stablyai/orca

Orca solves a specific coordination problem that emerges once you start running multiple AI coding agents simultaneously: they scatter across terminal windows, git branches, and tool interfaces with no shared view of what any of them is doing. The answer here is a desktop application that wraps any CLI-based agent — Claude Code, Codex, Cursor, Cline, Qwen Code, and roughly two dozen others — inside a unified workspace built around git worktrees.

The worktree model is the architectural choice worth understanding. Rather than running agents in the same working directory and hoping they don't collide, Orca fans a single prompt across up to five isolated worktrees simultaneously. Each agent gets its own branch, its own filesystem state, its own terminal. You compare the outputs and merge whichever one won. This is not a novel idea in version control, but applying it as the primary unit of agent orchestration is a practical response to a real problem: parallel agents writing to the same files is chaos.

The diff annotation feature extends that review loop. You can drop inline comments on any diff line and route them back to the agent without leaving the app — the review-revise cycle stays inside one interface rather than bouncing between a terminal, a code editor, and a PR interface. GitHub and Linear integration follows the same logic: open a worktree directly from a ticket, review the result, commit, without a context switch.

The mobile companion is an unusual addition for a developer tool. The premise is that long-running agents finish on their own schedule, and you may not be at your desk. The iOS and Android apps let you monitor agent status and send follow-up instructions from a phone. Whether that workflow actually fits how people use coding agents is an open question, but the infrastructure — push notifications, remote steering — is there.

Design Mode is the most distinctive feature in the list. It embeds a real Chromium window and lets you click any UI element to inject its HTML, CSS, and a cropped screenshot directly into an agent prompt. That closes a loop that normally requires manual copy-paste or screenshot attachment: the agent sees exactly what you see, with the underlying markup attached.

The agent compatibility list is deliberately broad. The README states plainly that anything running in a terminal works, and the named integrations span models from Anthropic, OpenAI, Google, Mistral, Qwen, Xiaomi, and others. Orca is not betting on any single provider.

The project ships under MIT, runs on macOS, Windows, and Linux, and the team describes a daily release cadence — the changelog is explicitly called the real feature list because the README can't keep up.