AgentVera Brain: a memory of why the project is built this way
The reason behind a decision usually lives in a PR description, a commit message or a forgotten agent conversation. Brain reads those sources, lists decisions with their reasons, flags the ones that no longer match the code and, if you want, gives new agents in the project a short context pack.
Sources
Build the memory with “Build memory” in the Brain tab of a project’s Details page, and update it with “Refresh memory”.
- Git-tracked docs and ADRs: README, ARCHITECTURE, CONTRIBUTING, CLAUDE, AGENTS, CHANGELOG, docs/, adr/, decisions/ and similar, up to 250 files.
- The newest 1,500 non-merge commits and the files they touched.
- Up to 200 PRs and 200 issues, if the repository is on GitHub and you are signed in.
- Up to 120 agent conversations in the project.
Decisions and their status
Sources that explain a reason (ADRs, long PR descriptions, commits with “because”, “instead of”, “switched to”, conversations) are read by Claude Haiku. Each decision is saved with a title, its reason, a date, its sources and related code terms; duplicates are merged.
- Statuses: current, stale, superseded. A decision is stale when its code terms no longer appear in the code or its files are gone.
- “Compare with code” reruns that check with git grep; “Verify with Claude” has a read-only Claude Sonnet run judge one decision from file evidence.
- The knowledge graph shows decisions and the code terms they share.
Ask the project, and the context pack
Ask a question like “Why do we use JWT?” and the most relevant decisions and sources are picked; Claude Sonnet answers read-only and cites them. The context pack sums up what the project is, its stack, key decisions, conventions and recent changes in at most 2,500 characters. With “Give new agents the context pack” on, every new agent started in the project gets this summary, plus a list of decisions that may no longer hold, at the end of its instructions. This option is off by default.
Get started
- 1
Open the Brain tab on a project’s Details page and click “Build memory”.
- 2
Filter the decisions list for stale ones and check any with “Verify with Claude”.
- 3
Review the pack in the context pack view and turn on giving it to new agents.
Questions
Does the memory update itself?
No. You update it with “Refresh memory”.
Where is the data kept?
On your computer, in one file per project in the app’s data folder. To extract decisions and answer questions, source text goes to Anthropic through your Claude Code login.
Does it work without GitHub?
Yes. Docs, commits and conversations are still read; PRs and issues are added only when the repository is on GitHub and you are signed in.
How do agents use Brain?
With the context pack on, it is added to new agents’ instructions. There is no separate Brain tool agents can query.
Related features
Bring your agents to one desk.
Start on the free plan. Download it, add a project folder and open your first agent.