Born inside Wexa AI. Hardened in production.
CognoDB didn't start as a product. It started as a need we couldn't satisfy with anything on the market.
At Wexa AI, we build autonomous agents that have to remember, reason, and act across long horizons — across tools, across users, across days.
We needed a context store that could spin up per agent, run at the edge, be shared across a fleet, and never block on a JVM warm-up. We tried vector DBs (no edges), document stores (no real traversal), Postgres (recursive CTEs everywhere), and Neo4j (a beautiful engine engineered for the opposite of what we needed). Nothing fit. So we built it.
Today, CognoDB powers the context-graph layer behind Wexa AI's agent platform in production, storing the relationships, memory, and task dependencies that make our agents feel like they actually know you.
We open-sourced it under Apache 2.0 because every team building agents deserves the same foundation, and because the right answer to "where does agent memory live?" should not be a vendor lock-in.
"We stopped treating agent memory as an afterthought bolted onto a vector store. CognoDB let us model context the way agents actually use it — as a living graph that's born, grows, and dies with the task."