Memori

Memori is a memory infrastructure layer designed specifically for AI agents. It captures conversations and execution traces and structures them into facts, preferences, rules, and summaries — the kinds of things agents need to recall in production.

The retrieval side uses semantic search with targeted recall instead of stuffing entire transcripts into the context window. Memori reports a 95% reduction in token usage versus full-context retrieval, with 81.95% accuracy on its benchmarks.

Memori Labs ships a freemium tier and integrates with CockroachDB for storage. The target is teams running agents in production who hit the wall of cost and latency from naive context management.