Single-pass ADD-only extraction: each add() fires one LLM call (default gpt-5-mini) that accumulates facts — no UPDATE or DELETE on ingestion. Entity linking via spaCy (installed as `mem0ai[nlp]`) extracts named entities from every stored memory and writes them to a parallel vector sub-collection (`{collection}_entities`); retrieval folds entity-match scores into the ranking alongside dense and keyword signals. External graph DB support (Neo4j, Memgraph, Kuzu, AGE, Neptune) was removed in 2026; graph-style association is now vector-native. Three memory scopes: user (cross-session persistence), session (ephemeral per run), agent (runtime state). Two deployment paths: OSS library (BYO LLM + BYO vector store) or fully managed cloud (Qdrant-backed, auto-scaled). 23+ official integrations span LangChain, LangGraph, CrewAI, AutoGen, Vercel AI SDK, OpenAI Agents SDK, Google ADK, AWS Bedrock, LiveKit, ElevenLabs, Dify, Flowise, and more.
Deep diveMem0 deep dive: how the token-efficient memory layer works in productionRead →- Storage
- Cloud: Mem0-hosted Qdrant vectors with automatic scaling, reranking, and no infrastructure work. OSS: pluggable vector store (Qdrant, Chroma, Pinecone, Weaviate, PGVector, Valkey, and others) plus an auto-created entities sub-collection (`{collection}_entities`) for entity-linked memories. Requires BYO LLM and BYO embedder in OSS mode (default text-embedding-3-small; Qwen 600M+ recommended for hybrid search). No external graph database required — entity storage is vector-native since Jun 2026.
- Retrieval
- Multi-signal fusion: dense semantic similarity, BM25 keyword matching, and entity matching (spaCy-extracted entities against the `{collection}_entities` sub-collection) are scored in parallel and fused at query time. Metadata filter support. No LLM call at retrieval time — retrieval is embed-then-search only. Reported p50 latency 0.88–1.09s per call. Token consumption under 7,000 per retrieval across all four benchmarks (LoCoMo 6,956; LongMemEval 6,787; BEAM 1M 6,719; BEAM 10M 6,914), vs 25,000+ for full-context approaches (self-reported).
- Self-host
- Self-host: moderate
- License
- Apache-2.0
- Pricing
- Free/Hobby (10k adds, 1k retrievals/mo, 1 project); Starter $19/mo (50k adds, 5k retrievals, 1 project); Pro $249/mo (500k adds, 50k retrievals, unlimited projects, private Slack, advanced analytics, graph memory); Enterprise custom (unlimited adds/retrievals/projects, on-prem, audit logs). The former Growth $79/mo tier has been discontinued — Starter now jumps directly to Pro. · Freemium
- GitHub stars
- 62,937
- Last release
- 2026-08-05
- Last commit
- 2026-08-07
- First catalogued
- 2026-06-28
Strengths
- ~30s to wire up: one API key and two lines of code via the cloud SDK; OSS mode requires BYO vector store but still ships a working config in minutes
- 23+ official framework integrations: LangChain, LangGraph, CrewAI, AutoGen, Vercel AI SDK, OpenAI Agents SDK, Google ADK, AWS Bedrock, LiveKit, ElevenLabs, Dify, Flowise, and more
- Token-efficient retrieval: under 7,000 tokens per retrieval call vs 25,000+ for full-context; extraction is one LLM call per add()
- Generous free tier: 10k memory adds and 1k retrievals per month at no cost with no credit card required
- Entity-linked recall without an external graph DB: spaCy NLP extracts entities into a parallel vector sub-collection and boosts precision on named-entity queries
Watch out
- ADD-only extraction accumulates facts without deduplication: contradictory or stale memories can pile up in long-running deployments; there is no automatic conflict resolution or overwrite on ingestion
- Each add() fires one LLM call (gpt-5-mini by default): extraction API costs accrue at production scale on top of embedding costs — budget for LLM usage on every memory write, not just on reads
- All published benchmark scores (LoCoMo 92.5, LongMemEval 94.4, BEAM 1M 64.1) are self-reported by Mem0; independently reproduced results in this catalog are lower; no neutral same-config multi-framework comparison exists yet
- OSS/platform feature gap: entity linking requires `mem0ai[nlp]` + spaCy in OSS; cloud platform adds automatic reranking, webhooks, audit logs, custom categories, and SOC 2/HIPAA controls not present in the OSS tier
Best for
