MemoryAtlas

An OS-inspired memory layer for personalized AI agents that organizes user memory into short-, mid-, and long-term tiers and migrates entries between them. Published as an EMNLP 2025 Oral.

Storage
Segmented-page storage with a heat-based updating mechanism that promotes/demotes memory across short / mid / long-term tiers.
Retrieval
Retrieves relevant user and assistant memory across the tiered store to condition responses; details per the paper and repo.
Self-host
Self-host: moderate
License
Apache-2.0
Pricing
Open source (Apache-2.0), free to self-host · Free / OSS
GitHub stars
1,487
Last release
2025-07-18
Last commit
2026-04-28
First catalogued
2026-06-28

Strengths

  • Peer-reviewed (EMNLP 2025 Oral)
  • OS-style tiered memory with heat-based promotion/demotion
  • Permissive Apache-2.0 license

Watch out

  • Name collides with the separate MemTensor MemOS project — they are different systems
  • Last repo push ~April 2026; maintenance cadence is research-project-paced

Best for

  • Personalized conversational agents needing tiered long-term user memory

Benchmark results

No sourced results yet.

Sources

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