279,11 @@ mod tests {

        // Create initial table with 3 rows
        let batch = record_batch!(("id", Int32, [1, 2, 3])).unwrap();
+        let schema = batch.schema();
+        let reader = RecordBatchIterator::new(vec![Ok(batch)], schema);
+
        let table = db
-            .create_table("test_overwrite", batch)
+            .create_table("test_overwrite", Box::new(reader))
            .execute()
            .await
            .unwrap();
@@ -312,9 +318,20 @@ mod tests {
        let db = connect(uri).execute().await.unwrap();

        // Create initial table
-        let batch = record_batch!(("id", Int32, [1, 2, 3])).unwrap();
+        let schema = Arc::new(ArrowSchema::new(vec![Field::new(
+            "id",
+            DataType::Int32,
+            false,
+        )]));
+        let batches = vec![RecordBatch::try_new(
+            schema.clone(),
+            vec![Arc::new(Int32Array::from(vec![1, 2, 3]))],
+        )
+        .unwrap()];
+        let reader = RecordBatchIterator::new(batches.into_iter().map(Ok), schema.clone());
+
        let table = db
-            .create_table("test_empty", batch)
+            .create_table("test_empty", Box::new(reader))
            .execute()
            .await
            .unwrap();
@@ -335,13 +352,12 @@ mod tests {
            false,
        )]));
        // Empty batches
-        let source_reader: Box<dyn arrow_array::RecordBatchReader + Send> =
-            Box::new(RecordBatchIterator::new(
-                std::iter::empty::<Result<RecordBatch, arrow_schema::ArrowError>>(),
-                source_schema,
-            ));
+        let source_reader = RecordBatchIterator::new(
+            std::iter::empty::<Result<RecordBatch, arrow_schema::ArrowError>>(),
+            source_schema,
+        );
        let source_table = db
-            .create_table("empty_source", source_reader)
+            .create_table("empty_source", Box::new(source_reader))
            .execute()
            .await
            .unwrap();
@@ -373,10 +389,20 @@ mod tests {
        let db = connect(uri).execute().await.unwrap();

        // Create initial table
-        let batch = record_batch!(("id", Int32, [1])).unwrap();
-        let schema = batch.schema();
+        let schema = Arc::new(ArrowSchema::new(vec![Field::new(
+            "id",
+            DataType::Int32,
+            true,
+        )]));
+        let batches =
+            vec![
+                RecordBatch::try_new(schema.clone(), vec![Arc::new(Int32Array::from(vec![1]))])
+                    .unwrap(),
+            ];
+        let reader = RecordBatchIterator::new(batches.into_iter().map(Ok), schema.clone());
+
        let table = db
-            .create_table("test_multi_batch", batch)
+            .create_table("test_multi_batch", Box::new(reader))
            .execute()
            .await
            .unwrap();
--- a/rust/lancedb/src/table/datafusion/udtf/fts.rs
+++ b/rust/lancedb/src/table/datafusion/udtf/fts.rs
@@ -97,7 +97,7 @@ mod tests {
        table::datafusion::BaseTableAdapter,
        Connection, Table,
    };
-    use arrow_array::{Int32Array, RecordBatch, StringArray};
+    use arrow_array::{Int32Array, RecordBatch, RecordBatchIterator, StringArray};
    use arrow_schema::{DataType, Field, Schema as ArrowSchema};
    use datafusion::prelude::SessionContext;

@@ -173,7 +173,14 @@ mod tests {

        // Create LanceDB database and table
        let db = crate::connect("memory://test").execute().await.unwrap();
-        let table = db.create_table("foo", batch).execute().await.unwrap();
+        let table = db
+            .create_table(
+                "foo",
+                RecordBatchIterator::new(vec![Ok(batch)].into_iter(), schema),
+            )
+            .execute()
+            .await
+            .unwrap();

        // Create FTS index
        table
@@ -316,7 +323,13 @@ mod tests {
            RecordBatch::try_new(metadata_schema.clone(), vec![metadata_col, extra_col]).unwrap();

        let _metadata_table = db
-            .create_table("metadata", metadata_batch.clone())
+            .create_table(
+                "metadata",
+                RecordBatchIterator::new(
+                    vec![Ok(metadata_batch.clone())].into_iter(),
+                    metadata_schema.clone(),
+                ),
+            )
            .execute()
            .await
            .unwrap();
@@ -380,7 +393,14 @@ mod tests {
        let batch =
            RecordBatch::try_new(schema.clone(), vec![id_col, text_col, category_col]).unwrap();

-        let table = db.create_table(table_name, batch).execute().await.unwrap();
+        let table = db
+            .create_table(
+                table_name,
+                RecordBatchIterator::new(vec![Ok(batch)].into_iter(), schema),
+            )
+            .execute()
+            .await
+            .unwrap();

        // Create FTS index
        table
@@ -526,7 +546,14 @@ mod tests {
        ]));
        let batch = RecordBatch::try_new(schema.clone(), vec![id_col, text_col]).unwrap();

-        let table = db.create_table("docs", batch).execute().await.unwrap();
+        let table = db
+            .create_table(
+                "docs",
+                RecordBatchIterator::new(vec![Ok(batch)].into_iter(), schema),
+            )
+            .execute()
+            .await
+            .unwrap();