- Fastest drop-in memory with the largest integration ecosystem when time-to-working-agent is the constraint
- Token-cost-sensitive production agents needing sub-7k retrieval tokens and single-pass extraction without full-context overhead
- SOC 2 / HIPAA workloads and AWS Agent SDK users needing managed, auditable memory with no infrastructure to run
How it integrates
Benchmark results
| Benchmark | Value | Backbone | Trust | Source |
|---|---|---|---|---|
| longmemeval | 94.4 accuracy | — | Self-reported | Mem0 ↗ |
| longmemeval | 49 accuracy | GPT-4o | Independent | Zep (competitor harness) ↗ |
| locomo | 92.5 accuracy | — | Self-reported | Mem0 ↗ |
| beam-1m | 64.1 accuracy | — | Self-reported | Mem0 ↗ |
| beam-10m | 48.6 accuracy | — | Self-reported | Mem0 ↗ |
| locomo | 66.88 accuracy | — | Independent | Hindsight/Vectorize (competitor re-run) ↗ |
| longmemeval | 62.6 accuracy | gpt-4o-mini | Independent | LiCoMemory (Huang et al., HKUST et al.) — competitor re-run ↗ |
| locomo | 54.68 accuracy | gpt-4o-mini | Independent | LiCoMemory (Huang et al., HKUST et al.) — competitor re-run ↗ |
Sources
- Pricing page re-checked — unchanged from the 2026-07-23 capture: Hobby free (10k adds, 1k retrievals, 1 project), Starter $19/mo (50k adds, 5k retrievals, 1 project), Pro $249/mo (500k adds, 50k retrievals, unlimited projects), Enterprise custom. Still no Growth tier; page also notes usage-based pricing is available for teams that don't map to a fixed tier (vendor)
- Monthly card re-verification — GitHub API repo metadata: 62,255 stars, license Apache-2.0 confirmed, archived=false, last commit on the default branch 2026-08-01 (third-party)
- https://mem0.ai/ (vendor)
- https://github.com/mem0ai/mem0 (vendor)
- Mem0 research page — single-pass ADD-only extraction; multi-signal retrieval (semantic + BM25 + entity); token counts per benchmark (<7k); self-reported scores LoCoMo 92.5, LongMemEval 94.4, BEAM 1M 64.1, BEAM 10M 48.6; p50 latency 0.88–1.09s (vendor)
- Mem0 pricing page — Free (10k adds, 1k retrievals), Starter $19/mo, Growth $79/mo, Pro $249/mo, Enterprise custom (vendor)
- Mem0 security page — SOC 2 Type I certified, HIPAA-ready, SOC 2 Type II audit in progress (as of Jun 2026) (vendor)
- Mem0 graph memory docs — entity linking now vector-native via spaCy; entities stored in {collection}_entities sub-collection; external graph DB (Neo4j, Memgraph, Kuzu, AGE, Neptune) removed Jun 2026 (vendor)
- Mem0 platform MCP server docs — cloud-hosted at https://mcp.mem0.ai/mcp; remote-http transport; npx mcp-add install command (vendor)
- Introducing OpenMemory MCP — local-first Docker deployment; SSE/HTTP on localhost:8765; full data ownership (vendor)
- Mem0 REST API reference — Memory CRUD, search, events, entities, webhooks; token-based auth (vendor)
- mem0ai on PyPI — Python SDK v2.0.10 (Jun 2026); mem0ai[nlp] extra for spaCy entity linking (vendor)
- Mem0 Node SDK quickstart — TypeScript/JavaScript SDK v3.0.12 (Jun 2026); MiniMax and LiteLLM providers added; platform + OSS modes (vendor)
- Mem0 integrations page — 23+ integrations: LangChain, LlamaIndex, LangGraph, AutoGen, CrewAI, Camel AI, Vercel AI SDK, OpenAI Agents SDK, Google ADK, Mastra, Agno, AgentOps, Respan, LiveKit, ElevenLabs, Pipecat, Dify, Flowise, ChatDev, Raycast, AWS Bedrock, MCP (vendor)
- Release cli-v0.2.10 (2026-07-13) titled 'Mem0 Python CLI (v0.2.10)' — confirms an official Mem0 Python CLI; sibling tag cli-node-v0.2.11 (2026-07-13) is the Node CLI track (vendor)
- GitHub releases — latest tags v2.0.12 (Python core) and ts-v3.1.0 (TypeScript SDK), both 2026-07-13 (third-party)
- GitHub API repo metadata — 60,925 stars, pushed 2026-07-15T08:57Z (third-party)
- GitHub API re-check — license still Apache-2.0, not archived, pushed 2026-07-23T16:24Z (third-party)
- Mem0 pricing page re-check — Growth $79/mo tier no longer listed; current tiers are Hobby (free), Starter $19/mo, Pro $249/mo, Enterprise custom (vendor)
- Mem0 OpenCode plugin docs — `opencode plugin @mem0/opencode-plugin`; 9 memory tools, lifecycle hooks, 9 skills incl. /mem0-scope (vendor)
- Mem0 OpenCode plugin release v0.2.2 (2026-07-22) — confirms the plugin is actively maintained; earliest found tag opencode-v0.1.3 (2026-06-13) (third-party)
Last verified 2026-08-01 · updated by refresh-framework-cards