Capturas de pantalla del proyecto
Resumen
Mem0 ('mem-zero') enhances AI assistants and agents with an intelligent memory layer that remembers user preferences, adapts to individual needs, and continuously learns over time. The project introduced a new memory algorithm (April 2026) that uses single-pass ADD-only extraction, entity linking, multi-signal retrieval (semantic, BM25 keyword, and entity matching), and temporal reasoning. Benchmark improvements reported on the managed platform include 92.5 on LoCoMo, 94.4 on LongMemEval, and 64.1 on BEAM (1M tokens). The open-source SDK may yield directionally similar but not identical results, as managed platform benchmarks include proprietary optimizations. The repository is written primarily in Python, is not archived, and is actively maintained.
Características principales
- Multi-level memory: User, Session, and Agent state retention with adaptive personalization
- Single-pass ADD-only extraction with no UPDATE/DELETE overwrites
- Entity linking across memories for retrieval boosting
- Multi-signal retrieval fusing semantic, BM25 keyword, and entity matching
- Temporal reasoning for time-aware retrieval
- Cross-platform SDKs (Python and npm) and a CLI
- Self-hosted server with auth and dashboard, or managed cloud platform
- Agent skills for AI coding assistants (Claude Code, Codex, Cursor, Windsurf, OpenCode, OpenClaw)
- Integrations with Langgraph, CrewAI, and a browser extension for ChatGPT, Perplexity, and Claude
Requisitos, instalación y guía rápida
Uso
Compatibilidad de modelos y casos de uso
Defaults to OpenAI gpt-5-mini as the LLM and text-embedding-3-small as the embedding model. Supports a variety of LLMs and embedders; configuration details are available in the repository's supported LLMs and supported embeddings documentation.
Notas sobre la licencia y los riesgos
Licensed under Apache-2.0. See the LICENSE file in the repository for details.
Editorial verification 2026-08-09: repository URL, owner, description, license and repository statistics were reviewed. License metadata: Apache-2.0. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.
Lanzamiento y mantenimiento
New memory algorithm introduced in April 2026 with single-pass ADD-only extraction, agent-generated facts as first-class data, entity linking, multi-signal retrieval, and temporal reasoning. A migration guide from OSS v2 to v3 is available. The evaluation framework is open-sourced for reproducibility.