        // Create FTS index with position information for phrase queries
        table
@@ -664,7 +691,14 @@ mod tests {
        let batch =
            RecordBatch::try_new(schema.clone(), vec![id_col, title_col, content_col]).unwrap();

-        let table = db.create_table("multi_col", batch).execute().await.unwrap();
+        let table = db
+            .create_table(
+                "multi_col",
+                RecordBatchIterator::new(vec![Ok(batch)].into_iter(), schema),
+            )
+            .execute()
+            .await
+            .unwrap();

        // Create FTS indices on both columns
        table
@@ -929,7 +963,13 @@ mod tests {
        let metadata_batch =
            RecordBatch::try_new(metadata_schema.clone(), vec![metadata_id, extra_info]).unwrap();
        let _metadata_table = db
-            .create_table("metadata", metadata_batch.clone())
+            .create_table(
+                "metadata",
+                RecordBatchIterator::new(
+                    vec![Ok(metadata_batch.clone())].into_iter(),
+                    metadata_schema,
+                ),
+            )
            .execute()
            .await
            .unwrap();
@@ -1318,7 +1358,14 @@ mod tests {
        ]));
        let batch = RecordBatch::try_new(schema.clone(), vec![id_col, text_col]).unwrap();

-        let table = db.create_table("docs", batch).execute().await.unwrap();
+        let table = db
+            .create_table(
+                "docs",
+                RecordBatchIterator::new(vec![Ok(batch)].into_iter(), schema),
+            )
+            .execute()
+            .await
+            .unwrap();

        // Create FTS index with position information
        table
@@ -1463,7 +1510,14 @@ mod tests {
        let batch =
            RecordBatch::try_new(schema.clone(), vec![id_col, title_col, content_col]).unwrap();

-        let table = db.create_table("docs", batch).execute().await.unwrap();
+        let table = db
+            .create_table(
+                "docs",
+                RecordBatchIterator::new(vec![Ok(batch)].into_iter(), schema),
+            )
+            .execute()
+            .await
+            .unwrap();

        // Create FTS indices on both columns
        table
@@ -1537,7 +1591,14 @@ mod tests {
        let batch =
            RecordBatch::try_new(schema.clone(), vec![id_col, title_col, content_col]).unwrap();

-        let table = db.create_table("docs", batch).execute().await.unwrap();
+        let table = db
+            .create_table(
+                "docs",
+                RecordBatchIterator::new(vec![Ok(batch)].into_iter(), schema),
+            )
+            .execute()
+            .await
+            .unwrap();

        // Create FTS indices
        table
@@ -1663,23 +1724,36 @@ mod tests {
            .unwrap();

        // Create table with simple text for n-gram testing
-        let data = RecordBatch::try_new(
+        let data = RecordBatchIterator::new(
+            vec![RecordBatch::try_new(
+                Arc::new(ArrowSchema::new(vec![
+                    Field::new("id", DataType::Int32, false),
+                    Field::new("text", DataType::Utf8, false),
+                ])),
+                vec![
+                    Arc::new(Int32Array::from(vec![1, 2, 3])),
+                    Arc::new(StringArray::from(vec![
+                        "hello world",
+                        "lance database",
+                        "lance is cool",
+                    ])),
+                ],
+            )
+            .unwrap()]
+            .into_iter()
+            .map(Ok),
            Arc::new(ArrowSchema::new(vec![
                Field::new("id", DataType::Int32, false),
                Field::new("text", DataType::Utf8, false),
            ])),
-            vec![
-                Arc::new(Int32Array::from(vec![1, 2, 3])),
-                Arc::new(StringArray::from(vec![
-                    "hello world",
-                    "lance database",
-                    "lance is cool",
-                ])),
-            ],
-        )
-        .unwrap();
+        );

-        let table = Arc::new(db.create_table("docs", data).execute().await.unwrap());
+        let table = Arc::new(
+            db.create_table("docs", Box::new(data))
+                .execute()
+                .await
+                .unwrap(),
+        );

        // Create FTS index with n-gram tokenizer (default min_ngram_length=3)
        table
@@ -1802,29 +1876,43 @@ mod tests {
            .unwrap();

        // Create table with two text columns
-        let data = RecordBatch::try_new(
+        let data = RecordBatchIterator::new(
+            vec![RecordBatch::try_new(
+                Arc::new(ArrowSchema::new(vec![
+                    Field::new("id", DataType::Int32, false),
+                    Field::new("title", DataType::Utf8, false),
+                    Field::new("content", DataType::Utf8, false),
+                ])),
+                vec![
+                    Arc::new(Int32Array::from(vec![1, 2, 3])),
+                    Arc::new(StringArray::from(vec![
+                        "Important Document",
+                        "Another Document",
+                        "Random Text",
+                    ])),
+                    Arc::new(StringArray::from(vec![
+                        "This is important information",
+                        "This has details",
+                        "Nothing special here",
+                    ])),
+                ],
+            )
+            .unwrap()]
+            .into_iter()
+            .map(Ok),
            Arc::new(ArrowSchema::new(vec![
                Field::new("id", DataType::Int32, false),
                Field::new("title", DataType::Utf8, false),
                Field::new("content", DataType::Utf8, false),
            ])),
-            vec![
-                Arc::new(Int32Array::from(vec![1, 2, 3])),
-                Arc::new(StringArray::from(vec![
-                    "Important Document",
-                    "Another Document",
-                    "Random Text",
-                ])),
-                Arc::new(StringArray::from(vec![
-                    "This is important information",
-                    "This has details",
-                    "Nothing special here",
-                ])),
-            ],
-        )
-        .unwrap();
+        );

