Mem0

Key features

• Add, search, and update memory loop Send messages to store facts, then search, fetch, update, or delete them per memory, with per-memory history on both editions. Agents read and revise stored context through the same calls instead of rebuilding session state by hand.[1][2]

• Entity scoping by user, agent, app, and run Memories separate along user, agent, and run identifiers, with app-level tenancy on the platform, so one deployment can serve many end users without leaking context. Multi-tenant products keep each user's recall isolated by construction.[1][2]

• Graph memory on the platform The managed edition keeps an entity graph over extracted memories and feeds those connections into ranking, which surfaces related facts plain similarity might miss. Open source still extracts entities for ranking boosts but no longer ships a queryable graph.[1][2]

• Dream memory consolidation A platform capability that consolidates memories over time, keeping long-lived recall current as facts change. Agents that run for months accumulate cleaner context instead of stale duplicates.[1][2]

• Memory decay and expiration Opt-in decay reinforces recently used memories while dampening stale ones at search time, and per-memory expiration dates bound how long a fact lives. Teams shape recall freshness without manual pruning jobs.[1][2]

• Python and JavaScript SDKs plus REST Both editions expose the same core operations through two SDKs and a REST API, with the homepage quickstart saving and searching a memory in a few lines. Existing codebases adopt recall without changing agent frameworks.[1][2]

• MCP and framework integrations Model Context Protocol support plus documented integrations with agent frameworks connect memory to coding agents and orchestration stacks. Teams wire recall into the tools agents already use rather than building adapters.[1][2]

• Enterprise governance on the platform Audit logs, workspace governance, SSO, on-prem deployment, and SLA support ship on the enterprise plan, with SOC 2 Type 1, HIPAA, and GDPR compliance stated on the site. Regulated deployments get the paper trail procurement asks for.[1][2]

Who will love it

• A few SDK lines store and retrieve memories through the same add and search calls, so teams keep their framework and gain cross-session context. The quickstart path reaches a first memory in minutes on either edition.[1][2]

• The Apache-2.0 open-source edition with pip install and swappable vector database, model, and embedder gives full data control to teams that already run their own infrastructure. Memory stores stay inside the team's own boundary.[1][2]

• Platform audit logs, workspace governance, SSO, on-prem deployment, and stated compliance coverage answer the controls that regulated rollouts require. Procurement gets a managed path with a documented paper trail.[1][2]

Who should avoid it

• The documented path is SDKs, REST, and server setup; no verified no-code builder exists, so visual-first teams will stall. Every customization assumes someone on the team writes code.[1][2]

• Graph memory and Dream consolidation are platform-only paid capabilities, and the Hobby free tier caps retrievals at 1,000 per month, so free-tier or self-hosted users plan around that boundary. Check the live pricing page before committing recall-heavy workloads.[1][2]

• If the task fits one stateless call, running a memory layer adds moving parts without a matching payoff. A direct model API call stays cheaper and simpler there.[1]

Compatibility and requirements