# MMW — persistent memory for AI agents

> MMW documentation. Every page is available as Markdown: append .md to the page URL.

- [What MMW is](https://docs.mmwhub.tech/index.md): Long-term memory for AI agents and teams over the Model Context Protocol, with sources, statuses and isolation.
- [How MMW works](https://docs.mmwhub.tech/start/how-it-works.md): Records, workspaces, projects, keys and statuses — the big picture of what happens to a fact after remember.
- [Quickstart](https://docs.mmwhub.tech/start/quickstart.md): From sign-up to the first fact your agent recalls in a new session, in five minutes.
- [Use cases](https://docs.mmwhub.tech/start/use-cases.md): Five realistic scenarios, from one developer with two editors to knowledge handover when someone leaves.
- [Why MMW](https://docs.mmwhub.tech/start/why-mmw.md): An honest comparison with built-in assistant memory, notes files such as CLAUDE.md, and building your own vector database.
- [ChatGPT (beta)](https://docs.mmwhub.tech/clients/chatgpt.md): Connect MMW to ChatGPT as an OAuth connector in developer mode. Supported in beta.
- [Claude Code](https://docs.mmwhub.tech/clients/claude-code.md): Connect MMW to Claude Code with one command, using the local agent and, with your consent, session history.
- [Claude Desktop](https://docs.mmwhub.tech/clients/claude-desktop.md): Connect MMW to the Claude Desktop app through the local mmw-agent.
- [claude.ai (web)](https://docs.mmwhub.tech/clients/claude-web.md): Add MMW to Claude on the web as a custom connector over OAuth, with nothing to install.
- [Codex](https://docs.mmwhub.tech/clients/codex.md): Connect MMW to Codex through the local agent for memory across sessions and, with your consent, session history.
- [Cursor](https://docs.mmwhub.tech/clients/cursor.md): Connect MMW to Cursor through the local agent or over direct HTTP, globally or for a single project.
- [Other MCP clients](https://docs.mmwhub.tech/clients/other-mcp.md): Connect any MCP client to MMW over Streamable HTTP with an API key. The server is stateless.
- [Clients overview](https://docs.mmwhub.tech/clients/overview.md): Which assistants connect to MMW, how they connect, and what you get in each.
- [Telegram bot](https://docs.mmwhub.tech/clients/telegram.md): Save notes to MMW and ask your memory questions from Telegram with the @mmwhub_bot bot.
- [VS Code](https://docs.mmwhub.tech/clients/vscode.md): Connect MMW to VS Code (agent mode) over direct HTTP with an mcp.json file.
- [Agent configuration](https://docs.mmwhub.tech/agent/configuration.md): Every mmw-agent setting, with ready configs for Claude Code, Claude Desktop, Cursor and Codex.
- [The mmw-agent local agent](https://docs.mmwhub.tech/agent/overview.md): What mmw-agent does, how to run it with uvx, and its commands and environment variables.
- [Session history](https://docs.mmwhub.tech/agent/session-history.md): How mmw-agent uploads your Claude Code and Codex history with your consent, what never leaves your machine, and how to stop it.
- [File anchoring and the Stop hook](https://docs.mmwhub.tech/agent/source-anchoring.md): How mmw-agent ties facts to local files, notices when they change at search time, and reminds Claude Code to save the outcome of a turn.
- [Agent troubleshooting](https://docs.mmwhub.tech/agent/troubleshooting.md): Symptom, cause and fix for common mmw-agent problems, based on how the agent actually behaves.
- [Archivist](https://docs.mmwhub.tech/memory/archivist.md): What the MMW archivist does after every write — graph links, staleness marks and refinements — and what it never does.
- [Best practices](https://docs.mmwhub.tech/memory/best-practices.md): What an agent should and should not remember, how to name fact_key, work with sources and summaries — plus a ready instruction for CLAUDE.md or AGENTS.md.
- [Core concepts](https://docs.mmwhub.tech/memory/concepts.md): What an MMW memory record is made of — fields, statuses, the scope label, confidence, fact_key versions, summaries and graph relations.
- [Conflicts and summaries](https://docs.mmwhub.tech/memory/conflicts-and-summaries.md): How MMW surfaces disagreement between versions of a fact, and why a summary never hides unresolved conflicts.
- [Memory Guard](https://docs.mmwhub.tech/memory/memory-guard.md): What Memory Guard rejects, what it replaces with placeholders before saving, and what the agent sees in each case.
