Mem0

Mem0 is a memory layer for AI applications and agents. An application sends useful conversation turns to Mem0 after an interaction, then searches for relevant memories before the next model call. It supplies reusable facts, preferences, decisions or goals instead of replaying an entire transcript. Memories can be scoped by user, agent, run and metadata, which helps keep context separate across people and workflows. Mem0 extracts facts by default rather than storing a verbatim transcript, creates embeddings for retrieval and checks related memories to reduce duplication; applications can also store raw content or explicitly update and delete records.

Mem0 is offered as a managed Platform and as Apache-2.0-licensed open-source software. The open-source edition can run as an embedded Python or TypeScript library or as a self-hosted server with a dashboard, API keys and request auditing. The library exposes configurable LLM, embedding, vector-store and reranking components, while the managed service operates the storage and retrieval infrastructure. Platform-only capabilities and their OSS equivalents differ, particularly Graph Memory and multi-signal retrieval.

Top features

• Scoped persistent memory: Associate memories with a user_id, agent_id, run_id and application metadata instead of treating every conversation as one global history.
• Extraction and deduplication: Turn messages into reusable facts with metadata and embeddings, checking existing context before adding another memory.
• Search before generation: Retrieve context by query and filters, then let the application decide which results belong in the model prompt.
• Memory controls: Add, search, inspect, update and delete memories; disable inference when exact raw content must be stored.
• Flexible deployment: Use the hosted API, install the Python or TypeScript library, or run the self-hosted server stack on infrastructure you control.
• Configurable OSS components: Choose LLMs, embedding models, vector stores and optional rerankers rather than being locked to the default local configuration.

Use cases

• Preserve preferences and account context for personal assistants without sending every prior conversation to the model.
• Let customer-support agents recall earlier issues, resolutions and durable customer details across sessions.
• Carry project decisions, goals and feedback through long-running coding, research or operations workflows.
• Build separate memory scopes for multiple users, agents or runs in the same application.