The gist

EverOS is an open-source Python library and local-first memory runtime for AI agents, developed by EverMind-AI. It provides a portable memory layer that solves the challenge of persistent agent memory across applications and sessions. It stores conversations, files, and agent trajectories as human-readable Markdown files, using local SQLite and LanceDB indexes for fast, self-evolving retrieval without relying on external managed databases. This approach makes agent memory directly editable, diffable, and versionable with Git.

What it does

Store agent memory as human-readable, versionable Markdown files.
Index memory locally using SQLite and LanceDB for fast, orthogonal retrieval.
Ingest and process multimodal files including images, PDFs, audio, and office documents.
Evolve memory offline by merging conversation clusters and refining agent skills between sessions.
Provide an OpenAI-compatible API for adding and searching memory.
Visualize the memory lifecycle from ingest to recall with a built-in terminal UI.

How it works

EverOS runs as a local server with API endpoints. Developers integrate it as a Python library or send data (conversations, files) to its API. EverOS saves this data as structured Markdown files and updates local SQLite and LanceDB indexes for retrieval. The system is open-source and self-hosted, but relies on external API keys from providers like OpenRouter and DeepInfra for LLM, embedding, and reranking capabilities to process the memory content.

Best for

Developers building AI agents who need a local-first, persistent long-term memory solution. It is ideal for projects that require a Git-friendly, file-based source of truth for agent memory without dependencies on managed vector databases.

Watch out for

Processing office documents (.doc, .ppt, .xls) requires a system-level installation of LibreOffice to function.