Agent Memory
Memory and coordination for a fleet of agents: one service on TessariDB, reached over MCP, REST and a command line.
TessariDB Agent Memory keeps what an agent framework usually keeps in files — goals and the evidence behind them, plans and waves, tasks and claims, checklists, decisions, risks, and what each agent remembers — as records in TessariDB. One service holds them; agents reach it over MCP, services over REST, and people and scripts over a command line.
Status. This section describes Agent Memory 0.4.1-alpha, on TessariDB 0.33.1-beta. It is alpha: the contract may change in any 0.x release.
Why not files
Files work for one agent and stop working when several run at once. There is no transaction, so two agents writing one registry race and the second wins; a lock file left by a process that died freezes the project; nothing records who changed what. None of it is an error anybody sees.
In the store, the same state gets real transactions instead of advisory locks, search over the same records, a history of every change attributed to the session that made it, claims that lapse rather than linger, and a view of which agent holds what right now.
Where to go next
- What it holds — the record families, and the two pairs that look alike and are not.
- Getting started — the container, and connecting Claude Code or Codex to it.
- Surfaces and verbs — the one contract the three surfaces render, and the verbs in it.
The product site is am.tessaridb.com. Its store is TessariDB, documented here.