A lifelong memory stack for LLM agents built on 'semantically lossless compression' — store dense, high-information memory so an agent recalls more while spending far fewer tokens. Ships as one `simplemem` Python package that auto-routes across three pillars: SimpleMem (text efficiency core), Omni-SimpleMem (multimodal: text/image/audio/video), and EvolveMem (self-evolving retrieval). Also offered as a cloud-hosted and self-hostable MCP server. Backed by arXiv papers (2601.02553, 2604.01007, 2605.13941).
- Storage
- A three-stage write pipeline: Semantic Structured Compression distills interactions into self-contained atomic memory units (resolved coreferences, absolute timestamps) indexed through multiple complementary views; Online Semantic Synthesis merges related context within a session into abstract representations, removing redundancy at write time. Vectors persist in LanceDB; the multimodal path adds entropy-driven selective ingestion and knowledge-graph augmentation.
- Retrieval
- Intent-Aware Retrieval Planning infers the search intent behind a query to decide what to retrieve and assemble a compact context; the multimodal path uses hybrid FAISS + BM25 with pyramid token-budget expansion. EvolveMem runs a closed-loop Evaluate→Diagnose→Propose→Guard cycle that mutates retrieval config (top_k, fusion mode, verification, reflection) against a dev set.
- Self-host
- Self-host: moderate
- License
- MIT
- Pricing
- Open-source MIT Python package, free to self-host (needs an OpenAI-compatible LLM + embedding endpoint). A cloud-hosted MCP server exists at mcp.simplemem.cloud; its pricing was not captured. · Free + paid
- GitHub stars
- 3,566
- Last release
- 2026-05-21
- Last commit
- 2026-06-23
- First catalogued
- 2026-06-28
Strengths
- Compression-first design targets large token savings — semantically lossless memory units retrieved by inferred intent
- Three integrated pillars in one package: text core, multimodal (image/audio/video) Omni backend, and EvolveMem self-evolving retrieval
- Research-backed (three arXiv papers) with reproducible LoCoMo/MemBench/Mem-Gallery harnesses; MIT-licensed
Watch out
- README headlines many self-reported benchmark gains (LoCoMo F1, ~30x token reduction, Mem-Gallery SOTA, 'outperforms Claude-Mem by 64%') — record via harvest-benchmarks with selfReported:true, not here
- Requires an OpenAI-compatible LLM + embedding endpoint for both memory construction and retrieval; no fully-offline default
- A hosted MCP service (mcp.simplemem.cloud) exists but its pricing/relationship to the OSS core was not captured — priceTier and pricing need reviewer confirmation
Best for
- Token-budget-constrained agents needing dense lifelong memory with intent-aware retrieval, optionally across modalities
Benchmark results
No sourced results yet.
Sources
- SimpleMem README (vendor)
- GitHub API repo metadata (stars, MIT license, v0.3.0 release) (third-party)
- SimpleMem: Efficient Lifelong Memory for LLM Agents (arXiv) (paper)
Last verified 2026-06-28 · updated by discover-frameworks