Know what your agents remember.
MMW gives AI agents persistent, portable memory through MCP — with provenance, isolation, conflict states, and controlled deletion built into the model.
From a source to a governed memory.
The map is the product thesis: every remembered fact should retain context, cross a clear boundary, and remain inspectable after retrieval.
Sources
MMW control plane
Memory workspace
- Provenance and confidence
- Verified, stale, conflict
- Tenant · project · workspace
- Export and controlled deletion
Consumers
Documents enter with a source ID, hash, and revision — not as anonymous chunks.
Built far enough to test the hard parts.
Current evidence is technical, not commercial. MMW is a working pre-seed staging product seeking its first design partners.
- 193
- backend tests passed in the current staging release
- 26
- frontend unit tests
- 20
- browser end-to-end tests
- 1.0
- current API contract
External MCP lifecycle verified with synthetic data
- Remember
- Search
- Forget
The product works. The company is being formed around it.
MMW is pre-revenue, has no claimed customers or pilots, and is not yet production-ready. The next milestone is a small group of design partners who need memory they can inspect, move, and delete.
- Working now
- MCP memory operations, provenance, isolation, conflict and stale states, APIs, dashboard
- Next
- Design-partner validation, production hardening, customer onboarding, usage metering
- Founder
- Mikhail Singatullin — solo technical founder, 22 years across infrastructure, software, AI, and automation