Orca
A desktop ADE for running and reviewing parallel CLI coding agents in isolated Git worktrees.
Per-dimension scores and reasoning
Evidence shows: README mentions privacy and telemetry docs but lacks specifics; package.json includes posthog-node telemetry dependency without data flow details; workflow files show careful handling of GitHub tokens and signing credentials but not user data. Deductions: no user confirmation mechanism, insufficient data flow transparency, missing sensitive data handling details, no dependency security audit shown, external effects (e.g., SSH, computer use) lack explicit permission boundaries, rollback not mentioned, source attribution partial (publisher unverified).
Evidence shows: README and package.json are consistent, version number present, extensive test scripts, workflow files include detailed failure handling. Deductions: no dependency availability guarantees (e.g., lockfile, mirrors), failure messages only in CI workflows, not at user interface level.
Evidence shows: README clearly targets '100x builders' and multiple scenarios (parallel worktrees, mobile companion, SSH), supports many CLI agents, provides multi-platform install instructions. Deductions: capability boundaries not explicit (e.g., computer use permissions), trigger precision (e.g., CLI commands) not detailed, environment fit (e.g., headless Linux) only partially mentioned.
Evidence shows: README well-structured, install notes detailed, naming stable (orca), examples and FAQ via docs links, MIT license clear, version number present, maintenance responsibility via contributing guide and community channels. Deductions: known limitations not explicitly listed, changelog not directly provided (only via releases link).
Evidence shows: output usability (e.g., CLI commands, mobile app) described, marginal value (parallel agent orchestration) clear, cost-benefit (use own subscription) mentioned. Deductions: cost-benefit not quantified, lack of specific performance data.
Evidence shows: feature claims in README have doc links but no independent verification; test files exist but no test results shown; facts and inferences not clearly separated. Deductions: claim traceability insufficient, cross-source corroboration lacking, fact-inference separation unclear.
- Publisher identity unverified; treat source with caution.
- Telemetry data flow opaque; review privacy policy.
- Dependency security not audited; check for known vulnerabilities.
- External effects (e.g., SSH, computer use) lack explicit permission boundaries.
What does this agent do, and when should you use it?
Orca is an agent development environment for coordinating parallel coding agents, with each agent assigned its own Git worktree. It explicitly supports Codex, Claude Code, OpenCode, Pi, and other CLI agents that run in a terminal. The desktop application combines terminal splits, file editing, an embedded Chromium-based Design Mode, GitHub and Linear workflows, and line-level annotations on AI-generated diffs. It is distributed for macOS, Windows, and Linux, with iOS and Android companion apps for status monitoring, notifications, and follow-up messages. For remote development, SSH worktrees run agents on a remote machine with file editing, Git, terminals, automatic reconnection, and port forwarding.
A user sends prompts to one or more CLI coding agents in Orca; the same prompt can be fanned out across agents, each running in an isolated Git worktree. Agents work through terminals, and their output is surfaced as workspace file changes, Git diffs, and changes that can be committed; users can inspect and edit work, annotate individual diff lines, and send that feedback back to an agent. Orca documents CLI workflows including orca worktree create, snapshot, click, and fill. Its Design Mode lets a user select an element in a Chromium window and send that element's HTML, CSS, and a cropped screenshot into an agent prompt. SSH worktrees move the agent, file editing, Git, and terminal activity onto a remote machine.
- A developer evaluating several fixes for the same bug can fan one prompt out to multiple agents, compare their isolated worktrees, and merge the preferred result.
- An engineer who already uses Codex, Claude Code, OpenCode, or Pi can run those terminal agents side by side in one desktop workspace.
- A team with a powerful remote Linux machine can use SSH worktrees for agents while retaining file editing, Git, terminal access, reconnection, and port forwarding.
- A reviewer assessing AI-produced changes can attach comments to precise diff lines and send the requested revisions back to the agent.
- A developer away from their desk can use the iOS or Android companion app to receive completion notifications and send follow-up instructions.
What are this agent's strengths and limitations?
- Runs parallel agents in separate Git worktrees, allowing alternatives to be compared before a result is merged.
- Supports multiple terminal-based coding agents instead of one named provider; the README explicitly includes Codex, Claude Code, OpenCode, and Pi.
- Brings terminal splits, file editing, AI-diff annotations, GitHub/Linear browsing, and Design Mode into one desktop workspace.
- Provides SSH worktrees with documented automatic reconnection and port forwarding.
- The mobile companion can notify users when an agent finishes and accept follow-up instructions.
- Its core workflow depends on users having terminal-runnable coding agents installed and their own subscriptions.
- Parallel isolation is based on Git worktrees, so projects outside a Git-based workflow do not directly receive that model.
- The supplied material lacks complete setup details for first launch, CLI-agent authentication, SSH configuration, and command arguments for the Orca CLI.
- GitHub, Linear, SSH, mobile, and Design Mode workflows depend on the relevant services, network access, and platform conditions.
How do you install or deploy this agent?
Install the documented desktop package for your platform:
macOS: brew install --cask stablyai/orca/orca
Arch Linux: yay -S stably-orca-bin
Desktop builds are explicitly listed for macOS (Apple Silicon and Intel), Windows (.exe), and Linux (AppImage). The mobile companion is available through the iOS App Store/TestFlight or the Android APK listed in the README. You also need a terminal-runnable CLI agent and that agent's own subscription. The supplied evidence does not document complete first-launch, agent-authentication, or mobile-pairing steps.
How do you use this agent?
Open a Git repository in Orca and start a terminal CLI coding agent. To explore alternatives in parallel, send the same prompt to multiple agents so each works in its own Git worktree. Compare the resulting file changes and diffs, add line comments where revisions are needed, send feedback to the agent, then choose a result to merge or commit. Documented CLI workflow names include orca worktree create, orca snapshot, orca click, and orca fill; the supplied material does not include their arguments or a complete runnable workflow example.
How does this agent compare with similar options?
Compared with using Codex, Claude Code, OpenCode, or Pi directly in a single terminal, Orca is positioned as the workspace that runs, compares, and reviews multiple terminal agents in separate Git worktrees; it does not replace those CLI agents themselves.
FAQ
Is Orca tied to one model or agent provider?
Do I need separate subscriptions for coding agents?
Can it run work on a remote server?
What survives a restart or disconnection?
Is the CLI fully documented in the supplied material?
orca worktree create, snapshot, click, and fill. Their argument syntax and a complete end-to-end example are not included.