LightMem vs MemMachine
ICLR 2026 memory framework for LLMs/agents: LLMLingua pre-compression, topic segmentation and offline memory updates — leading LoCoMo/LongMemEval results at lower token cost. — versus — Long-term memory layer for AI agents — episodic (graph), profile (SQL) and working memory behind Python/TS SDKs, REST and MCP; ships LangChain, LangGraph, CrewAI and LlamaIndex integrations.
Both are agent memory layers with MCP servers; MemMachine ships product-shaped SDKs and framework integrations, LightMem ships a compression-first pipeline with published benchmark wins.
| LightMem | MemMachine | |
|---|---|---|
| Stars | 1.0k | 3.3k |
| Forks | 92 | 196 |
| Language | Python | Python |
| License | MIT | Apache-2.0 |
| Last activity | 3 days ago | 3 days ago |
| Topics | memory | memory |
| Curated connections | 2 | 7 |
LightMem — the curator's take
Research-grade memory with receipts: reproduction scripts for LoCoMo/LongMemEval plus a baseline harness that benchmarks Mem0, A-MEM and LangMem side by side — useful even if you adopt none of them. The compression-first pipeline (LLMLingua before storage) is the differentiating idea. Expect paper-adjacent ergonomics: manual model downloads and config dicts, not a polished product.
MemMachine — the curator's take
Pick it when memory is a product requirement, not a cache: separating episodic (graph) from profile (SQL) from working memory maps to how assistants actually personalize, and the documented LangGraph/CrewAI/LlamaIndex integrations mean you don't write the glue. NOT worth the footprint for a single-user tool — it wants a server plus Neo4j and SQL; a vector store or a JSON file gets a prototype further. Watch the open-core boundary: the managed platform is the business model.