memoryaicm Pro

memoryaicm gives long-term memory to any AI assistant or agent: Claude Desktop, Claude Code, Cowork, a local model (Ollama, llama.cpp), OpenAI, Gemini, or your own agent loop. Everything runs on your machine, in pure Python, with no vector database: a single SQLite file is the source of truth. - Nothing leaves your machine, no vendor lock-in. - Auditable: append-only, SHA-256 hash-chained log. Forgetting means de-indexing, never destroying. - "Said, not inferred": a fact is kept only if it is grounded in what you actually wrote. - An out-of-model validator blocks secrets, hidden instructions and stale values. - Consolidation ("sleep"): deduplication, abstraction, pruning, compiled profile. - GDPR: full export, AES-256-GCM encrypted backup, erasure by key destruction. - Benchmarked on LoCoMo (1,536 questions): beats a BM25 baseline up to k = 20, 97 ms per recall, no GPU. THE PRO VERSION ($55 / 49 €, one-time payment, per seat) - One-click Windows installer for Claude Desktop, Cowork and Claude Code - One year of updates - Email support, reply within two business days - Internal use with no obligation to publish your code The source code is free (AGPL-3.0): https://github.com/GotoRyugen/memoryaicm To embed memoryaicm in a product or service without publishing your code: Enterprise license, quote on request (see the website). https://gotoryugen.github.io/memoryaicm/