Archilas gives AI a memory of the team. For engineering teams, that means the agent in Claude Code, Cursor, or Codex can ask what was decided about a service, which pull request changed it, and what is still open, and get an answer with dated sources instead of a guess.
What the agent can answer
- "Why did we move the job queue?" from the Slack thread and the pull request.
- "What is still open on billing?" using outcome words such as open, blocked, and todo.
- "Was the Postgres upgrade rejected?" using words such as rejected, dropped, and ruled out.
- "What changed this week?" through archilas.changes.
How to set it up for a team
- One person creates the workspace and connects GitHub, Slack, Linear, and anything else the team uses.
- Each developer adds the MCP server to their editor with a key, or runs the installer.
- Agents save decisions with archilas.remember as work happens, so the next person's agent has them.
- Plans differ by questions and seats. Free is one seat with 100 questions a month; Studio and Team add seats.
Rules files still matter
Keep CLAUDE.md and Cursor rules for fixed conventions like the lint command. Archilas holds the decisions that change, with a date and a source.
Questions
- Does Archilas read our code?
- It syncs GitHub activity from repositories you connect, such as commits, pull request reviews, and comments. It is not a code search index.
- Does it change anything in GitHub or Slack?
- Connectors only create webhooks so updates arrive, and remove them on disconnect.