CortexDB V1: The Experience Layer for AI Agents — CortexDBLoginGet Started

Blog/V1 Launch<br>AnnouncementMay 16, 202618 min read<br>CortexDB V1: The Experience Layer for AI Agents<br>V1 ships today. 93.8% on LongMemEval-S — past Mem0's published 93.4% — and 86.9% on LoCoMo, with a five-layer memory model, hierarchical scopes, PASETO authentication, and 53 integrations. Eight weeks from first commit to state-of-the-art.

93.8%<br>LongMemEval-S<br>469 / 500

86.9%<br>LoCoMo cats 1–4<br>1,339 / 1,540

53<br>Integrations<br>37 frameworks + 16 connectors

8 wks<br>First commit to V1<br>498 commits

Prashant Malik<br>Co-creator of Apache Cassandra, among the first 30 engineers at Facebook. Two decades of distributed-infrastructure work for the systems that power the internet. Building CortexDB now — the database for what an AI agent has lived through.

The missing third layer<br>Today's AI agents are amnesiacs. Open ChatGPT, tell it your name, log out, come back tomorrow — it has no idea who you are. For a chatbot that's annoying. For an agent — one that books travel, manages a customer pipeline, debugs production systems — it's broken.<br>People have tried to paper over this by stuffing more text into the prompt. The prompt is a clipboard, not a memory. You can't enforce "delete this customer's data" on a clipboard. You can't audit who saw what. You can't answer why do you think the deal is at risk? from a clipboard.<br>A real AI agent needs three layers, stacked together:<br>Layer 1<br>Intelligence<br>The LLM: reasoning, language, planning. Built by OpenAI, Anthropic, Google, Meta.<br>Solved — commoditizing fast.

Layer 2<br>Knowledge<br>RAG over docs: what's true about the world. Pinecone, Weaviate, vector pipelines.<br>Solved — becoming standard infra.

Layer 3<br>Experience<br>What this agent has lived through. Events, episodes, facts, beliefs — specific to this user, this org.<br>Missing. This is what we build.

Layers 1 and 2 are commodities. Layer 3 is the moat — and Layer 3 is what CortexDB builds.<br>Memory isn't storage. It's a cycle.<br>Real memory — the kind your own brain runs — isn't a single operation. It's a continuous cycle of operations the brain performs, over and over. Skip any one and you're back to a vector DB.<br>Capture ──▶ Extract ──▶ Reconcile ──▶ Forget ──▶ Consolidate<br>(WAL) (facts + (bi-temporal (lifecycle ("sleep" —<br>entities) supersession) methylation) the brain)<br>Capture. Ingest raw experience. Conversations, events, observations — preserved exactly as they happened in a tamper-proof WAL.<br>Extract. Pull facts, entities, relationships out of raw events. Turn experience into structured knowledge.<br>Reconcile. When memories contradict — Bob said yes Tuesday, no Friday — decide what's true now. Newest wins, with the older version preserved as true until Tuesday.<br>Forget. Prune the irrelevant. Not all memories matter. The brain forgets on purpose.<br>Consolidate. When the agent "sleeps," build a coherent worldview from raw events. Synthesis. Generalization. Procedure formation. This is where memory becomes intelligent.<br>V1 ships the first four stages as stable surface. Consolidation is in flight as a beta endpoint — POST /v1/understanding/synthesize.<br>The five layers<br>If the cycle is what memory does, the layers are what memory contains. Five layers, stacked from raw to refined, each addressable through its own API endpoint:<br>Events<br>"14:22 — message received from [email protected]"<br>Immutable, atomic captures. The WAL is the source of truth.

Episodes<br>"The Acme deal — May 2 to May 13"<br>Bounded spans of related events. Sealed once consolidated.

Facts<br>"(ent_acme, deal_stage, signed)" — valid 2026-05-13 → now<br>Bi-temporal triples. Supersedable. Older versions preserved.

Beliefs<br>"Acme is likely to renew · confidence 0.62"<br>Probabilistic claims with a supports[] graph. Walkable evidence trail.

Understanding<br>"Concept: Q3 renewal motion — version 7"<br>Synthesized concepts that span many beliefs and episodes.

A system with only Events is a log. A system with only Facts is a database. A system with all five layers, with the right relationships between them, is the experience layer that completes the agent.<br>Almost no AI memory product handles Beliefs properly. They have facts. They don't track confidence or evidence. They can't answer "why do you think that?" — and that question is one of the biggest reasons enterprise buyers don't trust agents yet. CortexDB can: GET /v1/beliefs/why?belief_id=… returns the full support graph plus a narrative rendering.<br>The benchmarks<br>Anyone can claim great AI memory. There are exactly two public, standardized benchmarks the research community has rallied around — LongMemEval-S and LoCoMo — and V1 hits state-of-the-art on both.<br>LongMemEval-S<br>ICLR 2025 — 500 questions, six memory-skill categories.

93.8%<br>469 / 500 · server parity

single-session-assistant56 / 56100.0%<br>knowledge-update76 / 7897.4%<br>single-session-user67 / 7095.7%<br>single-session-preference28 / 3093.3%<br>temporal-reasoning122 / 13391.7%<br>multi-session120 / 13390.2%

Single run, no retry targeting, no...