Brennan Benson
d5ef633adb
* fix(crash-reporting): keep a pre-gone process-metrics sample so the crashed process's working set survives its crash report * feat(crash-reporting): renderer peak/private bytes and gone-time system memory in crash details * fix(crash-reporting): macOS system-memory fields and an era-invariant pin for peak/private metrics * test(crash-reporting): kill five mutation survivors in the pre-gone sampler coverage Adversarial review found these mutations survived the suite: - dropping the immediate sample at startPreGoneProcessMetricsSampling() - removing the double-start idempotence guard - widening renderer peak/private aggregation to all buckets - a failed sweep erasing the previous good sample - the recorder hardcoding 'renderer' instead of event.processType Each now has a binding assertion; also documents that the crashed-process-absent flag is bucket-level only. * fix(crash-reporting): prove crasher absence by vanished pid, not bucket count alone The absent flag was bucket-level, so any surviving same-type process (a webview guest, the dashboard popout, another utility) silently cleared it — and webviewTag guests make multi-renderer sessions the norm. The pre-gone sample now keeps per-process pid/bucket/workingSet identities; a sampled same-bucket pid missing from the live set proves absence and reports the vanished process's own working set (processMetricsVanished*), so the crasher's size is no longer summed with surviving guests. Also: split gone-time system memory into its own module (max-lines), pin peak/private aggregation as a true max, clamp garbage negative working sets and backwards clocks, pin live-metrics precedence over incoming detail keys, and verify the sampler timer is unref'd by behavior. * fix(crash-reporting): bucket-aware vanished-pid check with consume-once attribution Loop-3 hardening of the vanished-pid logic: - Live pids now carry their bucket: a recycled pid living on as a different process type still reads as a vanished sampled process (the bare pid set misread the crasher as alive). - Vanished pids are attributed once. In a crash loop with no sweep between deaths, record #2 confidently inherited the FIRST crasher's pid and working set (dedupe window is only 2s, so both records ship); it now degrades to the honest bucket-count arm instead. - An ambiguous multi-process VanishedWorkingSetMB sum is bounded by VanishedLargestWorkingSetMB so no single-process reading of the sum survives triage. - Killer tests for the remaining mutation survivors: unreadable gone-time metrics prove nothing (flag/vanished stay off), pid-less sampled metrics never vanish, fractional-MB rounding, negative system-memory clamp, and full per-family precedence over colliding incoming detail keys. * fix(crash-reporting): flag consumed and blind-era vanished attribution instead of going silent Loop-4 hardening of the consume-once attribution: - processMetricsVanishedAlreadyReportedCount: a record whose vanished pids were consumed by a prior report now says so, instead of being indistinguishable from "nothing vanished" while its PreGone mirrors still show the prior crasher's era. - processMetricsVanishedAmbiguousWithEarlierCrash: consume-once only consumed when the gone-time read succeeded; a crash recorded blind (getAppMetrics threw) left its pid unconsumed, so the next record in the same era confidently emitted THAT crash's pid and working set as its own. Blind buckets now taint the era until a fresh sweep. - Pin two behaviors that were correct but unpinned: a failed sweep must not clear attribution state, and an ambiguous vanished pair is consumed too. - Document gone-time system memory reading healthier than at kill time, and the bound on the attribution set. - Split the suppressed-breadcrumb builder into its own module (max-lines). * fix(crash-reporting): extend vanished ambiguity to consumed eras and pin two unpinned behaviors Loop-5 findings: - processMetricsVanishedAmbiguousWithEarlierCrash fired only for the blind era; a partially-consumed era has the same shape (an earlier crash's unsampled respawn is as plausible a crasher as the newly vanished pid), yet emitted a confident VanishedPid with no flag. - Pin the > largest tie-break (first-enumerated wins) instead of re-accepting it as an equivalent mutant every loop. - The suppressed-breadcrumb type field had zero coverage after the module split — removing the whole block passed the suite. * refactor(crash-reporting): cut per-pid vanished attribution, keep the stateless absence proof Five review loops found defects in the same subsystem: the per-pid vanished attribution outputs (consume-once set, blind-era taint, consumed-era ambiguity). The absence proof they fed does not need any of it — a sampled same-bucket pid missing from the live enumeration (including cross-bucket pid recycle) is stateless and idempotent, so it stays true for every record of a crash loop with zero module-level attribution