memkeeper blog https://memkeeper.ai/blog/ Notes on building a persistent memory layer for AI agents: retrieval, benchmarks, and local-first design. en-us Fri, 26 Jun 2026 18:00:00 GMT A harder benchmark, and it still tells you when it can't answer https://memkeeper.ai/blog/benchmarking-agent-memory-longmemeval.html https://memkeeper.ai/blog/benchmarking-agent-memory-longmemeval.html Fri, 26 Jun 2026 18:00:00 GMT memkeeper on LongMemEval, a long-term-memory benchmark with about fifty sessions of history per question: competitive answer accuracy, local and zero marginal cost, and the full abstention number, 0.833, the category most leaderboards leave out. What your AI coding subscription actually buys https://memkeeper.ai/blog/what-your-ai-subscription-buys.html https://memkeeper.ai/blog/what-your-ai-subscription-buys.html Fri, 26 Jun 2026 15:00:00 GMT A receipt for your AI coding plan: about 41x list-price-equivalent leverage on a $100 subscription, and the surprise that 96% of the tokens are the model re-reading context it has already seen. Cheap, local, and it tells you when it can't answer https://memkeeper.ai/blog/cheap-local-agent-memory.html https://memkeeper.ai/blog/cheap-local-agent-memory.html Fri, 26 Jun 2026 12:00:00 GMT memkeeper matches the published answer-accuracy of commercial agent-memory systems on LoCoMo, running locally at zero marginal cost, and reports its abstention behavior, including the adversarial category most leaderboards leave out. Documents aren't memories https://memkeeper.ai/blog/documents-arent-memories.html https://memkeeper.ai/blog/documents-arent-memories.html Wed, 24 Jun 2026 12:00:00 GMT A memory and a document are different things. Why memkeeper keeps curated atomic memories and ingested document chunks in separate tiers, with a one-way promotion bridge between them. Benchmarking AI agent memory on LoCoMo https://memkeeper.ai/blog/benchmarking-agent-memory-locomo.html https://memkeeper.ai/blog/benchmarking-agent-memory-locomo.html Tue, 23 Jun 2026 12:00:00 GMT We ran memkeeper against LoCoMo, a long-term conversational QA benchmark, and published the retrieval numbers, the method, and the script to reproduce them. Why hybrid retrieval beats pure vector search https://memkeeper.ai/blog/hybrid-retrieval-vs-vector-search.html https://memkeeper.ai/blog/hybrid-retrieval-vs-vector-search.html Tue, 23 Jun 2026 12:00:00 GMT Vector search is necessary but not sufficient for agent memory. Why memkeeper pairs dense embeddings with BM25 and a cross-encoder reranker. Local-first memory for AI agents https://memkeeper.ai/blog/local-first-memory-for-ai-agents.html https://memkeeper.ai/blog/local-first-memory-for-ai-agents.html Tue, 23 Jun 2026 12:00:00 GMT Your agent's memory is your context. The default in this space is a hosted vector database. We think the default should be your own machine. A capability broker for AI coding agents https://memkeeper.ai/blog/warden-capability-broker.html https://memkeeper.ai/blog/warden-capability-broker.html Tue, 23 Jun 2026 12:00:00 GMT Warden is a deny-by-default gate that decides what an AI coding agent is allowed to do. We logged 27,860 real decisions, listed the evasions it provably stops, and wrote down where its edges are.