A universal, persistent memory layer for AI assistants — designed, built, and open-sourced solo.
AI assistants forget everything between sessions. MnemOS solves that by extracting facts from conversations via LLM, embedding them with bge-m3, and injecting relevant context before each new message — invisibly, across ChatGPT and Claude. Memories are stored per-user rather than per-platform, so context follows the person across tools instead of resetting every session.
1 · ConnectSetup starts by looking for the local MnemOS server on localhost:8765 — the FastAPI process that stores and retrieves memories. Nothing leaves the machine at this stage.
1 · ConnectServer found and confirmed running. This local-first handshake is what lets MnemOS work without any cloud dependency for storage.
2 · EngineThe user picks how conversations get turned into memories: Gemini Flash for speed over the cloud, or Ollama for a fully local, private extraction pipeline.
2 · EngineWith Ollama selected, models are listed by size and quality with live download status — so setup adapts to whatever hardware it's running on, from a 0.4GB model to 9GB.
3 · DoneSetup complete in under two minutes. From here, MnemOS silently extracts and injects memory across ChatGPT, Claude, and Gemini — no further action needed from the user.
02 / Extension
The extension popup surfaces the memory count, the most recently extracted facts with their inferred type (skill, goal, preference, workflow), and lets the user switch extraction models without leaving the browser.
03 / Dashboard
OverviewThe dashboard's overview gives memory stats at a glance — total memories, sessions stored, and the semantic-vs-episodic split — broken down per connected app.
MemoriesEvery extracted fact is listed here, searchable and filterable by app and memory type, each with a confidence score from the extraction model.
SessionsRaw conversation history behind every extraction — auto-refreshing and exportable per session, so any stored memory can be traced back to its source conversation.
SettingsServer connection and extraction engine configuration, including one-click local model downloads via Ollama — the same controls exposed during onboarding, always reachable later.
