Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Ibid Remember: for Acme, always use Net 30 payment terms."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Told twice is a rule.
Ibid is a self-managing memory and rulebook for AI assistants. The owner says how they work, in passing, while working. The database decides when a statement has become a rule: said explicitly, said about one client, or said twice in distinct contexts within 90 days. One-offs never count. Rules live on a tree of spaces (a client, a job under the client) and a set of task kinds. recall returns the rules that apply, widest first; the narrower rule wins on conflict. A rule told for two siblings moves up a level and supersedes its narrower copies.
Nothing is edited or deleted. Rows are append-only and corrections are supersessions. The owner reads a manual the system wrote, and everything it added on its own is listed with an undo.
Alongside the rulebook sits a memory store: append-only facts with confidence and importance, salience that rises with use and decays without it, and a resolver that finds near-duplicate memories and links them as restatements or supersessions. Kill signals, registered as SQL, say when a feature has failed.
It runs on plain Postgres and speaks MCP, so it attaches to Claude, ChatGPT, Claude Code, Cursor or anything else that adds an MCP server.
This is a partial fork from the internal code of stetworks.com.
Taken from Stet: the learning loop (lessons, evidence, the told-twice bar, promotion), spaces, kinds, recall, the manual, the MCP server, and the test discipline (every rule the database enforces has a test that tries to break it and is refused). Added here: the memory store, salience, the near-duplicate resolver, staged approvals, the plain-Postgres and stdio targets, the CLI and export/import. Left behind: everything about quoting, drafting, mailboxes, credentials and hosting that made Stet a product. No customer, tenant or credential data crossed over; scripts/scrub-check.mjs runs in CI to keep it that way.
Related MCP server: Munin Memory
Item 0 of the build plan: repository, license, harness, CI. See docs/PLAN.md for the build order and acceptance criteria.
npm install
npm test # real Postgres via PGlite, no services neededMIT. See LICENSE.
This server cannot be deployed
Shared memory for AI agents, as a graph in your own Postgres. Writes never call an LLM.
Versioned agent memory in your own Postgres: portable context, permissioned, audit trail.
Persistent knowledge graph for AI-augmented teams. Store decisions, findings, and standing rules across agent sessions with semantic search and typed connections. Includes cross-session memory, audit trail, workspace isolation, and secret detection. Built for teams running agents that need to remember. Free until launch with team tier as default, anon trial available.
Personal wiki and memory layer for AI assistants. Persistent, structured memory across sessions.
