↓ Skip to main content
9 – 17 UHR +49 8031 3508270 LUITPOLDSTR. 9, 83022 ROSENHEIM
DE / EN
Orchestrator Stably AI (Lovecast Inc.) USA Desktop app, SSH, mobile

Orca

Orca by Stably AI: open-source desktop app that runs coding agents like Claude Code and Codex in parallel in separate Git worktrees, with review and PRs.

Licence MIT
First release March 2026
Interfaces Desktop app (Electron), CLI (orca), SSH (remote hosts), Mobile companion (iOS, Android)
Platforms macOS, Windows, Linux

The seven building blocks in Orca

How Orca implements the building blocks of an AI harness

  1. 1

    Agent loop

    Orca does not run an agent loop of its own; it launches the installed agent CLIs in terminals or, optionally, in a structured chat surface over supported agent sessions.

  2. 2

    Tools and execution environment

    Tools come from the respective coding agent; Orca adds the `orca` CLI, a skills registry with MCP, an embedded browser and, per its docs, computer use for local desktop apps.

  3. 3

    Context management

    Every task runs in its own Git worktree with its own branch; files Git ignores, such as local configuration, can be carried into new worktrees via `.worktreeinclude` or `orca.yaml`.

  4. 4

    State and memory

    After a restart, Orca restores worktrees, splits, scrollback and the focused tab according to its docs; past agent sessions can be browsed and resumed from the sidebar.

  5. 5

    Permissions

    According to its docs, Orca pre-fills the agents' bypass flags for new launches, such as `--dangerously-skip-permissions` for Claude Code and `--dangerously-bypass-approvals-and-sandbox` for Codex; you can switch to "Manual" in the settings.

  6. 6

    Verification and feedback

    A diff review with staging by hunk or line, line comments handed back to the agent in one go, commit, push and pull requests with checks supports human review before the merge.

  7. 7

    Orchestration

    Orca's core contribution: several agents in parallel, one worktree per task; plus an experimental orchestration with runs, tasks and supervised workers, and bots that accept tasks from Slack, Linear or GitHub.

Models and model access

Which models the harness uses and how they are obtained

Models and model access
Models Model-agnostic: Orca starts the agents; the models are configured there
Access routes Model-agnostic (depends on the coding agent started)
Free choice of model Yes
Custom endpoint / LLM gateway No
Orca does not call language models itself for agent work; the coding agent that is started determines model, provider and endpoint. According to the maker, prompts and code go directly to the configured providers.
Features
MCP servers Sub-agents Hooks Skills Plugins / extensions Headless / CI Checkpoints Resume sessions

Enterprise use

Data processing, controls and pricing model according to the vendor

Data processing and location

According to the maker, Orca works locally; prompts and code go directly to the configured model providers and are neither inspected nor stored by Orca. Bot connectors (Slack, GitHub, Linear, webhooks) run through Orca Cloud, and the mobile relay requires a sign-in and, per the privacy policy, uses infrastructure in the US and Hong Kong.

Permissions

By default, Orca launches agents with their bypass flags. A Git worktree only isolates files in the repository, not processes, network access or credentials. For company use, the "Manual" setting should be chosen deliberately.

Telemetry

Product telemetry is on by default for new installations and goes pseudonymized to PostHog in the US; it can be switched off in the settings or via `DO_NOT_TRACK=1`. The docs do not state a maximum retention period.

Identity and team

Orca does not document its own user management for developers; bots have an allow list ("Who can ask"), and a Remote Orca Server can serve as a shared runtime for several clients.

Pricing

Orca is MIT-licensed and free. An enterprise offer is available on request; documented there is "SOC 2 Readiness", not a certification.

Strengths
  • One task, one worktree: parallel agents do not get in each other's way in the repository
  • Complete working environment with terminal, editor, embedded browser with design mode and diff review
  • Pull requests for GitHub, GitLab and Bitbucket Cloud with checks, plus auto-merge and stacked PRs on GitHub; Linear and Jira integrated
  • Uses existing subscriptions and logins, with account switching and usage and rate-limit display
  • More than 40 preconfigured agents, deeper integration with Claude Code, Codex, Cursor and Muse Code
  • Remote operation via SSH, as a Remote Orca Server or with the mobile companion
Limitations
  • The agents' bypass flags are set by default and must be changed deliberately for company use
  • Remote Orca Server is in beta, orchestration and cloud VM per workspace are experimental
  • Bots and mobile relay require Orca Cloud with infrastructure outside the EU
  • Feature-rich Electron app rather than a lean terminal runtime
  • Telemetry on by default for new installations

Use cases

Typical tasks for this harness

Let several coding agents work in parallel on separate tasks, each in its own worktree
Review agent results in the diff, comment on them and ship them as a pull request
Put different agents on the same task and compare the results
Run agents on a remote development machine via SSH and review locally
Pass tasks from Slack, Linear or GitHub to agents via bots

Orca is an open-source desktop app by Stably AI that runs several AI coding agents in parallel and starts every task in its own Git worktree. Orca calls itself an “Agent Development Environment”: besides the agent terminals, the app brings an editor, an embedded browser, a diff review and pull requests. As of 22 November 2026.

What is Orca?

Orca is developed by Stably AI (Lovecast Inc.) from San Francisco, a company from the Y Combinator program. The stablyai/orca repository is MIT-licensed; the first release came out in March 2026, and the current version is v1.4.269 from 21 November 2026, with several releases per week. The app is based on Electron and runs on macOS (Apple Silicon and Intel), Windows 10/11 and Linux (AppImage, .deb, .rpm), complemented by mobile apps for iOS and Android.

