Memory that proves its answers.

A memory and recall layer that turns language into typed mathematical structure — and answers only when it can prove the path.

What this is

CleverMemory is a memory and recall system for language models. It ingests text — notes, documents, conversations — and converts it into rich, typed mathematical structure: entities linked to identities, events with typed roles, time as intervals, every claim carrying its provenance. When you ask a question, it answers only when it can prove the matching path through that structure. When it can't, it says so — with a certificate naming exactly what was missing.

No embeddings guessing at similarity. No plausible-sounding wrong answers. Exact meaning over string matching. Proof or honest abstention.

Why it matters

Language models are extraordinary readers and writers with unreliable memories. Giving them a memory layer that never fabricates — that hands back proven facts with citations, or an honest "I don't know" — changes what you can build on top of them.

The open standard

The pack format — one addressable field for language and world knowledge, built entirely from friendly-licensed sources — is being opened as a standard: the spec, the reader, and the importers. The common language for language.

Read the blog · About the packs

Latest writing

All posts →