Metobe turns your team's meetings, chats, email and tools into a living model of commitments, decisions and people, then checks every promise against what actually happened.
Illustration. Names and data are fictional.
Memory is not chat history
Most AI memory finds a similar piece of past text. Metobe turns raw information into data about the world: who promised what to whom, by when, and whether it happened.
Typical AI memory
“Find something that looks like this.”
Metobe
The model proposes an interpretation. A deterministic layer decides what changes in memory.
Commitments are first-class objects
Memory is append-only. Nothing is overwritten; current state is rebuilt from events. So Metobe can always explain why it believes something.
Event log
Current state, rebuilt from events
Auditable by design
Every fact has a source, every guess has a confidence, and the expensive questions are answered by queries, not by asking the model to reread everything.
Ask why Metobe believes something and get the exact moment it came from.
“Fine, I'll get the prototype to Maria by Friday.”Meeting · Oct 7 · 14:32
A possible match is stored with its confidence until evidence settles it.
Many names, one person.
Deterministic checks find what needs attention. Claude explains it and suggests what to do.
select * from commitments
where deadline < now()
and status = 'promised'
and evidence_count = 0;For teams
With a whole team connected, Metobe sees who waits on whom, which decisions are still in force, and where work actually gets stuck.
Waiting chains surface on their own.
Until someone decides to revisit them.
Knowledge
The company uses PostgreSQL.
Max works on the backend.
Memory
On Oct 7 the team decided to use PostgreSQL for Metobe.
Max took over the backend integration and promised it by Friday.
Built with Claude
Bigger context windows don't mean everything belongs in them. Metobe rebuilds just the context a question needs, from events, at the moment it's asked.
Meetings · chats · email · documents · calendar · GitHub/GitLab · CRM · tasks · other agents
Reads long, messy conversations. Tells a promise from a maybe. Explains state and drafts actions.
Append-only events, resolved entities, commitments and decisions with lifecycles, deterministic detectors.
Answers with sources, proactive check-ins, actions that wait for approval.
Storage for agent memory exists. Metobe decides what is worth remembering, ties it to real people and work, and checks it against what happens.
Where we are
We're looking for a few teams to shape commitment memory with us.