aq.dev / docs/memory

Memory

Published October 28, 2026 · by the AQ team

Memory in AQ has two parts: shared rules every agent reads, and notes every agent can search. Both live in Settings, Organization, Memory. Nothing from memory is injected into a turn; agents search it with a tool when they need it.

Shared rules (AGENTS.md)

Write your organization's rules once. AQ keeps one shared AGENTS.md per person per machine and mirrors it into the home of every coding agent (Claude Code, Codex, OpenCode, Kimi Code, Grok and more), and into a work item's worktree only when the repository has no AGENTS.md of its own. A tracked file in your repository is never modified.

Notes from every agent

AQ reads the memory each CLI keeps on your machines: Claude Code's memory files, Codex memories, Grok, Muse Code, Command Code's taste notes and Antigravity knowledge items. It keeps a copy as AQ notes (the CLIs keep theirs), so search_memory reaches them from any agent. Cursor and Devin keep memory on their vendors' servers, and Kimi Code and OpenCode keep none; the Sources card says so.

  • Personal by default. A CLI's memory is yours. Turn on "Share project notes with the team" to make project and reference notes your team's; notes about you always stay personal.
  • Editable. Your edit wins over the file and never expires. Write a change back to the CLI's own memory where it supports that.
  • Forget for good. A forgotten note never comes back from the file.
  • Needs a look. Notes whose cited code moved, notes saved twice, and notes unused for weeks.

Agents saving notes

Agents can save a note with save_memory. Saves go under the person the thread runs as, and anything shaped like a secret is refused. Background: shared memory across coding agents.

Keep reading

  • AQ Docs: how the multiplayer agent workspace worksHow AQ works: workspaces as isolated git worktrees, real agent CLIs in persistent sessions, multiplayer access, live previews with comments, Linear and GitHub integration, and runners in your own cloud.
  • AI Coding Agent Workspaces: Isolation and LifecycleA coding agent workspace is the isolated checkout and session state one AI task runs inside. How Cursor, Devin, Codex Cloud, Conductor, and AQ each isolate and manage it.
  • Live Previews and Feedback for AI Coding AgentsHow to see what an AI coding agent actually built: run its dev server, tunnel it, or use preview deployments, and route comments back to the agent as prompts.
  • Linear Integration for AI Coding AgentsFive ways to run coding agents from Linear issues: open-in-tool handoffs, the MCP server, cloud agent delegation, Linear coding sessions, and label-driven workspaces.
  • What are multiplayer coding agents?Multiplayer coding agents are AI coding agents whose sessions are shared live places a team can enter together: watch, steer, and review one run, instead of receiving diffs from single-player tools.