Mnemosyne OS
Local memory layer that lets any AI model recall your files across sessions.
Mnemosyne OS watches a folder or vault on your machine and builds a persistent memory graph from it, so any connected model — local GGUF or cloud APIs like Claude, GPT, or Gemini — can answer a question and name the file it came from. It runs with no account and no mandatory server; you can plug in your own API key, use metered credits, or stay fully offline. On its own LongMemEval-M benchmark it measures 77.1% (37 of 48) under a strict judge it built after an earlier 72.9% figure, reproduced across two runs and confirmed on a held-out set.
What holds up
- +No account needed — the download itself is the signup, and everything runs locally
- +Publishes its own benchmark protocol and reproduces the score across two runs, not one quoted number
- +Works with your own API key, metered credits, or fully local GGUF models — no forced cloud dependency
Mind the limits
- −Its 77.1% benchmark is self-administered on 48 questions — no independent party has reproduced it
- −Encryption at rest isn't automatic — you have to arm it yourself, or an unarmed vault stays unprotected
Featured here? Take the badge
Put it on your site — it links back to this review. Free for every listed product, always.
<a href="https://stillworks.dev/products/p/mnemosyne-os/"><img src="https://stillworks.dev/badge/mnemosyne-os.svg" alt="Picked by StillWorks" width="250" height="54"></a>