* refactor(terminal): own multiplex ACK window accounting in one module * perf(terminal): pace alternate-screen multiplex output to the measured link BDP Slow SSH links queued up to 2MB of superseded TUI frames ahead of keystroke echo. orcad now estimates delivery rate and base RTT from existing output ACK timing, caps in-flight output for alternate-screen streams near 2x BDP on links under 4 MB/s, and replaces a backlog larger than that budget with the existing recovery snapshot. Normal-screen streams and fast links keep the legacy windows. ORCA_TERMINAL_ADAPTIVE_ACK_WINDOW=0 restores legacy behavior. * perf(terminal): back off unavailable superseding snapshots and bound legacy echo checks * perf(terminal): gate ACK pacing to measured remote links and keep supersedes history-neutral Pacing now needs a base RTT of at least 40ms, 16 fresh window-limited samples, and a sample within 15s, so LAN/local links and stale estimates keep the legacy windows. Superseding snapshots are screen-only, only replace frames that never switch screen mode (rechecked after serializing), and leave client history untouched. Paced streams keep today's total buffering before overflow, so overflow recovery fires no earlier than with legacy windows. * fix(terminal): close supersede gaps found in review Scan for screen switches and OSC side effects across frame and send boundaries (raw and JSON-escaped), back off after an aborted supersede so it cannot re-serialize in a loop, and keep paced pending headroom until the queue drains so leaving the alternate screen cannot force an overflow recovery. * fix(terminal): judge supersedable backlogs on decoded text across every seam Scan the joined text of the whole queue (plus the sent tail) so screen switches split across any frame, send, or covered/retained boundary are seen; never supersede span frames; reject bells (via the shared bell detector), OSC clipboard/notification writes and 8-bit OSC; back off for 1s after a rejected scan instead of rescanning on every ACK. * perf(terminal): make backlog superseding opt-in and forget the sent tail after snapshots Every recovery snapshot makes current clients replay their reattach reset (cursor shape/visibility, focus reporting), which is visible when it fires routinely, so superseding is opt-in via ORCA_TERMINAL_SUPERSEDE_ALT_BACKLOG=1 until clients can skip it. Pacing stays on by default: on a slow link the existing overflow recovery then reaches the client promptly instead of queuing behind a 2MB window. After any snapshot the sent tail is unknown (it can end inside an escape), so superseding waits for the next text frame. * test(terminal): cover default pacing and opt-in superseding separately * perf(terminal): read the supersede switch once per connection and skip its bookkeeping when off * perf(terminal): make adaptive ACK pacing opt-in and fix review findings - Default off: ORCA_TERMINAL_ADAPTIVE_ACK_WINDOW=1 enables it; when off every new path is a single flag check, so behavior matches legacy windows. - Base RTT is a windowed minimum kept honest by periodic probes that drain our own queue; a paced RTT beyond what our window can add means the path changed, so legacy windows hold while the base re-learns (no more self-tightening to 64KiB after an RTT increase). - Paced pending headroom is capped by the connection window legacy would have used and released on drain, pause and detach. - Superseding refuses snapshots that end inside an escape. - Replace the latency-threshold unit test with hardcoded estimator tests and move the link simulation to bench:terminal-multiplex-ack-pacing. * perf(terminal): drop backlog superseding and harden opt-in ACK pacing - Remove the opt-in backlog-superseding snapshot path (code, env switch, frame classifier and tests); it needs a client capability to skip the reattach reset before it can ship. - Paced-ness is decided once at admission (alt screen checked before the estimate) and reused by the send, so a probe starting in between cannot mislabel a sample. - Base RTT is fed only by sends with a near-empty connection queue; probes end only on such samples from paced streams, and a probe that cannot drain backs off. A path change forgets the base and holds legacy windows until a drained sample re-learns it. - Paced pending headroom is sized from the steady window (never a probe's zero) and released whenever the queue drains back within the legacy cap. - Flow-level tests: disabled pacing keeps legacy frames on a measured slow link; drained headroom returns so later output overflows at the legacy cap. * fix(terminal): release paced headroom whenever the queue drains, including after recovery
Orca
中文 · 日本語 · 한국어 · Español · Français · Português · Türkçe · Русский
The AI Orchestrator for 100x builders.
Run Claude Code, Codex, and any other coding agent in parallel, each in its own worktree. Chat or terminal, on your laptop or in the cloud, with diffs, a browser, and a CLI in one app.
Download Orca
Features
Also in the box:
- Drag Files to Agents: drag files from the explorer, or images from your desktop, straight into an agent's prompt.
- Jump Anywhere: one shortcut jumps to any workspace or tab across every repo and host.
- Rich Previews: open images, PDFs, and Markdown right in the workspace, next to the agent that made them.
- Mobile Emulator: run an iOS Simulator or Android emulator beside your agent, and let it tap, type, and check its own work.
- Dictation: press a shortcut and talk. Orca types what you say into the agent's prompt or terminal.
- Usage & Accounts: see how much of your agent limits you've used and when they reset, and switch accounts without logging out.
- Automations: schedule prompts to run on any repo: nightly dependency bumps, weekly triage, flaky-test sweeps.
- Computer Use: when there's no CLI or API, agents can see and drive desktop apps and external browsers.
- And many, many more: we ship daily, so this list is perpetually behind. The changelog is the real feature list.
Supported Agents
Works with any CLI agent — if it runs in a terminal, it runs in Orca.
Claude Code
Codex
Grok
Cursor
GitHub Copilot
Muse
DeepSeek Harness
ZCode
OpenCode
MiMo Code
Amp
OpenClaude
Antigravity
Pi
oh-my-pi
Hermes Agent
Devin
Goose
Auggie
Autohand Code
Charm
Cline
CodeBuddy
Codebuff
Freebuff
Command Code
Continue
Droid
Kilocode
Kimi
Kiro
Mistral Vibe
Qwen Code
Rovo Dev
+ any CLI agent
Install
Desktop — macOS, Windows, Linux
- Download from onOrca.dev
- Or grab a build directly: macOS Apple Silicon · macOS Intel · Windows (.exe) · Linux AppImage · All builds
Or via a package manager:
# macOS (Homebrew)
# Always use the full name; bare `orca` is an unrelated Plotly cask.
brew install --cask stablyai/orca/orca
# Upgrade an existing install
brew upgrade --cask stablyai/orca/orca
# Arch Linux (AUR) — or stably-orca-git to build from source
yay -S stably-orca-bin
Mobile App — iOS, Android
Pair with your desktop app to monitor and steer your agents from your phone.
- iOS: Download on the App Store
- Android: Download APK · Install guide
Community & Support
-
Discord: Join the community on Discord.
-
Twitter / X: Follow @orca_build for updates and announcements.
-
WeChat: Scan to join an Orca community WeChat group. If one group is full, join another.
-
Feedback & Ideas: We ship fast. Missing something? Request a new feature.
-
Privacy: See the privacy & telemetry docs for what anonymous usage data Orca collects and how to opt out.
-
Show Support: Star this repo to follow along with our daily ships.
Developing
Want to contribute or run locally? See our CONTRIBUTING.md guide.
Signed Builds
Windows code signing sponsored/provided by SignPath.io, certificate by SignPath Foundation.
License
Orca is free and open source under the MIT License.













