Agents on your desktop. Streamed to your phone. Organized per repo.
OrchestraCode runs the Claude Code and Codex CLIs on your desktop (macOS, Windows or Linux) and streams every task, chat, diff and tests, live to the web app and your phone. Or move a task to a cloud sandbox with your own key. Every task gets its own branch, preview and pull request, organized per repo, with review comments pulled in and a share link for anyone who needs to see it.
Your machine first. Everywhere second.
Runs where your code is.
The desktop app drives the Claude Code or Codex CLI on your machine, one git worktree per task. Your repo, your login, your files. Nothing is uploaded unless you say so.
Watch from anywhere.
Turn on streaming and the web app and your phone mirror chat, diffs and test output live. Send follow-ups from the train; the laptop does the work.
Keep going without the laptop.
Need it to run overnight or for the whole team? Start the task in a cloud sandbox with your own API key. Same branch, same preview, same PR.
Everything a task needs, in one place.
The same shell on desktop, web and phone. Collapse the sidebar to a rail, switch dark or light, and every screen is a click away.
Pick where a task runs when you start it. Local keeps everything on your machine; stream mirrors it to web and phone; cloud takes over with your own key. Offline laptop? Messages queue until it is back.
Repos are groups in the sidebar, each with its own filters: latest, without PR, open changes, archive. One task can target several repos, frontend and API in one go.
Every task is a branch. Every finished task can open the pull request. GitHub, Bitbucket and Azure DevOps.
PR review threads sync into the task. The agent proposes a fix, you approve, it pushes, replies and resolves the thread.
Send a read-only transcript with diff and tests to a teammate. They can comment; you can turn a comment into a follow-up.
Let a random model take each task and score the result before its name is revealed. Your own scoreboard tells you which model earns its keep.
Expanded, each repo shows its filter row and tasks with status, open changes, PR and review counts. Collapsed, it is a rail of repo initials with running badges.
The desktop app finds your Claude Code or Codex install, checks it is unmodified, and waits while you sign in inside the CLI. Then connect GitHub, Bitbucket or Azure DevOps.
Claude through Claude Code, GPT through Codex, and Kimi, GLM or anything else through OpenRouter or your own keys. Cost per task, always visible.
Your laptop's tasks, in your pocket.
The phone mirrors whatever the agent is doing on your desktop or in the cloud. Dictate a task, answer a question the agent has, read the diff. If the laptop goes offline, messages queue until it is back, or move the task to the cloud with one tap.
Tap the mic, describe the change in plain words.
A task finishes, a test fails, an agent needs input.
Local tasks stream chat and diffs only; files stay put.
Works with the Claude Code and Codex you already have.
OrchestraCode launches the Claude Code and Codex binaries exactly as their vendors publish them. You sign in inside each CLI with your own account or key; the app never sees, stores or proxies credentials. Other models run through your own OpenRouter or provider keys. Usage is billed to you, under your own agreements.
- Unmodified Claude Code binary, detected and checksummed
- Sign in inside the CLI · subscription, API key, Bedrock, Vertex, Foundry
- Every built-in auth method stays available; none are hidden
- No credential sharing, pooling or resale
- Codex CLI with your own OpenAI login or key
- Other models through your OpenRouter or provider keys
- Cloud mode runs on your own API key
OrchestraCode is an independent product and is not built, endorsed by, or affiliated with Anthropic. Claude and Claude Code are trademarks of Anthropic, PBC; Codex is a trademark of OpenAI.
One platform, many ways to ship.
Run Claude Code untethered from your laptop. Tasks keep running after you close the tab.
Read more →/claude-code-background-agentClaude Code for your whole team.Hosted sessions, audit trail, shared task history, no per-laptop setup.
Read more →/claude-code-as-a-serviceA managed runtime.We handle the sandbox, the git plumbing and the previews. You bring the key.
Read more →/ai-coding-agent-apiAI coding agent API.Spin up agents from your own product or workflow, programmatically.
Read more →/build/coding-agentBuild with us.White-label the OrchestraCode runtime inside your own app.
Read more →/agentic-codingAgentic coding tools, compared.Sixteen tools across four categories, honest tradeoffs, real prices.
Read more →Get early access.
Private beta, limited to 200 teams. Desktop (macOS, Windows, Linux), web and mobile apps ship together.
Private beta · no card needed