CueMap
A deterministic, Rust-built memory engine for AI agents, offering sub-millisecond recall over millions of stored memories.
🔗 Visit CueMap
Description
AI agents that need to remember things across sessions usually lean on a vector database, but plain similarity search doesn't capture how memory actually works — some things matter because they're recent, others because they keep coming up, others because of the specific context they happened in. CueMap is a purpose-built memory engine that models that more directly: it filters memories through overlapping contextual cues, balances fresh information against high-impact past events, and reinforces memories that get accessed frequently — while staying fully deterministic, so the same query reliably returns the same result.
It's built in Rust for speed, claiming sub-3-millisecond average recall even across a million stored memories, and ships with an MCP server so AI coding assistants can plug into it directly, plus a CLI and HTTP API for everything else. It supports multi-project isolation, encryption at rest, and tree-sitter-powered code analysis for codebase-context use cases. Licensing is BSL-1.1 (converting to Apache 2.0 after four years) — free for development, testing, and self-hosting, with a separate commercial license required to resell it as a closed-source SaaS.
💬 Our review
The short version: a technically serious piece of infrastructure if you're building AI agents that need genuinely reliable, fast, contextual memory rather than a generic vector-search bolt-on — the determinism and sub-millisecond recall claims are the differentiators worth verifying yourself on your own workload.
Compared to general-purpose vector databases (Pinecone, Weaviate, Chroma) that store embeddings and let you handle recall logic yourself, CueMap bakes in the memory-specific ranking (recency, reinforcement, contextual cues) as a first-class feature rather than something you build on top. The BSL-1.1 license is worth reading carefully — free to self-host and develop with, but if your plan is to resell it as a hosted service, you'll need a commercial license. A strong pick for teams self-hosting AI agent infrastructure; probably overkill if you just need basic semantic search.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Gratuit en développement/test/auto-hébergement (BSL-1.1, passe en Apache 2.0 après 4 ans) ; licence commerciale pour usage SaaS fermé.
Pros
Rappel sous la milliseconde même à 1M+ mémoires (Rust)
Classement déterministe (récence, renforcement, contexte) au-delà de la similarité simple
Serveur MCP pour intégration directe aux assistants de code IA
Gratuit pour auto-hébergement, dev et test
Cons
Licence BSL-1.1 exige une licence commerciale pour revente en SaaS fermé
Pas d'offre hébergée gérée mentionnée, auto-hébergement seulement
Produit technique de niche, difficile à évaluer sans test pratique