-        let table = Arc::new(db.create_table("docs", data).execute().await.unwrap());
+        let table = Arc::new(
+            db.create_table("docs", Box::new(data))
+                .execute()
+                .await
+                .unwrap(),
+        );

        // Create FTS indices on both columns
        table
--- a/rust/lancedb/src/table/dataset.rs
+++ b/rust/lancedb/src/table/dataset.rs
@@ -57,6 +57,15 @@ impl DatasetRef {
        matches!(self, Self::Latest { .. })
    }

+    async fn need_reload(&self) -> Result<bool> {
+        Ok(match self {
+            Self::Latest { dataset, .. } => {
+                dataset.latest_version_id().await? != dataset.version().version
+            }
+            Self::TimeTravel { dataset, version } => dataset.version().version != *version,
+        })
+    }
+
    async fn as_latest(&mut self, read_consistency_interval: Option<Duration>) -> Result<()> {
        match self {
            Self::Latest { .. } => Ok(()),
@@ -109,21 +118,6 @@ impl DatasetRef {
        Ok(())
    }

-    fn is_up_to_date(&self) -> bool {
-        match self {
-            Self::Latest {
-                read_consistency_interval,
-                last_consistency_check,
-                ..
-            } => match (read_consistency_interval, last_consistency_check) {
-                (None, _) => true,
-                (Some(_), None) => false,
-                (Some(interval), Some(last_check)) => last_check.elapsed() < *interval,
-            },
-            Self::TimeTravel { dataset, version } => dataset.version().version == *version,
-        }
-    }
-
    fn time_travel_version(&self) -> Option<u64> {
        match self {
            Self::Latest { .. } => None,
@@ -211,7 +205,18 @@ impl DatasetConsistencyWrapper {
    }

    pub async fn reload(&self) -> Result<()> {
-        self.0.write().await.reload().await
+        if !self.0.read().await.need_reload().await? {
+            return Ok(());
+        }
+
+        let mut write_guard = self.0.write().await;
+        // on lock escalation -- check if someone else has already reloaded
+        if !write_guard.need_reload().await? {
+            return Ok(());
+        }
+
+        // actually need reloading
+        write_guard.reload().await
    }

    /// Returns the version, if in time travel mode, or None otherwise
@@ -230,20 +235,35 @@ impl DatasetConsistencyWrapper {
        }
    }

-    async fn is_up_to_date(&self) -> bool {
-        self.0.read().await.is_up_to_date()
+    async fn is_up_to_date(&self) -> Result<bool> {
+        let dataset_ref = self.0.read().await;
+        match &*dataset_ref {
+            DatasetRef::Latest {
+                read_consistency_interval,
+                last_consistency_check,
+                ..
+            } => match (read_consistency_interval, last_consistency_check) {
+                (None, _) => Ok(true),
+                (Some(_), None) => Ok(false),
+                (Some(read_consistency_interval), Some(last_consistency_check)) => {
+                    if &last_consistency_check.elapsed() < read_consistency_interval {
+                        Ok(true)
+                    } else {
+                        Ok(false)
+                    }
+                }
+            },
+            DatasetRef::TimeTravel { dataset, version } => {
+                Ok(dataset.version().version == *version)
+            }
+        }
    }

