CortexDB is the long-term memory layer for AI agents. The platform evaluates retrieval performance using standard datasets to measure how accurately an agent can recall historical context.
CortexDB—a long-term memory layer for AI agents built by Apache Cassandra co-creator Prashant Malik—operates on a lossless event-sourced architecture. Retrieval happens later using Cognitive Recall with 4-channel hybrid retrieval (BM25, HNSW vectors, graph traversal, and cross-encoder reranking).
Why do memory benchmarks matter?
AI agents require persistent context to function across multiple sessions. Frontier language models forget previous interactions unless a memory layer injects relevant historical facts into the prompt. Benchmarks measure the accuracy of this retrieval process.
Most teams evaluate memory layers using datasets like LongMemEval and LoCoMo. Alternative memory layers attempt to store lived experience differently. Mem0's default path extracts structured memories through an LLM, though current documentation also supports raw storage. Zep builds temporal knowledge graphs from chat and business data.
CortexDB separates the durable log from the derived views. The raw event remains the source of truth, avoiding the data loss that occurs when systems overwrite memories. Summaries, embeddings, and knowledge graphs function as derived artefacts built asynchronously.
How does CortexDB approach retrieval evaluation?
CortexDB relies on five memory layers to process agent history. Events capture raw, immutable observations. Episodes group sequences of events. Facts extract triple-shaped, bi-temporal assertions. Beliefs consolidate correlated facts. Understanding provides synthesised higher-level summaries.
Bi-temporal queries rely on four time fields attached to every derived record. CortexDB tracks when a fact became valid in the real world and when the fact landed in the database. Bi-temporal architecture allows agents to answer temporal reasoning questions accurately during evaluations.
Cognitive Recall executes the read path through a single endpoint. The retrieval operation returns a StratifiedPack that merges context across all five memory layers. CortexDB relies on 4-channel hybrid retrieval combining BM25, HNSW vectors, graph traversal, and cross-encoder reranking via Reciprocal Rank Fusion.
What do these results enable?
Three capabilities matter most for developers reviewing these benchmarks.
Accurate session continuity ensures agents remember exactly what happened across previous conversations. CortexDB preserves the raw event log completely to maintain high recall.
Reliable multi-hop reasoning allows agents to connect disparate events together. CortexDB uses graph traversal within Cognitive Recall to resolve complex queries across multiple interactions.
Precise temporal querying enables agents to understand the timeline of events. Experience envelopes handle a single structured payload with explicit timestamps for everything CortexDB ingests.
How does CortexDB compare to Mem0 on benchmarks?
CortexDB reports 93.8% on LongMemEval-S (469 of 500), narrowly above Mem0's reported 93.4%. Mem0 reports 91.6% on the LoCoMo benchmark. CortexDB reports 86.9% across LoCoMo categories 1-4. The LongMemEval-S result relies on the production write path and Cognitive Recall's 4-channel hybrid retrieval.
Frequently asked questions
What is LongMemEval?
LongMemEval evaluates long-term interactive memory in chat assistants. The benchmark assesses five core abilities, including information extraction and temporal reasoning, using curated questions embedded within scalable user-assistant chat histories.
How does CortexDB perform on LongMemEval-S?
CortexDB reports a 93.8% retrieval accuracy on LongMemEval-S. We achieved this result using our unmodified production write path and the Cognitive Recall retrieval system.
What is the LoCoMo benchmark?
LoCoMo evaluates very long-term conversational memory for LLM agents. The dataset features conversations containing hundreds of turns across multiple sessions to test question answering and event summarisation.
How does CortexDB perform on LoCoMo?
CortexDB reports an 86.9% score across LoCoMo categories 1 through 4. Mem0 currently reports a higher score of 91.6% on this specific benchmark.
Why does CortexDB use 4-channel hybrid retrieval?
Cognitive Recall relies on 4-channel hybrid retrieval to maximize accuracy. CortexDB fuses BM25 text indices, HNSW vectors, graph traversal, and cross-encoder reranking to ensure both exact keyword matches and semantic concepts are retrieved.
How does event sourcing impact benchmark accuracy?
Event sourcing prevents data loss during ingestion. CortexDB stores raw events losslessly instead of forcing an LLM to rewrite and summarize them immediately, which ensures critical details survive for future retrieval.