Open sourcePython · ≈69k stars · Apache-2.0
Mem0
The memory layer for AI apps and agents.
Mem0 extracts, stores and retrieves salient facts from conversations and documents — a managed memory layer with an open-source core that has become a default choice for agent memory.
Pros & cons
Each point cites its sources and says what kind of evidence backs it. A second reviewer tried to refute every claim before it was kept.
Pros
- User reportsOpen source and self-hostable; a commenter also calls its hosted memory API one of the more developed managed offerings.[1][2]
- User reportsUsers report wiring it easily into local stacks (for example with Qdrant) and into a personal MCP memory server for Claude Code.[1][2]
- User reportsWidely integrated: agent runtimes and backend-agnostic memory SDKs list it as a supported backend.[1][2]
- MetadataApache-2.0 with active Python and TypeScript SDKs (ts-v3.3.1, Sep 2026).[1]
Cons
- User reportsAn OpenClaw user found the extracted 'facts' store became a mess of incomplete and invalid entries, and stopped using it.[1]
- Independent testA MemBench test (4,000 cases, gpt-5-nano) found Mem0 and Zep 14-77x more expensive and 31-33% less accurate than passing the full history.[1]
- User reportsWrites trigger LLM extraction calls (2-3 per memory, according to authors of competing tools), so cost scales with volume and self-hosting still needs a model key.[1][2][3]
- User reportsResearchers in issue #2800 could not reproduce the published LOCOMO scores with the open-source library; maintainers said the hosted platform has extra features.[1]
- Best for
- Adding personalization memory (user preferences, facts) to an app quickly, hosted or self-hosted.
- Watch out for
- Extraction quality and per-write LLM cost; benchmark it against a plain long-context or summary baseline on your own data.
Assessed Oct 29, 2026 · adversarially reviewedSuggest a correction
Activity
- Last push
- · 3d ago
- Latest release
- ts-v3.3.1
· 8w ago - Last 90 days
- 20+ releases
ts-v3.3.1Mem0 Node SDK (v3.3.1)
· 8w agoBug Fixes: Config (OSS): ConfigManager no longer injects OpenAI's default baseURL and model into non-OpenAI LLM providers. The defaults now apply only to openai and openai_structured, so a DeepSeek, xAI, or other provider config without an explicit baseURL or model falls back…
v2.2.1Mem0 Python SDK (v2.2.1)
· 8w agoBug Fixes: Core: add() (Memory and AsyncMemory) no longer reports records the vector store rejected as successful ADD events. Only records that were actually inserted are written to history, entity-linked, and returned. If none of the extracted memories could be inserted, add()…
deepseek-plugin-v0.3.2Mem0 DeepSeek Plugin (v0.3.2)
· 8w agoImprovements: Search: The search_memory tool description no longer asks the agent to search proactively before answering anything that may depend on earlier context. It now asks for a search before repeating investigation or when earlier decisions, fixes, commands, or results…
pi-agent-v0.3.2Mem0 Pi Agent Plugin (v0.3.2)
· 8w agoImprovements: Search: The memory policy, mem0_memory tool description, and prompt guidelines no longer ask the agent to search before answering anything that may depend on earlier context or to run several searches per question. They now ask for a search before repeating…
openclaw-v1.2.1Mem0 OpenClaw Plugin (v1.2.1)
· 8w agoImprovements: Search: The memory_search tool description no longer asks the agent to search proactively or to run several searches for multi-part questions. It now asks for a search before repeating investigation or when earlier decisions, fixes, commands, or results may help.…
Checked against GitHub on Nov 21, 2026. Release notes are shortened to plain text.
The team behind it
San Francisco, CA · Est. 2024
The memory layer for AI apps and agents.
Mem0 (from the EmbedChain team) provides a managed + open-source memory layer that extracts, stores and retrieves user and entity memories across sessions — a standard building block for personalization.
In research (7)
- Jetriever: Shifting Agent Memory Retrieval from Similarity to UtilityVerdict: Incremental
Nov 16, 2026 · Mem0 is a published InMind baseline (3.2% and 6.4% Application, Table 6) and a MAB baseline (21.1 overall, Table 7).
- LEDGER: Auditable Agent Memory Under ChangeVerdict: Useful
Nov 16, 2026 · Evaluated directly as the memory-system baseline (OSS 1.0.11, default and coding prompt).
Nov 16, 2026 · Mem0 (Chhikara et al., 2025) is cited as a memory layer that stores and retrieves past experience, comparable to the memory-file and retrieved-memory arms.
Oct 27, 2026 · Production memory layer that selects and compresses facts by relevance. The paper shows relevance-based selection readmits on-topic decoys.
- ETAMem: Decoupling and Aligning Heterogeneous Memory for Long-Term Conversational AgentsVerdict: Incremental
Oct 27, 2026 · Mem0 is a direct baseline (57.79 on LoCoMo, 64.96 on LongMemEval-S).
Oct 23, 2026 · Mem0 is a baseline in Table 1, where it averages 24.2.
Showing the 6 newest of 7 papers.
Neighbours on the workbench
JavaScript · ≈308k stars · MIT
affaan-m/ECC
Agent harness performance optimization system: skills, instincts, memory and security for coding agents.
Maintained by Affaan Mustafa
TypeScript · ≈265k stars · MIT
NousResearch/hermes-agent
The agent that grows with you.
Maintained by Nous Research
v0.23.3releasedPython · ≈151k stars · Apache-2.0
Graphify-Labs/graphify
A /graphify skill that turns a codebase, docs and schemas into a queryable knowledge graph.
Maintained by Graphify-Labs
v0.9.114releasedConnections
Relations from the Harness Hub knowledge graph, as charted in the State of Harness reports.
