MemoryAtlas

A self-hosted cognitive memory engine that stores memories across five sectors (episodic, semantic, procedural, emotional, reflective) rather than flat chunks. It layers a temporal knowledge graph with valid_from/valid_to facts and point-in-time queries, plus an associative 'waypoint' graph of traversable links, on top of embeddings — with a decay and reinforcement engine in place of TTLs. Recall returns an explainable trace showing which nodes fired. Ships as a Python or Node SDK against local SQLite, or as a multi-user server with REST API, dashboard, and MCP.

Storage
SQLite by default (local-first, zero external infra); Postgres in server mode. Temporal knowledge-graph facts are stored alongside vector embeddings, with pluggable embedders (OpenAI, Gemini, Ollama, AWS, or a synthetic fallback).
Retrieval
Composite scoring over salience, recency, and coactivation across the multi-sector store; graph traversal of associative waypoint links for multi-hop recall; and point-in-time temporal fact queries via /api/temporal/* that resolve what was true at a given date.
Self-host
Self-host: trivial
License
Apache-2.0
Pricing
Free and open source, self-hosted only — no vendor-hosted tier found. Cost is your own infra; the project's own comparison estimates roughly $5–8/month of storage at 100k memories. · Free / OSS
GitHub stars
4,421
Last release
2025-12-12
Last commit
2026-07-28
First catalogued
2026-07-25

Strengths

  • Genuinely differentiated retrieval — temporal knowledge graph plus an associative waypoint graph and a decay/reinforcement engine, not bare vector similarity
  • True local-first default (SQLite, zero cloud dependency) with a documented upgrade path to a multi-user server
  • Unusually broad shipped surface for a free project: Python and Node SDKs, REST API, MCP, CLI, and an official VS Code extension
  • Built-in migration tooling from Mem0, Zep, and Supermemory lowers switching cost
  • Point-in-time fact queries answer 'what did we believe on this date', which flat vector stores cannot

Watch out

  • The README's first line still reads '🚧 This project is currently being rewritten. Expect breaking changes and potential bugs' — verified live 2026-07-25, and the rewrite branch exists; carded as needs-review for that reason
  • Name collision: this is the independent CaviraOSS project, distinct from mem0's separately-branded 'OpenMemory MCP' that lives inside the mem0 repo — easy to conflate when searching
  • The marketing site's footer claims MIT while the repo LICENSE is Apache-2.0; the repo is authoritative, and the discrepancy is a mild reliability flag for other site claims
  • Release hygiene is loose — a prerelease-flagged v1.3.0 sits ahead of GitHub's 'latest' v1.2.3 while PyPI and npm shipped further versions with no matching tag
  • Small maintainer footprint (a single primary author on releases and commits) for a project mid-rewrite — real bus-factor risk

Best for

  • Self-hosted local-first agent memory that needs time-aware facts and point-in-time truth
  • Teams wanting a free, no-lock-in alternative to Mem0, Zep, or Supermemory with a documented migration path
  • VS Code, Claude Code, and Cursor users wanting an MCP-native memory tool

How it integrates

Benchmark results

No sourced results yet.

Sources

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