The control layer for agent‑written codebases
A shared whiteboard of your codebase: agents write the code, you keep understanding and control.
curl -fsSL https://getaltitude.dev/install.sh | shmacOS and Linux. On Windows, download the binary manually.
works with Claude Code · Cursor · Codex · any agent
You can't read everything agents write anymore. Reviewing every diff stopped being realistic a while ago.
Letting them run without structure doesn't work either. After a month, the codebase is something nobody can explain.
Altitude is the level in between. You describe the structure — the parts, what they do, how they connect. The agent works inside that structure. How much you see and control is decided by how you draw the board: a scope you drew is yours, everything inside it is delegated.
How it works
- 1
Either of you proposes a change. It appears on the board, green, before any code exists.
- 2
You look at it. Change it, accept it, or throw it away.
- 3
The agent builds it and marks it done. If code later drifts away from the board, the board shows you where.
Bring your own coding agent
Altitude never bundles or mandates an agent. Point whatever you use — Claude, Cursor, Codex — at Altitude over the Model Context Protocol, and it gets catalogue search, the architecture map, and the propose → build → built loop.
claude mcp add altitude -- altitude mcp --allow-ui-controlTry it
Hand this to your agent:
“Read https://getaltitude.dev/llms.txt, set up altitude in this repo, and map the codebase.”Setup, mapping, and the first build all happen in the chat. Then runaltitude ui and look at your own repo.
Install Altitude
One command on macOS or Linux. Then altitude init inside any app.
curl -fsSL https://getaltitude.dev/install.sh | sh