The IDE for agents.
You make the calls.

Plan, build, review and test with agents, each in its own sandbox on your machine. runcastle runs the whole loop. You step in only for the decisions.

curl -fsSL https://runcastle.dev/install.sh | bash
Read the docs

Talk to one orchestrator. It cuts the work.

Bring a raw idea. The orchestrator checks it against what's already built, cuts it into features, and gives each one its own agent.

Every ticket runs in its own sandbox.

Each feature agent works in rounds. Every ticket gets a fresh sandbox and its own worker. Only what failed goes into the next round.

See it in your app before it's built.

The agent runs your real frontend and designs with your own components. Tell it what to change in the chat. When you like it, it commits the design and builds from there.

Watch the agent use what it built.

Every round comes back with a review package: a video of the agent clicking through your app, a screenshot of each screen, and anything it found wrong.

Test drive from any state.

Pick a scenario and the feature's branch starts with that data already in place. A new account, a busy month, an overdue customer. No stashing, no migrations, no setup by hand.

Ctrl+J. Noted for later.

Jot down what you notice from anywhere in the app, without losing your place. When you're ready, the orchestrator groups your notes, asks about the unclear ones, and routes each to a feature.

Answer from anywhere.

Turn on the private link and scan the code. Your phone gets every chat, design and walkthrough over Tailscale, and a notification when an agent needs you.

Private link

Only your own Tailscale devices can reach it.

Start building.

One command installs it. Then run runcastle in any repo.

curl -fsSL https://runcastle.dev/install.sh | bash