MemoryAtlas

A local-first, Markdown-native memory runtime and Python library that gives agents one portable memory layer across coding assistants, apps, devices, and workflows. Stores conversations, files, and agent trajectories as canonical .md files (readable, editable, diffable, Git-versioned) and syncs local SQLite + LanceDB indexes for fast retrieval and self-evolving reuse. Exposes a local REST API server (OpenAPI-documented) for agent integration; optional multimodal ingest extra handles images, PDFs, audio, and office documents.

Storage
Markdown is the source of truth, with separate first-class tracks for the user (episodes/profile) and the agent (cases/skills) plus a source-backed Knowledge Wiki (editable Markdown pages with taxonomy, CRUD APIs, and topic search, added v1.1.0). Local SQLite + LanceDB indexes are derived from the .md files; a cascade watcher re-syncs indexes when files are edited directly. No MongoDB, Elasticsearch, or Redis required.
Retrieval
Orthogonal retrieval scoped by user_id, agent_id, app_id, project_id, and session_id, plus topic search over the Knowledge Wiki. An offline Reflection pass (added v1.1.0) merges episode clusters and refines profiles and skills between sessions. Knowledge search requires both embedding and rerank providers to be configured; missing providers now fail explicitly rather than silently returning degraded results.
Self-host
Self-host: moderate
License
Apache-2.0
Pricing
Open-source Python library (Apache-2.0), local-first and free to self-host; the real server-backed memory flow requires two external provider API keys (LLM/multimodal via OpenRouter and embedding/rerank via DeepInfra by default; any OpenAI-compatible provider supported). Managed cloud offering at evermind.ai; API keys obtainable at console.evermind.ai (cloud pricing not published). · Free + paid
GitHub stars
11,953
Last release
2026-08-07
Last commit
2026-08-07
First catalogued
2026-06-28

Strengths

  • Local-first and user-owned: Markdown source of truth backed only by SQLite + LanceDB — no managed service or cloud vector DB required to self-host
  • Direct .md file editing supported; a cascade watcher re-syncs the indexes automatically
  • Separate user (episodes/profile) and agent (cases/skills) tracks, orthogonally scoped by user_id, agent_id, app_id, project_id, and session_id
  • Knowledge Wiki (v1.1.0): editable source-backed Markdown knowledge pages with taxonomy, CRUD APIs, and topic search
  • Offline Reflection pass (v1.1.0) merges episode clusters and refines profiles and skills between sessions
  • Optional multimodal ingest (everos[multimodal]) handles images, PDFs, audio, HTML, email, and office documents via LibreOffice

Watch out

  • The real server-backed memory flow requires two external provider API keys (LLM/multimodal + embedding/rerank); the fully-local claim applies to storage only, not to extraction or semantic search
  • Knowledge search fails explicitly (HTTP error) if embedding and rerank providers are not configured — confirmed v1.1.0 compatibility note
  • Community MCP (evermemos-mcp) connects to EverMemOS Cloud, not the local OSS server — a cloud API key is required for that integration path, not self-hosting
  • Official Claude Code plugin (evermem-claude-code) is marked legacy in the EverOS README; it connects to EverMem Cloud rather than the EverOS 1.0.0 OSS API
  • Reflection and self-evolving claims are vendor-described; no independent benchmark evaluation found
  • Cloud pricing at evermind.ai not published; console at console.evermind.ai

Best for

  • Makers wanting a portable, local-first, Git-versioned Markdown memory layer shared across multiple agents, apps, and AI coding assistants
  • Agents that need user-profile memory and agent-skill memory as separate, orthogonally-scoped tracks in one runtime

How it integrates

Benchmark results

No sourced results yet.

Sources

Last verified 2026-08-01 · updated by refresh-framework-cards