    /// Ensures that the dataset is loaded and up-to-date with consistency and
    /// version parameters.
    async fn ensure_up_to_date(&self) -> Result<()> {
-        if !self.is_up_to_date().await {
-            // Re-check under write lock — another task may have reloaded
-            // while we waited for the lock.
-            let mut write_guard = self.0.write().await;
-            if !write_guard.is_up_to_date() {
-                write_guard.reload().await?;
-            }
+        if !self.is_up_to_date().await? {
+            self.reload().await?;
        }
        Ok(())
    }
@@ -331,60 +351,4 @@ mod tests {
        let stats = io_stats.incremental_stats();
        assert_eq!(stats.read_iops, 1);
    }
-
-    /// Regression test: before the fix, the reload fast-path (no version change)
-    /// did not reset `last_consistency_check`, causing a list call on every
-    /// subsequent query once the interval expired.
-    #[tokio::test]
-    async fn test_reload_resets_consistency_timer() {
-        let db = connect("memory://")
-            .read_consistency_interval(Duration::from_secs(1))
-            .execute()
-            .await
-            .unwrap();
-        let io_stats = IoStatsHolder::default();
-        let schema = Arc::new(Schema::new(vec![Field::new("id", DataType::Int32, false)]));
-        let table = db
-            .create_empty_table("test", schema)
-            .write_options(WriteOptions {
-                lance_write_params: Some(WriteParams {
-                    store_params: Some(ObjectStoreParams {
-                        object_store_wrapper: Some(Arc::new(io_stats.clone())),
-                        ..Default::default()
-                    }),
-                    ..Default::default()
-                }),
-            })
-            .execute()
-            .await
-            .unwrap();
-
-        let start = Instant::now();
-        io_stats.incremental_stats(); // reset
-
-        // Step 1: within interval — no list
-        table.schema().await.unwrap();
-        let s = io_stats.incremental_stats();
-        assert_eq!(s.read_iops, 0, "step 1, elapsed={:?}", start.elapsed());
-
-        // Step 2: still within interval — no list
-        table.schema().await.unwrap();
-        let s = io_stats.incremental_stats();
-        assert_eq!(s.read_iops, 0, "step 2, elapsed={:?}", start.elapsed());
-
-        // Step 3: sleep past the 1s boundary
-        tokio::time::sleep(Duration::from_secs(1)).await;
-
-        // Step 4: interval expired — exactly 1 list, timer resets
-        table.schema().await.unwrap();
-        let s = io_stats.incremental_stats();
-        assert_eq!(s.read_iops, 1, "step 4, elapsed={:?}", start.elapsed());
-
-        // Step 5: 10 more calls — timer just reset, no lists (THIS is the regression test).
-        for _ in 0..10 {
-            table.schema().await.unwrap();
-        }
-        let s = io_stats.incremental_stats();
-        assert_eq!(s.read_iops, 0, "step 5, elapsed={:?}", start.elapsed());
-    }
 }
--- a/rust/lancedb/src/table/delete.rs
+++ b/rust/lancedb/src/table/delete.rs
@@ -34,7 +34,7 @@ pub(crate) async fn execute_delete(table: &NativeTable, predicate: &str) -> Resu
 #[cfg(test)]
 mod tests {
    use crate::connect;
-    use arrow_array::{record_batch, Int32Array, RecordBatch};
+    use arrow_array::{record_batch, Int32Array, RecordBatch, RecordBatchIterator};
    use arrow_schema::{DataType, Field, Schema};
    use std::sync::Arc;

@@ -53,7 +53,10 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_delete", batch)
+            .create_table(
+                "test_delete",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -99,7 +102,10 @@ mod tests {
        let original_schema = batch.schema();

        let table = conn
-            .create_table("test_delete_all", batch)
+            .create_table(
+                "test_delete_all",
+                RecordBatchIterator::new(vec![Ok(batch)], original_schema.clone()),
+            )
            .execute()
            .await
            .unwrap();
@@ -120,8 +126,13 @@ mod tests {
        // Create a table with 5 rows
        let batch = record_batch!(("id", Int32, [1, 2, 3, 4, 5])).unwrap();

+        let schema = batch.schema();
+
        let table = conn
-            .create_table("test_delete_noop", batch)
+            .create_table(
+                "test_delete_noop",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
--- a/rust/lancedb/src/table/optimize.rs
+++ b/rust/lancedb/src/table/optimize.rs
@@ -212,7 +212,7 @@ pub(crate) async fn execute_optimize(

 #[cfg(test)]
 mod tests {
-    use arrow_array::{Int32Array, RecordBatch, StringArray};
+    use arrow_array::{Int32Array, RecordBatch, RecordBatchIterator, StringArray};
    use arrow_schema::{DataType, Field, Schema};
    use rstest::rstest;
    use std::sync::Arc;
@@ -236,7 +236,10 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_compact", batch)
+            .create_table(
+                "test_compact",
+                RecordBatchIterator::new(vec![Ok(batch)], schema.clone()),
+            )
            .execute()
            .await
            .unwrap();
@@ -250,7 +253,11 @@ mod tests {
                ))],
            )
            .unwrap();
-            table.add(batch).execute().await.unwrap();
+            table
+                .add(RecordBatchIterator::new(vec![Ok(batch)], schema.clone()))
+                .execute()
+                .await
+                .unwrap();
        }

        // Verify we have multiple fragments before compaction
@@ -315,7 +322,10 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_prune", batch)
+            .create_table(
+                "test_prune",
+                RecordBatchIterator::new(vec![Ok(batch)], schema.clone()),
+            )
            .execute()
            .await
            .unwrap();
@@ -329,7 +339,11 @@ mod tests {
                ))],
            )
            .unwrap();
-            table.add(batch).execute().await.unwrap();
+            table
+                .add(RecordBatchIterator::new(vec![Ok(batch)], schema.clone()))
+                .execute()
+                .await
+                .unwrap();
        }

        // Verify multiple versions exist
@@ -391,7 +405,10 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_index_optimize", batch)
+            .create_table(
+                "test_index_optimize",
+                RecordBatchIterator::new(vec![Ok(batch)], schema.clone()),
+            )
            .execute()
            .await
            .unwrap();
@@ -409,7 +426,11 @@ mod tests {
            vec![Arc::new(Int32Array::from_iter_values(100..200))],
        )
        .unwrap();
-        table.add(batch).execute().await.unwrap();
+        table
+            .add(RecordBatchIterator::new(vec![Ok(batch)], schema.clone()))
+            .execute()
+            .await
+            .unwrap();

