One project per company or client. The brain creates the project ID, the first API key and the MCP connection for your agent from the name alone.
Keep it in the agent's configuration; it stays viewable on the Overview and the Agents page. The key carries the project ID, the role and the agent ID; the agent sends it as a Bearer token on every MCP request.
Connect the agent: pick its type and paste.
Every agent needs the MCP endpoint and its own key. Pick the agent, copy the command for its type and paste it into that agent. The key identifies the agent on every call, so give each agent its own.
No agent yet. Add agent creates the first key.
The commands above carry a placeholder where the key goes.
One row is one agent and its key. The key is the agent's identity: it goes in the Authorization header of the agent's MCP connection, and the brain reads the project, the role and the agent from it. connect shows the key and the commands at any time, new key replaces one.
| Agent | Role | Key | Status | Last sync | Synced to | |
|---|---|---|---|---|---|---|
| No agents yet. | ||||||
| Agent | Role | Transport | Client | Connected | Last seen |
|---|---|---|---|---|---|
| No live sessions. Agents appear here while they talk to the brain. | |||||
Merging needs a connected LLM (GitHub & LLM page). Replace works without one; the file needs at least one ## heading.
Upload the documents your agents must know: SOPs, specs, contracts, research. Agents read them over MCP with list_documents, search_documents and read_document; you decide which agents see each file.
Drop a file here or
Markdown, text, CSV, JSON, YAML, HTML, PDF or Word (.docx), up to 20 MB.
Agents added later with one of the ticked roles see it too.
| Title | File | Size | Shared with | Uploaded | |
|---|---|---|---|---|---|
| all agents | download | ||||
| No documents yet. Everything you upload becomes part of the agents' knowledge base. | |||||
Ask questions over the brain files, the findings and the uploaded documents, or tell the assistant what to change in a brain file and apply its edits. Connect an LLM on the GitHub & LLM page first.
Answers come from the brain files, recent activity and the documents shared in this project. Edits are proposed first and applied only when you approve them.
Click an entry to see its commit diff.
Writes from roles with rw-approve wait here. Approving applies, logs, commits and broadcasts the change.
These sections of ACCESS_PERMISSIONS.md travel to every agent through the MCP connection. Edit them to change what agents update and when; they apply on the next whoami.
Connect the company repository. Every push lands in the company log, and with an LLM the brain reads the code and fills the design system, stack, schema and product facts itself, so the brain stays current even when no agent is working.
Connect an LLM below to analyse the repository.
Used only by this server: to analyse the repository and to merge uploaded documents into the right sections. Your agents bring their own models and never need it. The key is stored in .env on this server.
No backup repository yet. One click creates a private repository on your GitHub and pushes every change there.
To delete this project use the Delete button at the top of the page.Delete permanently
This removes the vault with all brain files, every API key, the documents, the ledger and the proposals of . Connected agents lose access immediately.
The backup repository stays on GitHub with the full history.