Open sourcePython · ≈60k stars · MIT
MemPalace
Local-first AI memory: verbatim storage with scoped semantic retrieval, exposed to agents via MCP and skills.
MemPalace is an MIT-licensed, local-first memory system for AI agents. It stores conversation history verbatim, without summarising or extracting, and retrieves it with semantic search over a structured index. People and projects become 'wings', topics become 'rooms' and original content sits in 'drawers'. ChromaDB is the default store, with a pluggable backend interface and an optional native Rust engine. Agent skills install it and configure MCP for coding agents such as Claude Code. The README reports 96.6% R@5 on LongMemEval with zero API calls and commits the benchmark results to the repo.
Pros & cons
Each point cites its sources and says what kind of evidence backs it. A second reviewer tried to refute every claim before it was kept.
Pros
- DocsLocal-first memory that stores conversations verbatim (ChromaDB by default) and needs no API calls for retrieval.[1]
- User reportsA user is adopting it for locally managed, portable memory across small agents, despite the criticism of its benchmarks.[1]
- MetadataMIT licensed, about 60k stars, and active (v3.11.0, Oct 2026), with a pluggable retrieval backend.[1]
Cons
- User reportsOpen issue #27 says the headline 96.6% LongMemEval R@5 is raw ChromaDB retrieval without the palace structure, and that AAAK mode drops it to 84.2%.[1][2]
- User reportsA HN commenter summarizing the GitHub issues says scores dropped sharply when third parties ran end-to-end QA instead of retrieval metrics.[1]
- Independent testIn a cold-start comparison of six tools (by the author of a rival tool), it took about 3 hours to index, 2.6 s per query, and scored 14/100 hit@1.[1]
- User reportsA look-alike domain promoted on HN was not affiliated with the project and served malware.[1]
- Best for
- Local, verbatim conversation recall for personal agents where you control the data.
- Watch out for
- Ignore the headline benchmarks until you test end-to-end QA on your own data, and install only from the official repo.
Assessed Oct 29, 2026 · adversarially reviewedSuggest a correction
People
Engineer
Co-creator, MemPalace
Context engineering & memory
Creator/Architect of MemPalace
Context engineering & memory
Neighbours on the workbench
JavaScript · ≈292k stars · MIT
affaan-m/ECC
Agent harness performance optimization system: skills, instincts, memory and security for coding agents.
Maintained by Affaan Mustafa
TypeScript · ≈258k stars · MIT
NousResearch/hermes-agent
The agent that grows with you.
Maintained by Nous Research
Python · ≈138k stars · Apache-2.0
Graphify-Labs/graphify
A /graphify skill that turns a codebase, docs and schemas into a queryable knowledge graph.
Maintained by Graphify-Labs
Connections
Relations from the Harness Hub knowledge graph, as charted in the State of Harness reports.
Maintained by · 1
- MemPalaceOrganisation
Built on · 2
- Claude CodeProject
- Model Context ProtocolProject
Sources
Verified Oct 29, 2026