MemoryAtlas

A neurobiologically inspired long-term memory framework that builds a knowledge graph over documents and retrieves with Personalized PageRank, enabling continual integration of knowledge. HippoRAG 2 improves multi-hop associativity and sense-making.

Storage
Open knowledge graph built from extracted entities/relations over a document corpus, plus passage embeddings. Pluggable vector-database backends (Qdrant or ChromaDB) were added June 2026 (PR #184), beyond the default in-memory store.
Retrieval
Personalized PageRank over the knowledge graph for single-step multi-hop retrieval. HippoRAG 2 adds a 'Recognition Memory' step (Sec. 3.4) that uses an LLM to filter candidate triples matched to the query before they become PPR seed nodes — so retrieval is graph traversal with one online LLM-filtering call in the loop, not purely deterministic.
Self-host
Self-host: moderate
License
MIT
Pricing
Open source (MIT), free to self-host · Free / OSS
GitHub stars
3,933
Last release
2025-02-27
Last commit
2026-07-29
First catalogued
2026-06-28

Strengths

  • Peer-reviewed (NeurIPS'24); HippoRAG 2 follow-up paper (Feb 2025)
  • Single-step multi-hop retrieval via Personalized PageRank
  • More efficient offline indexing than GraphRAG/RAPTOR/LightRAG per the authors
  • Actively maintained — community feature merges through June 2026 (e.g. pluggable Qdrant/ChromaDB backends)

Watch out

  • Primarily a research artifact (no managed/hosted offering); PyPI package is still on a 2.0.0 alpha line
  • Targets corpus QA / continual knowledge integration more than conversational user memory

Best for

  • Multi-hop retrieval and knowledge integration over large document corpora

How it integrates

Benchmark results

No sourced results yet.

Sources

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