Loading this screen…
Brain indexes committed repository source and offers context to your coding . Reindex when source changes, and verify retrieved evidence in the code. These six steps configure it; changes go straight to the Brain server.
4 of 6 steps doneNext: Connect your agent
Done Postgres, Redis and Neo4j answer.
Done /home/you/code/project-brain.
Done Both providers answered a real call.
Done Last run completed.
Your coding agent (Claude Code, Cursor, Codex…) gets a key and a config, then can ask Brain for context through its tools. It still needs to inspect source and validate changes.
Now No agent has connected yet. Next Create a key below and paste the config into your agent.
Name the key after the machine or agent that will use it. You get a separate key that can be revoked, limited to agent permissions, plus a config to paste into your agent. Self-check calls the same tools from the server to prove they work. Every key is listed in Access.
Connect the Brain API to run actions — this is demo data.
Describe a task the way you would give it to an agent. Brain selects files, rules and decisions for a context pack; the selection may be incomplete or noisy. Verify it against current source. Build a new pack for each task and reindex committed source changes. This step checks that a pack can be returned; it does not establish coding or cost improvements.
Now No context pack yet. Next Describe a task below and build one — that is the whole point of the install.