        // Verify index stats before optimization
        let indices = table.list_indices().await.unwrap();
@@ -453,7 +474,10 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_optimize_all", batch)
+            .create_table(
+                "test_optimize_all",
+                RecordBatchIterator::new(vec![Ok(batch)], schema.clone()),
+            )
            .execute()
            .await
            .unwrap();
@@ -467,7 +491,11 @@ mod tests {
                ))],
            )
            .unwrap();
-            table.add(batch).execute().await.unwrap();
+            table
+                .add(RecordBatchIterator::new(vec![Ok(batch)], schema.clone()))
+                .execute()
+                .await
+                .unwrap();
        }

        // Run all optimizations
@@ -531,13 +559,20 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_deferred_remap", batch.clone())
+            .create_table(
+                "test_deferred_remap",
+                RecordBatchIterator::new(vec![Ok(batch.clone())], schema.clone()),
+            )
            .execute()
            .await
            .unwrap();

        // Add more data
-        table.add(batch).execute().await.unwrap();
+        table
+            .add(RecordBatchIterator::new(vec![Ok(batch)], schema.clone()))
+            .execute()
+            .await
+            .unwrap();

        // Create an index
        table
@@ -613,13 +648,20 @@ mod tests {
        let original_schema = batch.schema();

        let table = conn
-            .create_table("test_schema_preserved", batch.clone())
+            .create_table(
+                "test_schema_preserved",
+                RecordBatchIterator::new(vec![Ok(batch.clone())], schema.clone()),
+            )
            .execute()
            .await
            .unwrap();

        // Add more data
-        table.add(batch).execute().await.unwrap();
+        table
+            .add(RecordBatchIterator::new(vec![Ok(batch)], schema.clone()))
+            .execute()
+            .await
+            .unwrap();

        // Run compaction
        table
@@ -661,7 +703,10 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_empty_optimize", batch)
+            .create_table(
+                "test_empty_optimize",
+                RecordBatchIterator::new(vec![Ok(batch)], schema.clone()),
+            )
            .execute()
            .await
            .unwrap();
@@ -707,12 +752,19 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_checkout_optimize", batch.clone())
+            .create_table(
+                "test_checkout_optimize",
+                RecordBatchIterator::new(vec![Ok(batch.clone())], schema.clone()),
+            )
            .execute()
            .await
            .unwrap();

-        table.add(batch).execute().await.unwrap();
+        table
+            .add(RecordBatchIterator::new(vec![Ok(batch)], schema.clone()))
+            .execute()
+            .await
+            .unwrap();

        table.checkout(1).await.unwrap();

--- a/rust/lancedb/src/table/schema_evolution.rs
+++ b/rust/lancedb/src/table/schema_evolution.rs
@@ -89,7 +89,7 @@ pub(crate) async fn execute_drop_columns(

 #[cfg(test)]
 mod tests {
-    use arrow_array::{record_batch, Int32Array, StringArray};
+    use arrow_array::{record_batch, Int32Array, RecordBatchIterator, StringArray};
    use arrow_schema::DataType;
    use futures::TryStreamExt;
    use lance::dataset::ColumnAlteration;
@@ -105,9 +105,13 @@ mod tests {
        let conn = connect("memory://").execute().await.unwrap();

        let batch = record_batch!(("id", Int32, [1, 2, 3, 4, 5])).unwrap();
+        let schema = batch.schema();

        let table = conn
-            .create_table("test_add_columns", batch)
+            .create_table(
+                "test_add_columns",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -165,9 +169,13 @@ mod tests {
        let conn = connect("memory://").execute().await.unwrap();

        let batch = record_batch!(("x", Int32, [10, 20, 30])).unwrap();
+        let schema = batch.schema();

        let table = conn
-            .create_table("test_add_multi_columns", batch)
+            .create_table(
+                "test_add_multi_columns",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -197,9 +205,13 @@ mod tests {
        let conn = connect("memory://").execute().await.unwrap();

        let batch = record_batch!(("id", Int32, [1, 2, 3])).unwrap();
+        let schema = batch.schema();

        let table = conn
-            .create_table("test_add_const_column", batch)
+            .create_table(
+                "test_add_const_column",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -243,9 +255,13 @@ mod tests {
        let conn = connect("memory://").execute().await.unwrap();

        let batch = record_batch!(("old_name", Int32, [1, 2, 3])).unwrap();
+        let schema = batch.schema();

        let table = conn
-            .create_table("test_alter_rename", batch)
+            .create_table(
+                "test_alter_rename",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -288,7 +304,10 @@ mod tests {
        .unwrap();

        let table = conn
-            .create_table("test_alter_nullable", batch)
+            .create_table(
+                "test_alter_nullable",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -313,9 +332,13 @@ mod tests {
        let conn = connect("memory://").execute().await.unwrap();

        let batch = record_batch!(("num", Int32, [1, 2, 3])).unwrap();
+        let schema = batch.schema();

        let table = conn
-            .create_table("test_cast_type", batch)
+            .create_table(
+                "test_cast_type",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -356,9 +379,13 @@ mod tests {
        let conn = connect("memory://").execute().await.unwrap();

        let batch = record_batch!(("num", Int32, [1, 2, 3])).unwrap();
+        let schema = batch.schema();

        let table = conn
-            .create_table("test_invalid_cast", batch)
+            .create_table(
+                "test_invalid_cast",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -380,9 +407,13 @@ mod tests {
        let conn = connect("memory://").execute().await.unwrap();

