EverMind-AI/EverOS

One portable memory layer for every AI agent: local-first, Markdown-native, user-owned, and self-evolving across apps, tools, and workflows.

What it solves

EverOS provides a portable, local-first memory layer for AI agents and developers. It solves the problem of fragmented agent memory by creating a unified system where conversations, files, and agent trajectories are stored in a way that is human-readable, editable, and persistent across different apps, devices, and workflows.

How it works

EverOS uses a three-part local stack consisting of Markdown files, SQLite, and LanceDB. Markdown serves as the canonical source of truth, allowing users to edit memories directly as files which then sync to the indexes. The system separates user tracks (episodes and profiles) from agent tracks (cases and skills). It supports orthogonal retrieval based on IDs (user, agent, app, project, and session) and includes a reflection process that merges memory clusters and refines profiles offline between sessions.

Who it’s for

It is designed for AI agent makers and developers building coding assistants, personal AI companions, or any agentic system that requires long-term, evolving memory.

Highlights

  • Markdown Source of Truth: Memories are stored as .md files that are Git-versioned and directly editable.
  • Local-First Architecture: Operates without needing managed services like MongoDB or Redis, relying on a local SQLite and LanceDB stack.
  • Huma-Centric Memory: Distinguishes between user-specific profiles and agent-specific skills.
  • Multimodal Support: Can ingest images, PDFs, audio, and Office documents (via LibreOffice) into the memory stream.
  • Self-Evolving: Features offline reflection to consolidate and refine memories over time.