Orca as an AI harness

As an AI harness, Orca mainly covers the orchestration building block; model access and tools stay with the agent that is started. The organizing principle is “worktree-first”: every task gets its own worktree and branch, so parallel agents do not overwrite each other in the repository. Orca also supports the main checkout and folder workspaces. Around the agents, Orca builds a working environment of a GPU-rendered terminal with splits, an editor, a Chromium browser with design mode and a diff review. Line comments in the diff go back to the agent in one go, and commit, push and pull requests run from within the app.

For other machines, Orca offers four modes according to its docs: local, via SSH (agents and Git run on the host, editor and review stay local), as a Remote Orca Server in beta, and experimentally as a cloud VM per workspace with your own provider. Through the orca CLI, agents can control Orca themselves; an orchestration with runs, tasks and supervised workers is experimental.

Models and model access

Orca is model-agnostic. It launches the installed agent CLIs, which determine model, provider and access route themselves. Orca is explicitly built for subscriptions: Claude Code uses the existing login, several Claude and Codex accounts can be stored and switched, and the app shows usage and rate limits. Whether model access runs through a central proxy such as innFactory’s AI Gateway depends on the configuration of the respective agent.

Company use: data protection and governance

Companies should settle three points before a broad rollout. First, permissions: according to its documentation, Orca pre-fills the agents’ bypass flags for new launches, such as --dangerously-skip-permissions for Claude Code, on the grounds that worktrees are disposable. A worktree only isolates files in the repository, not processes, network access or credentials, so the “Manual” setting should be chosen deliberately. Second, the cloud components: bot connectors and the mobile relay run through Orca Cloud with infrastructure in the US and Hong Kong. Third, telemetry, which is on by default for new installations and can be switched off. Independently of this, source code and prompts of the launched agents go to their model providers and belong in your records of processing.

Which agents does Orca support?

According to the documentation, more than 40 agents are preconfigured and can be set up with one click. Integrated more deeply are Claude Code and Codex (usage, account switching, plus hooks for Claude Code), Cursor CLI (deep integration per the table) and Muse Code (hooks, status, usage). Further agents such as Pi, OpenCode, Droid or Qoder CLI report status and in part hooks; many others are only launched. Any CLI that is not listed can be stored as a plain command. Claude Agent Teams are available but disabled by default.

How it fits: Orca, herdr and the AI Gateway

Orca and herdr solve the same problem with different approaches: herdr relies on long-lived terminals, minimal intervention in the agents and an interface you can shape freely, Orca on a guided environment from task to pull request. You will find a detailed comparison with further tools in the article herdr vs. Orca. None of the orchestrators captures developers’ costs and agent time across teams and projects; the AI Gateway does that in proxy and companion mode. For business users outside development, CompanyGPT is the right harness; our AI harness overview gives an overview.

From developer harness to enterprise harness

Coding harnesses are tools for developers. Business users, cost control and governance need two more layers.

CompanyGPT – the harness for business users

CompanyGPT gives any language model tools via MCP servers, integrations and n8n, company knowledge via companyRAG, roles from Entra ID or Keycloak, and audit logs – operated in your own cloud tenant (Azure, AWS or Google Cloud) or sovereignly on STACKIT.

Discover CompanyGPT
AI Gateway – cost and governance for every agent

The AI Gateway bundles the model access of your tools and coding agents behind an OpenAI-compatible API: sign-in via Entra ID, Keycloak or Cognito, budgets and hard limits per team and user, cost attribution per project and repository via the Companion, and logging of every request.

View the AI Gateway

Frequently Asked Questions

What is Orca?

Orca is an open-source desktop app by Stably AI that calls itself an "Agent Development Environment". It launches coding agents such as Claude Code, Codex or Cursor in parallel, every task in its own Git worktree, and brings terminal, editor, browser, diff review and pull requests into one interface.

Is Orca open source and free?

Yes. The stablyai/orca repository is MIT-licensed, and the app is free according to the maker. An enterprise offer is available on request.

Which coding agents does Orca support?

According to the documentation, more than 40 agents are preconfigured, including Claude Code, Codex, Cursor CLI, Gemini, GitHub Copilot CLI, OpenCode, Pi, Amp, Kiro, Devin, Droid, Qwen Code and Aider. Claude Code, Codex, Cursor and Muse Code are integrated more deeply with usage display and hooks. Any other CLI can be launched as well.

Does Orca use my Claude or ChatGPT subscription?

Yes. Orca launches the installed CLIs with their existing login, including a subscription. Several Claude and Codex accounts can be stored and switched with a click, and Orca shows usage and rate limits.

Can Orca be used in a GDPR-compliant way?

That depends on the configuration. According to the maker, prompts and code go directly to the configured model providers. Bot connectors and the mobile relay run through Orca Cloud with infrastructure in the US and Hong Kong, and telemetry is on for new installations. Both belong in your records of processing or should be switched off; the model providers of the launched agents need to be assessed separately.

Does Orca work with innFactory's AI Gateway?

Yes, through the agents it launches. Orca itself makes no model calls for agent work. Claude Code or Codex can be routed through the AI Gateway via configuration, and the gateway's companion measures tokens and agent time from the local session logs, including subscription usage and regardless of whether Orca started the session.

Introduce coding agents safely in your company

We support selection, model access in the EU, permission design and cost control.