state. Dropped: processMetricsVanished{Count,WorkingSetMB,Pid, LargestWorkingSetMB,AlreadyReportedCount,AmbiguousWithEarlierCrash}, attributedVanishedPids, metricsBlindCrashBuckets, and the tests that existed only to defend them. Kept and still pinned: the 60s pre-gone sampler and its lifecycle, PreGone* mirrors + SampleAgeMs, renderer peak/private, gone-time system memory, the recorder processType binding, the live/PreGone era invariant, and the browser-pane case (webview guests keep the renderer bucket alive) that motivated the PR — now asserted via the absence flag plus PreGone mirrors alone. Documented the two honest limits: PreGone values are sample-time (up-to-60s understatement, bounded by AgeMs and lifetime peaks), and same-bucket pid recycle inside the sweep window is a false negative for the absence proof. * test(crash-reporting): pin PreGoneLargest to the crasher's own size, not the bucket's running sum Loop-6 mutation battery found one survivor in the cut's re-anchored suite: mutating Largest to carry the bucket's running sum survived every test, because no fixture put a same-bucket sibling BEFORE the largest process. That is the summing-bug family loop 2 found live. The webview-guest test now enumerates the guest first and asserts PreGoneLargest{Pid,Type,WorkingSetMB} carry the crasher's individual 4380, alongside the 4680 bucket total. Also restores the false-positive caveat the cut's comment dropped: a legitimately closed sampled process can trip the absence flag if the crasher's row somehow survives the live enumeration (pre-existing, unchanged by the cut). * fix(crash-reporting): mark pre-gone attribution ambiguous PreGone mirrors are whole-app snapshots, so a larger surviving Tab can own Largest and renderer-wide peak/private fields. Emit an explicit ambiguity boundary, prove the counterexample, and use Electron's pid plus creationTime identity to catch same-bucket PID reuse without adding stateful attribution.
Orca
中文 · 日本語 · 한국어 · Español · Français · Português
The AI Orchestrator for 100x builders.
Run Codex, ClaudeCode, OpenCode or Pi side-by-side — each in its own worktree, tracked in one place.
Download Orca
Features
Mobile CompanionMonitor and steer your agents from your phone — get notified when an agent finishes and send follow-ups from anywhere. |
 |
Parallel WorktreesFan one prompt across five agents, each in its own isolated git worktree — compare the results and merge the winner. |
 |
Terminal SplitsGhostty-class terminals with WebGL rendering, infinite splits, and scrollback that survives restarts. |
 |
Design ModeClick any UI element in a real Chromium window to send its HTML, CSS, and a cropped screenshot straight into your agent's prompt. |
 |
GitHub & Linear, NativeBrowse PRs, issues, and project boards in-app — open a worktree from any task and review without a context switch. |
 |
SSH WorktreesRun agents on a beefy remote box with full file editing, git, and terminals — auto-reconnect and port forwarding included. |
 |
Annotate AI DiffsDrop comments on any diff line and ship them back to the agent — review, edit, and commit without leaving Orca. |
 |
Drag Files to AgentsVS Code's editor with autosave everywhere — drag files or images straight into an agent prompt. |
 |
Orca CLIAgents drive Orca too — script every workflow with |
 |
Also in the box:
- Quick open — Search across worktrees, files, agents, commands, and repo context without leaving your flow.
- Account switcher & usage tracking — See Claude and Codex usage and rate-limit resets, and hot-swap accounts without re-logging in.
- Rich repo previews — Preview Markdown, images, PDFs, and repo docs in the workspace.
- Computer Use — Let agents operate desktop apps and visible UI when a workflow needs real interaction.
- Notifications and unread state — Know when an agent finishes or needs attention, then mark threads unread to come back later.
- 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
OpenCode
MiMo Code
Amp
OpenClaude
Antigravity
Pi
oh-my-pi
Hermes Agent
Devin
Goose
Auggie
Autohand Code
Charm
Cline
Codebuff
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
- Running
orca serveon a headless Linux server? See the headless Linux server guide.
Or via a package manager:
# macOS (Homebrew)
brew install --cask stablyai/orca/orca
# Arch Linux (AUR) — or stably-orca-git to build from source
yay -S stably-orca-bin
Mobile Companion — iOS, Android
Pair with your desktop app to monitor and steer your agents from your phone.
- iOS: Download on the App Store or join TestFlight
- Android: Download APK 0.0.43
Community & Support
-
Discord: Join the community on Discord.
-
Twitter / X: Follow @orca_build for updates and announcements.
-
WeChat: Scan to join the Orca community WeChat group 7.
-
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 sponored/provided by SignPath.io, certificate by SignPath Foundation.
License
Orca is free and open source under the MIT License.