Building the memory layer for AI
AI agents are remarkably capable — but they forget everything between sessions. CortexDB fixes that. We are building the infrastructure that gives AI systems durable, accurate, structured memory.
Our story
CortexDB was created by Prashant Malik, co-creator of Apache Cassandra. At Facebook, Prashant helped build the distributed infrastructure that serves billions of users. That experience taught a fundamental lesson: the hardest problems in infrastructure are about what happens to data over time.
When the AI agent revolution began, we saw the same mistakes being repeated. Memory systems that rewrite data through LLMs, losing information on every write. Vector databases used as memory layers, with no understanding of time, causality, or relationships. Systems that work in demos but fail at production scale.
We built CortexDB to bring database-grade rigor to AI memory. Event sourcing instead of LLM rewriting. Write-ahead logging for crash durability. Hybrid retrieval for accuracy. Distributed architecture from day one. The same principles that power financial systems and distributed databases, applied to the unique demands of AI workloads.
Our principles
Store the truth, never rewrite it
Every piece of data enters CortexDB as an immutable event. The original content is the source of truth. Everything else — knowledge graphs, vector indexes, search indexes — is a derived view that can be rebuilt at any time.
Production-grade from day one
CortexDB is built in Rust with crash durability, write-ahead logging, and distributed consensus. We test against SIGKILL. We measure in milliseconds. We do not ship features that are not production-ready.
Accuracy over convenience
We run three retrieval engines in parallel and fuse results with Reciprocal Rank Fusion. It is more complex than a single vector search, but it achieves 84% accuracy where simpler approaches manage 32%.
Built for engineers
CortexDB is infrastructure, not a product demo. We provide SDKs, not dashboards. We document edge cases, not just happy paths. We respect our users' time and intelligence.
Join us
We are building the memory layer that every AI system will depend on. If that excites you, we would love to hear from you.