        let batch = record_batch!(("a", Int32, [1, 2, 3]), ("b", Int32, [4, 5, 6])).unwrap();
+        let schema = batch.schema();

        let table = conn
-            .create_table("test_alter_multi", batch)
+            .create_table(
+                "test_alter_multi",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -410,9 +441,13 @@ mod tests {

        let batch =
            record_batch!(("keep", Int32, [1, 2, 3]), ("remove", Int32, [4, 5, 6])).unwrap();
+        let schema = batch.schema();

        let table = conn
-            .create_table("test_drop_single", batch)
+            .create_table(
+                "test_drop_single",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -443,9 +478,13 @@ mod tests {
            ("d", Int32, [7, 8])
        )
        .unwrap();
+        let schema = batch.schema();

        let table = conn
-            .create_table("test_drop_multi", batch)
+            .create_table(
+                "test_drop_multi",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -472,9 +511,13 @@ mod tests {
            ("extra", Int32, [10, 20, 30])
        )
        .unwrap();
+        let schema = batch.schema();

        let table = conn
-            .create_table("test_drop_preserves", batch)
+            .create_table(
+                "test_drop_preserves",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -524,9 +567,13 @@ mod tests {
        let conn = connect("memory://").execute().await.unwrap();

        let batch = record_batch!(("existing", Int32, [1, 2, 3])).unwrap();
+        let schema = batch.schema();

        let table = conn
-            .create_table("test_drop_nonexistent", batch)
+            .create_table(
+                "test_drop_nonexistent",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -546,9 +593,13 @@ mod tests {
        let conn = connect("memory://").execute().await.unwrap();

        let batch = record_batch!(("existing", Int32, [1, 2, 3])).unwrap();
+        let schema = batch.schema();

        let table = conn
-            .create_table("test_alter_nonexistent", batch)
+            .create_table(
+                "test_alter_nonexistent",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
@@ -572,8 +623,13 @@ mod tests {
        let conn = connect("memory://").execute().await.unwrap();

        let batch = record_batch!(("a", Int32, [1, 2, 3]), ("b", Int32, [4, 5, 6])).unwrap();
+        let schema = batch.schema();
+
        let table = conn
-            .create_table("test_version_increment", batch)
+            .create_table(
+                "test_version_increment",
+                RecordBatchIterator::new(vec![Ok(batch)], schema),
+            )
            .execute()
            .await
            .unwrap();
--- a/rust/lancedb/src/table/update.rs
+++ b/rust/lancedb/src/table/update.rs
@@ -117,8 +117,9 @@ mod tests {
    use crate::query::{ExecutableQuery, Select};
    use arrow_array::{
        record_batch, Array, BooleanArray, Date32Array, FixedSizeListArray, Float32Array,
-        Float64Array, Int32Array, Int64Array, LargeStringArray, RecordBatch, StringArray,
-        TimestampMillisecondArray, TimestampNanosecondArray, UInt32Array,
+        Float64Array, Int32Array, Int64Array, LargeStringArray, RecordBatch, RecordBatchIterator,
+        RecordBatchReader, StringArray, TimestampMillisecondArray, TimestampNanosecondArray,
+        UInt32Array,
    };
    use arrow_data::ArrayDataBuilder;
    use arrow_schema::{ArrowError, DataType, Field, Schema, TimeUnit};
@@ -166,46 +167,51 @@ mod tests {
            ),
        ]));

-        let batch = RecordBatch::try_new(
+        let record_batch_iter = RecordBatchIterator::new(
+            vec![RecordBatch::try_new(
+                schema.clone(),
+                vec![
+                    Arc::new(Int32Array::from_iter_values(0..10)),
+                    Arc::new(Int64Array::from_iter_values(0..10)),
+                    Arc::new(UInt32Array::from_iter_values(0..10)),
+                    Arc::new(StringArray::from_iter_values(vec![
+                        "a", "b", "c", "d", "e", "f", "g", "h", "i", "j",
+                    ])),
+                    Arc::new(LargeStringArray::from_iter_values(vec![
+                        "a", "b", "c", "d", "e", "f", "g", "h", "i", "j",
+                    ])),
+                    Arc::new(Float32Array::from_iter_values((0..10).map(|i| i as f32))),
+                    Arc::new(Float64Array::from_iter_values((0..10).map(|i| i as f64))),
+                    Arc::new(Into::<BooleanArray>::into(vec![
+                        true, false, true, false, true, false, true, false, true, false,
+                    ])),
+                    Arc::new(Date32Array::from_iter_values(0..10)),
+                    Arc::new(TimestampNanosecondArray::from_iter_values(0..10)),
+                    Arc::new(TimestampMillisecondArray::from_iter_values(0..10)),
+                    Arc::new(
+                        create_fixed_size_list(
+                            Float32Array::from_iter_values((0..20).map(|i| i as f32)),
+                            2,
+                        )
+                        .unwrap(),
+                    ),
+                    Arc::new(
+                        create_fixed_size_list(
+                            Float64Array::from_iter_values((0..20).map(|i| i as f64)),
+                            2,
+                        )
+                        .unwrap(),
+                    ),
+                ],
+            )
+            .unwrap()]
+            .into_iter()
+            .map(Ok),
            schema.clone(),
-            vec![
-                Arc::new(Int32Array::from_iter_values(0..10)),
-                Arc::new(Int64Array::from_iter_values(0..10)),
-                Arc::new(UInt32Array::from_iter_values(0..10)),
-                Arc::new(StringArray::from_iter_values(vec![
-                    "a", "b", "c", "d", "e", "f", "g", "h", "i", "j",
-                ])),
-                Arc::new(LargeStringArray::from_iter_values(vec![
-                    "a", "b", "c", "d", "e", "f", "g", "h", "i", "j",
-                ])),
-                Arc::new(Float32Array::from_iter_values((0..10).map(|i| i as f32))),
-                Arc::new(Float64Array::from_iter_values((0..10).map(|i| i as f64))),
-                Arc::new(Into::<BooleanArray>::into(vec![
-                    true, false, true, false, true, false, true, false, true, false,
-                ])),
-                Arc::new(Date32Array::from_iter_values(0..10)),
-                Arc::new(TimestampNanosecondArray::from_iter_values(0..10)),
-                Arc::new(TimestampMillisecondArray::from_iter_values(0..10)),
-                Arc::new(
-                    create_fixed_size_list(
-                        Float32Array::from_iter_values((0..20).map(|i| i as f32)),
-                        2,
-                    )
-                    .unwrap(),
-                ),
-                Arc::new(
-                    create_fixed_size_list(
-                        Float64Array::from_iter_values((0..20).map(|i| i as f64)),
-                        2,
-                    )
-                    .unwrap(),
-                ),
-            ],
-        )
-        .unwrap();
+        );