- [Source and freshness](https://docs.mmwhub.tech/memory/source-and-freshness.md): Link a record to a document with source_id and source_hash, and find out when the document changed and the fact went stale.
- [Workspaces and projects](https://docs.mmwhub.tech/memory/workspaces-and-projects.md): How MMW accounts, projects and workspaces fit together, how a key is bound to a project, and what the default workspace is.
- [Organization audit log](https://docs.mmwhub.tech/organizations/audit-log.md): What MMW records in the organization audit log, who can see it, and why it never contains memory content.
- [Organization billing](https://docs.mmwhub.tech/organizations/billing.md): Seat price, who takes a seat, the minimum, the 14-day pilot and how an organization gets activated.
- [Contractors and access terms](https://docs.mmwhub.tech/organizations/contractors.md): Time-limited access to organization memory that closes on its own, extends in one click and stops being billed when it ends.
- [Invitations and member keys](https://docs.mmwhub.tech/organizations/invitations-and-keys.md): How to add a person to an organization with a one-time link, and how they connect their agent with their own key.
- [Offboarding and knowledge handover](https://docs.mmwhub.tech/organizations/offboarding.md): When someone leaves, close their access instantly and hand all of their records, drafts included, to a chosen colleague.
- [Organizations in MMW](https://docs.mmwhub.tech/organizations/overview.md): Shared company memory next to your personal memory — one login, a space switcher and access rules enforced by the server.
- [Structure and roles](https://docs.mmwhub.tech/organizations/structure-and-roles.md): The organization's unit tree and the exact list of what an admin, head, member, contractor and auditor can do.
- [Record visibility](https://docs.mmwhub.tech/organizations/visibility.md): The four visibility levels of a record in an organization, the rule the server applies on every read, and how to change visibility.
- [What's new](https://docs.mmwhub.tech/reference/changelog.md): MMW change history — organizations, the local agent, the stateless server, memory statuses, the Telegram bot.
- [Errors](https://docs.mmwhub.tech/reference/errors.md): Every MMW server error and notice text — what it means and what to do.
- [forget](https://docs.mmwhub.tech/reference/forget.md): Soft-delete one record by ID or every record from a source — parameters, response, errors and pitfalls.
- [gateway_call](https://docs.mmwhub.tech/reference/gateway-call.md): Call a tool of a connected integration (such as GitHub) through MMW — parameters, response and errors.
- [Limits](https://docs.mmwhub.tech/reference/limits.md): Plan limits (records, calls per minute, session history) and the hard limits of the MMW server.
- [MCP tools](https://docs.mmwhub.tech/reference/mcp-tools.md): The full list of MMW tools, their annotations, transport and auth, plus a raw HTTP call example.
- [mmw_sync_status](https://docs.mmwhub.tech/reference/mmw-sync-status.md): The mmw-agent local tool — consent state and the result of the last session history sync cycle.
- [remember](https://docs.mmwhub.tech/reference/remember.md): Save a fact, decision or note — parameters, response, idempotency and every error the tool can return.
- [search](https://docs.mmwhub.tech/reference/search.md): Search memory — parameters, response fields, the unverified/stale/conflict statuses, graph relations and errors.
- [Troubleshooting](https://docs.mmwhub.tech/reference/troubleshooting.md): From symptom to fix — no tools in the client, 401, rate limits, subscription notices, claude.ai reconnect, empty search.
- [validate_memory](https://docs.mmwhub.tech/reference/validate-memory.md): Check freshness of every record from one source against its current hash — parameters, report and errors.
- [API keys and OAuth](https://docs.mmwhub.tech/account/api-keys-and-oauth.md): How mmw_ keys work, how to reveal, rotate and revoke them, how a key limits an agent to one project, and how claude.ai and ChatGPT sign in without your key.
- [Data and deletion](https://docs.mmwhub.tech/account/data-and-deletion.md): What data MMW stores, for how long, how forget differs from deleting your account, and how to request deletion.
- [Plans and billing](https://docs.mmwhub.tech/account/plans-and-billing.md): MMW personal plans with record, call and session history limits, how to pay, and what happens when a subscription ends.
- [Security](https://docs.mmwhub.tech/account/security.md): How MMW isolates data, stores keys, protects memory from secrets and poisoning, what goes into logs and what goes into backups.
- [Support](https://docs.mmwhub.tech/account/support.md): How to reach the MMW team, what to include in a request so we can help quickly, and what never to send.