GitHub radar
Mem0: Persistent Memory Layer for AI Agents
Open-source memory infrastructure that gives AI agents persistent recall — remembering user preferences, past tasks, and context across every session.
Mem0 is an open-source intelligent memory layer for AI agents and assistants, built by a YC S24 company (mem0ai). It makes agents remember user preferences, past tasks, and context across sessions — no more explaining the same things over and over. In April 2026, the team shipped a new memory algorithm: single-pass ADD-only extraction (one LLM call, no UPDATE/DELETE chains), entity linking across memories, and multi-signal retrieval combining semantic search, BM25 keyword matching, and entity matching. On the managed platform, benchmark accuracy on LoCoMo grew from 71.4 to 92.5; on LongMemEval — from 67.8 to 94.4 (managed platform includes proprietary optimizations; open-source SDK is directionally similar). Requires an LLM to function — gpt-5-mini from OpenAI by default, but any compatible API works. Three deployment options: pip/npm library, self-hosted server via Docker Compose, or cloud platform at app.mem0.ai. Plugs into Claude Code, Codex, Cursor, Windsurf via `npx skills add`.
Why a vibe-coder should care
Your agent stops asking the same questions and starts working like a colleague who knows you. Self-hosted option keeps all data on your machine. Open source, MIT license.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Deploy Mem0 from here: https://github.com/mem0ai/mem0 — use the Docker Compose option from the README, then install the skill with `npx skills add https://github.com/mem0ai/mem0 --skill mem0` and show me how the agent remembers context between sessions. Ask me for an API key if needed.
Self-hosted option runs via Docker on a regular laptop. Needs an LLM API key (OpenAI by default).
Open on GitHub▌ More finds