        let table = conn
-            .create_table("my_table", batch)
+            .create_table("my_table", record_batch_iter)
            .execute()
            .await
            .unwrap();
@@ -332,13 +338,15 @@ mod tests {
        Ok(FixedSizeListArray::from(data))
    }

-    fn make_test_batch() -> RecordBatch {
+    fn make_test_batches() -> impl RecordBatchReader + Send + Sync + 'static {
        let schema = Arc::new(Schema::new(vec![Field::new("i", DataType::Int32, false)]));
-        RecordBatch::try_new(
-            schema.clone(),
-            vec![Arc::new(Int32Array::from_iter_values(0..10))],
+        RecordBatchIterator::new(
+            vec![RecordBatch::try_new(
+                schema.clone(),
+                vec![Arc::new(Int32Array::from_iter_values(0..10))],
+            )],
+            schema,
        )
-        .unwrap()
    }

    #[tokio::test]
@@ -359,8 +367,12 @@ mod tests {
        )
        .unwrap();

+        let schema = batch.schema();
+        // need the iterator for create table
+        let record_batch_iter = RecordBatchIterator::new(vec![Ok(batch)], schema);
+
        let table = conn
-            .create_table("my_table", batch)
+            .create_table("my_table", record_batch_iter)
            .execute()
            .await
            .unwrap();
@@ -418,7 +430,7 @@ mod tests {
            .await
            .unwrap();
        let tbl = conn
-            .create_table("my_table", make_test_batch())
+            .create_table("my_table", make_test_batches())
            .execute()
            .await
            .unwrap();
--- a/rust/lancedb/src/test_utils.rs
+++ b/rust/lancedb/src/test_utils.rs
@@ -3,4 +3,3 @@

 pub mod connection;
 pub mod datagen;
-pub mod embeddings;
--- a/rust/lancedb/src/test_utils/datagen.rs
+++ b/rust/lancedb/src/test_utils/datagen.rs
@@ -34,7 +34,10 @@ impl LanceDbDatagenExt for BatchGeneratorBuilder {
            schema,
        ));
        let db = connect("memory:///").execute().await.unwrap();
-        db.create_table(table_name, stream).execute().await.unwrap()
+        db.create_table_streaming(table_name, stream)
+            .execute()
+            .await
+            .unwrap()
    }
 }

