lockstep
5 min readNaman Jain

Mem0 alternatives: 7 open-source and hosted options (2026)

Seven Mem0 alternatives compared from their own docs: Zep and Graphiti, Letta, Supermemory, Cognee, LangMem, Basic Memory, and native Claude Code memory.

People look for Mem0 alternatives for different reasons. Some want a fully local setup. Some need facts that change over time. Some are building on LangGraph, or want an agent platform rather than a memory component. "Alternative" means something different in each case.

This list groups seven options by the reason you'd switch, using each project's own repository and documentation, checked on 16 November 2026. For a deeper look at the four best-known, see Mem0 vs Zep vs Letta vs Supermemory.

The short answer

AlternativePick it if you needLicenseGitHub stars (16 Nov 2026)
Zep and GraphitiFacts that change over time, with historyGraphiti: Apache-2.0Graphiti: 32,835
LettaA platform that runs stateful agents end to endApache-2.025,457
SupermemoryOne API for memory, retrieval, and profiles, hosted or self-hostedMIT31,635
CogneeA free, local knowledge graph with no API keyApache-2.033,343
LangMemMemory tools for LangGraph agentsMIT1,771
Basic MemoryMarkdown notes people and agents both editAGPL-3.04,594
Native agent memoryA coding agent that remembers your preferencesBuilt inNot applicable

For reference, Mem0 itself is Apache-2.0 with 69,037 stars, and offers a hosted platform alongside the open-source version.

Zep and Graphiti: when facts change

Graphiti is Zep's open-source framework for temporal knowledge graphs. It stores entities and relationships with time attached, so it can tell what was true before a change and what is true now, instead of holding two conflicting notes. Zep builds a hosted context layer for enterprise data on top of it.

Graphiti runs on FalkorDB or Neo4j and needs an LLM provider for extraction. Its MCP server is labeled experimental. See temporal knowledge graph.

Letta: when you want a platform, not a component

Letta, formerly MemGPT, runs stateful agents that keep their state on the server. Each agent has core memory and archival memory and edits its own memory as it works. It runs on Letta Cloud or self-hosted. Choose it if you want the platform to run the agent; it's a bigger commitment than adding a memory component.

Supermemory: when you want one API for everything

Supermemory describes itself as context infrastructure for AI agents: memory, retrieval, and user profiles through one API. It indexes text, files, chats, and connector data into a semantic graph by namespace. It's hosted, or self-hosted as a single binary that can run offline, and it has a Claude Code plugin and an MCP server.

Cognee: when it must run locally and free

Cognee is an open-source AI memory platform that turns documents, code, and conversations into a self-hosted knowledge graph. Its README says it runs locally for free with no API key, using small models on your CPU. It suggests uses such as a company brain that connects decisions to the discussion and code behind them, and memory across agent runs.

LangMem: when you build on LangGraph

LangMem, from LangChain, helps agents extract information from conversations, refine their own prompts, and keep long-term memory. It provides memory tools agents use during a conversation, a background manager that consolidates knowledge, and native integration with LangGraph's long-term memory store. It works with any storage system through its core API.

Basic Memory: when people need to edit memory too

Basic Memory keeps knowledge as plain markdown files and builds a graph from notes and links, accessed over MCP. People can read and edit the same notes agents retrieve. It is free to run locally under AGPL-3.0, with a paid hosted workspace for syncing across devices.

Native agent memory: when it's a coding agent

If the agent is Claude Code or Codex, check what it already does before adding anything. Claude Code loads CLAUDE.md files and its own auto memory each session; Codex reads AGENTS.md and has an optional Memories feature. Both keep that memory on one machine. See how Claude Code memory works and Codex memory.

Add a memory layer when memory must cross tools, machines, or people, and connect it over MCP. See the best MCP memory server for Claude Code.

What every option shares

All seven are built for recall: return what an agent learned or was told. Some, like Graphiti, also track how facts change. None records whether a team agreed to something, who owns it, or what replaced it. For preferences, history, and user facts, that doesn't matter. For a team building against product and architecture decisions, it does.

Where Lockstep fits

Lockstep is shared team memory for Claude Code, built for decisions rather than recall. Decisions are captured from PRDs, Slack, and pull requests, become binding when their owner approves them, and are marked replaced when a newer version is approved. Claude Code is briefed on the current ones at the start of each session, and any MCP agent can query them.

It doesn't replace a memory layer for recall; many teams will use both. Session briefings work with Claude Code today. See how it works.

Sources checked 16 November 2026

For the four best-known in depth, see Mem0 vs Zep vs Letta vs Supermemory. For how memory differs from the context window and RAG, see agent memory vs context window vs RAG.

Frequently asked questions

What are the best open-source alternatives to Mem0?
Graphiti (Apache-2.0) for temporal knowledge graphs, Letta (Apache-2.0) for stateful agents, Supermemory (MIT) for a memory and context API, Cognee (Apache-2.0) for a local knowledge graph, LangMem (MIT) for LangGraph agents, and Basic Memory (AGPL-3.0) for markdown notes shared by people and agents.
Is Mem0 open source?
Yes. Mem0's GitHub repository is Apache-2.0 licensed, and Mem0 also offers a hosted platform. Many teams compare the open-source version with alternatives when they need a different memory model rather than a different license.
Which Mem0 alternative runs fully locally?
Cognee says it runs locally for free with no API key, using small models on your CPU. Supermemory can be self-hosted as a single binary, including offline. Basic Memory keeps plain markdown files on your disk. Graphiti runs on your own graph database but needs an LLM provider key.
Do coding agents need Mem0 or an alternative?
Often not at first. Claude Code already loads CLAUDE.md files and its own auto memory each session, and Codex has AGENTS.md and an optional Memories feature. Add a memory layer when memory must be shared across tools, machines, or people.
Can a memory layer track team decisions?
It can store the text of a decision, but memory layers are built for recall, not approval. They don't record who owns a decision, whether it was approved, or what replaced it, which is what a team needs to build against it.

Keep reading