The tool descriptions position krimto_write as the canonical memory call ahead of any local file or built-in session memory, on the grounds that a per-session memory directory is invisible to teammates and to your other editors — so the value on offer is not storage, it is one store several agents and people read from. Precedence is explicit rather than incidental: krimto_recall ranks user above team above org, and krimto_supersede makes a correction an object of its own with a reason attached and the superseded fact still in git. krimto_whoami is worth calling early, since the tool set assumes the caller's identity and scope list are known before anything is claimed about them.
A shared memory server with 6 tools, holding attributable facts under user, team and org scopes and keeping their history in git.
- Durable writes with provenance: krimto_write stores a fact with title, body, tags, scope and source, and can name the fact it supersedes.
- Retrieval built for a specific question: krimto_recall runs hybrid-ranked search across the scopes you name, with a limit, and resolves precedence in the order user, team, org; krimto_read fetches one fact by id with its full frontmatter.
- Correction that keeps the trail: krimto_supersede replaces a fact with a new title and body plus a reason, links the two through the supersedes field, and leaves the old one in git history.
- Orientation before you rely on any of it: krimto_list_scopes shows which scopes exist and what they contain, krimto_whoami reports the caller's identity and the scopes they can read and write.
A Krimto store the caller has an identity in, since every read and write is scoped to it — krimto_whoami reports that identity and the scopes attached to it.
One command — npx @krimto-labs/krimto init