Orca: the ADE that grounds parallel agents in git worktrees

What it is

Orca defines itself as an ADE, an Agentic Development Environment, under the tagline "the AI orchestrator for 100x builders". Its core move is to run Codex, Claude Code, OpenCode, and Pi side by side, each in its own git worktree, all tracked in one place. It has more than 75,000 stars, is MIT-licensed and written in TypeScript, and is YC-backed.

The positioning needs stating precisely: a harness answers "how does one agent run", Orca answers "how do many agents run at once and stay manageable". It is not a replacement for Codex or Claude Code - it treats them as things to be orchestrated. Its repo topics list codex, claude-code, opencode, and pi together, which tells you it is built for the reality that users already have several agent CLIs.

Parallel worktrees: the design that matters most

Fan one prompt across five agents, each isolated in its own git worktree, then compare results and merge the winner. That is Orca's central mechanism, and it is what separates the product from "open five more terminals".

Worktrees are native git: one repository can have several working directories checked out at once, each with its own branch and work area, sharing a single object store. Choosing it as the isolation unit buys three things. You do not copy the repository, so you save disk and clone time. Branch semantics come for free, so each agent's output is a branch you can diff, review, or throw away. Merging back to the trunk uses the git flow developers already know. Grounding "try several solutions in parallel" on worktrees turns agent uncertainty into an ordinary version-control object.

That is also why it calls itself an ADE rather than an orchestrator: what it orchestrates is working directories and branches, not a message queue.

The rest: everything oriented around not switching context

The remaining features share one throughline - letting you review and push work forward without leaving Orca.

At the terminal layer it offers Ghostty-class terminals with WebGL rendering, infinite splits, and scrollback that survives restarts. WebGL rendering is there so the UI does not stall when many parallel agents are flushing output at once; scrollback persisting across restarts is there because you will close the window while agents run long.

Design Mode is the interesting one: click any UI element in a real Chromium window and it sends that element's HTML, CSS, and a cropped screenshot straight into the agent's prompt. This addresses the hardest step in the frontend feedback loop - "the spacing on this button is wrong" is nearly impossible to convey precisely in words, while an element selector plus a screenshot can.

GitHub and Linear are native integrations: browse PRs, issues, and project boards in-app, open a worktree from any task, and review without a context switch. Annotate AI Diffs lets you drop comments on any diff line and ship them back to the agent, so review, edit, and commit all happen inside Orca. SSH worktrees run agents on a beefy remote box with full file editing, git, and terminals, including auto-reconnect and port forwarding.

The mobile companion is genuinely functional: you get notified when an agent finishes and can send follow-ups from anywhere, with iOS on the App Store and TestFlight and an Android APK. For parallel agents the mobile app is not decoration - when five agents run at once your attention is the scarce resource, and notifications decide whether you need to get back to the desk.

Where it sits in agientry

It is on the same layer as herdr, multica, and paperclip but at a different cut. herdr is a terminal runtime that takes over the terminal processes of your agents; Orca is an IDE shape that takes over working directories and branches; multica turns the workflow into issues on a board; paperclip goes one level higher and turns a team of agents into a company. The four stack: Orca or herdr runs agents underneath, multica or paperclip assigns work on top.

The README offers entry points in Chinese, Japanese, Korean, Spanish, French, and Portuguese, the site is onorca.dev, and the docs are complete. We have not benchmarked or reproduced Orca; its star count and its central position in the orchestration layer mean it has to be carried. This entry is graded as needing reproduction.