MemoryAtlas

A self-hostable 'memory operating system' that packages long-term memory into MemCube units and manages their lifecycle (store / retrieve / update / schedule) outside the model.

Storage
Pluggable backends; the local plugin uses persistent SQLite. Supports text, images, tool traces, and personas as memory.
Retrieval
Hybrid retrieval combining full-text search (FTS5) and vector similarity, with task summarization and cross-task skill reuse.
Self-host
Self-host: moderate
License
Apache-2.0
Pricing
Open source (Apache-2.0), free to self-host · Free / OSS
GitHub stars
10,018
Last release
2026-06-18
Last commit
2026-06-18
First catalogued
2026-06-28

Strengths

  • Active, high-traction project (10k+ stars)
  • Hybrid full-text + vector retrieval
  • Multi-modal memory (text, images, tool traces, personas)
  • Vendor reports ~35% token savings

Watch out

  • 'Memory OS' framing overlaps confusingly with the separate BAI-LAB MemoryOS project — they are different systems
  • A hosted/cloud offering (OpenMem) may exist; pricing for any managed tier not confirmed

Best for

  • Teams wanting a self-hosted memory layer with hybrid retrieval and skill reuse

Benchmark results

BenchmarkValueBackboneTrustSource
locomo75.8 accuracyGPT-4o-miniSelf-reportedMemOS (MemTensor et al.)
longmemeval77.8 accuracyGPT-4o-miniSelf-reportedMemOS (MemTensor et al.)

Sources

Last verified 2026-06-28 · updated by manual-refresh