How it works
Team memory and rules for your agents. What axs brain puts into a session, and how lessons are saved.
axs brain gives Claude Code, Codex, Cursor and other agents one shared memory. It saves what you teach them while you work, puts the right memories into the next session before the agent makes the same mistake again, and enforces the rules your team sets.
Get started
- Install the axs CLI and run
axs login. - Connect your agents:
axs install claude,codexorcursor. Other MCP clients use the connector URL. - Work as usual. Correct the agent when it gets something wrong. axs brain asks it to save the lesson, and the next session starts with it.
Already have a CLAUDE.md, AGENTS.md or Cursor rules? Your agent offers to import them in
the first sessions.
Memories come to the agent
Agents rarely search for what they do not know they are missing, so axs brain pushes memories in at the right moment:
| When | What the agent gets |
|---|---|
| Session start | Your always-on rules |
| First work on a project, server or tool | The memories of that project, server or tool |
| Every prompt | The memories that match it |
| After changes to infra, deploy files or services | The memories that may now be outdated |
The agent can also search on its own with memory_search and read a memory in full with memory_get.
Lessons are saved while you work
When you correct the agent, axs brain asks it at the end of its answer to save the lesson as a short memory. It checks for near-duplicates first and updates those instead. Memories with secrets are refused.
You can tell it directly too: "remember that we deploy with make release" or "forget that".
When axs brain is down
Your agent keeps working without memories: every hook gives up after a few seconds. Rule checks pause too, unless an
owner sets When axs brain is unreachable to Block risky commands in the team settings.