What is RememberOS?
RememberOS is privacy-first long-term memory for AI agents: store what your users and agents learn, search it semantically, and let memory evolve — facts get updated, extended, and forgotten over time, like real memory.
How it works (at a glance)#
- You send RememberOS text, files, or whole conversations.
- RememberOS extracts and indexes them — embeddings for semantic search, plus a graph of
typed facts (
fact/preference/episode) that supersede and extend each other. - At query time you fetch only the most relevant context and hand it to your model.
The pieces#
| Layer | What it does |
|---|---|
| Memory API | add stores verbatim; remember extracts atomic typed facts, links updates/extends, sets expiry — the graph-memory brain. |
| Search | Hybrid retrieval (vector + full-text fused with RRF), scopable by collection and container tag. |
| Profiles | An LLM-built summary of a collection's current facts — ready-made system-prompt context. |
| Connectors | File drop (browser or presigned direct-to-S3), SharePoint importer, a dlt destination for any data pipeline, MCP for agents, email-in. |
| Self-hosting & BYO | Run it on your own box; bring your own S3, Postgres, and extraction LLM — your data never has to leave your infrastructure. |
Why RememberOS#
- Privacy-first: EU-built, self-hostable, on-box embeddings, cookieless analytics, strict per-tenant isolation enforced by Postgres row-level security.
- Memory, not just storage: new facts supersede old ones (
is_latest), time-bound facts expire, profiles summarize what's currently true. - Zero-dependency SDKs: Python and TypeScript clients with no transitive deps.
Next steps#
QuickstartKey to first memory in five lines — Python, JS, or curl.
Install with AIOne prompt that wires RememberOS into Claude Code or Cursor.
Migrate to RememberOSComing from another memory layer? Map your calls and bulk-import.
PlaygroundTry add / search / remember live in the browser.
llms.txtThe agent-readable API summary at the site root.