EverOS

EverOS is a Python library and local-first memory runtime that gives agents one portable memory layer across coding assistants, apps, devices, and workflows. It stores conversations, files, and agent trajectories as readable Markdown, which serves as the source of truth, and keeps local SQLite and LanceDB indexes in sync for retrieval. The Markdown files can be edited directly and versioned in Git, and a watcher re-indexes the changes.

Memory is split into a user track (episodes and profile) and an agent track (cases and skills), and retrieval can be scoped by user, agent, app, project, and session. Between sessions, an offline reflection step merges clusters of episodes and refines profiles and skills, turning agent trajectories into reusable skill memories. A Knowledge Wiki holds editable, source-backed Markdown pages with a taxonomy, CRUD APIs, and topic search.

EverOS runs as a local server with an HTTP API for adding, flushing, and searching memory. Features scale with the providers configured: a single LLM key enables the core memory flow and keyword search, an embedding provider adds hybrid search, reflection, and skill extraction, a reranker adds agentic search and the Knowledge Wiki, and a multimodal model adds ingestion of images, PDFs, audio, and office documents. A standalone demo runs without any key.