@@ -45,5 +48,8 @@ pub async fn virtual_table(name: &str, values: &RecordBatch) -> Table {
        schema,
    ));
    let db = connect("memory:///").execute().await.unwrap();
-    db.create_table(name, stream).execute().await.unwrap()
+    db.create_table_streaming(name, stream)
+        .execute()
+        .await
+        .unwrap()
 }
--- a/rust/lancedb/src/test_utils/embeddings.rs
+++ b/rust/lancedb/src/test_utils/embeddings.rs
@@ -1,59 +0,0 @@
-// SPDX-License-Identifier: Apache-2.0
-// SPDX-FileCopyrightText: Copyright The LanceDB Authors
-
-use std::{borrow::Cow, sync::Arc};
-
-use arrow_array::{Array, FixedSizeListArray, Float32Array};
-use arrow_schema::{DataType, Field};
-
-use crate::embeddings::EmbeddingFunction;
-use crate::Result;
-
-#[derive(Debug, Clone)]
-pub struct MockEmbed {
-    name: String,
-    dim: usize,
-}
-
-impl MockEmbed {
-    pub fn new(name: impl Into<String>, dim: usize) -> Self {
-        Self {
-            name: name.into(),
-            dim,
-        }
-    }
-}
-
-impl EmbeddingFunction for MockEmbed {
-    fn name(&self) -> &str {
-        &self.name
-    }
-
-    fn source_type(&self) -> Result<Cow<'_, DataType>> {
-        Ok(Cow::Borrowed(&DataType::Utf8))
-    }
-
-    fn dest_type(&self) -> Result<Cow<'_, DataType>> {
-        Ok(Cow::Owned(DataType::new_fixed_size_list(
-            DataType::Float32,
-            self.dim as _,
-            true,
-        )))
-    }
-
-    fn compute_source_embeddings(&self, source: Arc<dyn Array>) -> Result<Arc<dyn Array>> {
-        // We can't use the FixedSizeListBuilder here because it always adds a null bitmap
-        // and we want to explicitly work with non-nullable arrays.
-        let len = source.len();
-        let inner = Arc::new(Float32Array::from(vec![Some(1.0); len * self.dim]));
-        let field = Field::new("item", inner.data_type().clone(), false);
-        let arr = FixedSizeListArray::new(Arc::new(field), self.dim as _, inner, None);
-
-        Ok(Arc::new(arr))
-    }
-
-    #[allow(unused_variables)]
-    fn compute_query_embeddings(&self, input: Arc<dyn Array>) -> Result<Arc<dyn Array>> {
-        todo!()
-    }
-}
--- a/rust/lancedb/tests/embedding_registry_test.rs
+++ b/rust/lancedb/tests/embedding_registry_test.rs
@@ -15,6 +15,7 @@ use arrow_array::{
 use arrow_schema::{DataType, Field, Schema};
 use futures::StreamExt;
 use lancedb::{
+    arrow::IntoArrow,
    connect,
    embeddings::{EmbeddingDefinition, EmbeddingFunction, EmbeddingRegistry},
    query::ExecutableQuery,
@@ -252,7 +253,7 @@ async fn test_no_func_in_registry_on_add() -> Result<()> {
    Ok(())
 }

-fn create_some_records() -> Result<Box<dyn arrow_array::RecordBatchReader + Send>> {
+fn create_some_records() -> Result<impl IntoArrow> {
    const TOTAL: usize = 2;

    let schema = Arc::new(Schema::new(vec![
--- a/rust/lancedb/tests/object_store_test.rs
+++ b/rust/lancedb/tests/object_store_test.rs
@@ -4,7 +4,7 @@
 #![cfg(feature = "s3-test")]
 use std::sync::Arc;

-use arrow_array::{Int32Array, RecordBatch, StringArray};
+use arrow_array::{Int32Array, RecordBatch, RecordBatchIterator, StringArray};
 use arrow_schema::{DataType, Field, Schema};

 use aws_config::{BehaviorVersion, ConfigLoader, Region, SdkConfig};
@@ -111,6 +111,7 @@ async fn test_minio_lifecycle() -> Result<()> {
        .await?;

    let data = test_data();
+    let data = RecordBatchIterator::new(vec![Ok(data.clone())], data.schema());

    let table = db.create_table("test_table", data).execute().await?;

@@ -126,6 +127,7 @@ async fn test_minio_lifecycle() -> Result<()> {
    assert_eq!(row_count, 3);

    let data = test_data();
+    let data = RecordBatchIterator::new(vec![Ok(data.clone())], data.schema());
    table.add(data).execute().await?;

    db.drop_table("test_table", &[]).await?;
@@ -245,6 +247,7 @@ async fn test_encryption() -> Result<()> {

    // Create a table with encryption
    let data = test_data();
+    let data = RecordBatchIterator::new(vec![Ok(data.clone())], data.schema());

    let mut builder = db.create_table("test_table", data);
    for (key, value) in CONFIG {
@@ -271,6 +274,7 @@ async fn test_encryption() -> Result<()> {
    let table = db.open_table("test_table").execute().await?;

    let data = test_data();
+    let data = RecordBatchIterator::new(vec![Ok(data.clone())], data.schema());
    table.add(data).execute().await?;
    validate_objects_encrypted(&bucket.0, "test_table", &key.0).await;

@@ -296,6 +300,7 @@ async fn test_table_storage_options_override() -> Result<()> {

    // Create table overriding with key2 encryption
    let data = test_data();
+    let data = RecordBatchIterator::new(vec![Ok(data.clone())], data.schema());
    let _table = db
        .create_table("test_override", data)
        .storage_option("aws_sse_kms_key_id", &key2.0)
@@ -307,6 +312,7 @@ async fn test_table_storage_options_override() -> Result<()> {

    // Also test that a table created without override uses connection settings
    let data = test_data();
+    let data = RecordBatchIterator::new(vec![Ok(data.clone())], data.schema());
    let _table2 = db.create_table("test_inherit", data).execute().await?;

    // Verify this table uses key1 from connection
@@ -413,6 +419,7 @@ async fn test_concurrent_dynamodb_commit() {
        .unwrap();

    let data = test_data();
+    let data = RecordBatchIterator::new(vec![Ok(data.clone())], data.schema());

    let table = db.create_table("test_table", data).execute().await.unwrap();

@@ -423,6 +430,7 @@ async fn test_concurrent_dynamodb_commit() {
        let table = db.open_table("test_table").execute().await.unwrap();
        let data = data.clone();
        tasks.push(tokio::spawn(async move {
+            let data = RecordBatchIterator::new(vec![Ok(data.clone())], data.schema());
            table.add(data).execute().await.unwrap();
        }));
    }