Mem0 extracts memories from conversations and exposes add, search, update, and delete. Zep builds temporal context graphs using its open-source Graphiti framework. Letta, formerly MemGPT, keeps memory blocks inside the agent's context that the agent edits itself. Archilas builds team memory from connected tools and serves it over MCP. Each fits a different job.
This page describes each project from its own public documentation, as read on 30 October 2026. Check the links, because these projects change quickly. There is no score here. For what I will and will not claim about Archilas, see what I am not claiming.
Mem0
Mem0 is a memory layer developers call from their application. Its hosted MCP server exposes tools such as add_memory, search_memories, update_memory, and delete_memory, and authenticates with a Mem0 account or API key (Mem0 MCP docs). Its earlier local OpenMemory server is being sunset in favour of a self-hosted server (repo notice).
Zep
Zep builds context graphs of entities, relationships, and source episodes, and tracks how facts change over time (Zep docs). The graph framework underneath, Graphiti, is open source and records when a fact became valid and when it stopped being valid (Graphiti README).
Letta
Letta (formerly MemGPT) uses memory blocks: labelled sections of the agent's context window with a character limit. The agent reads and rewrites them with built-in tools, and blocks can be shared between agents or set read-only (Letta memory blocks).
Archilas
Archilas gives AI a memory of the team. It syncs from tools a team connects (Slack, GitHub, Gmail, Notion, Linear, Calendar, Drive, AI chats), keeps dated notes with sources, and serves answers to any MCP client through ask, remember, changes, import_history, and forget. It is a hosted service. It is not an SDK for building your own app's memory.
Choosing
- Building a product that remembers its own end users: Mem0 or Zep are designed for that.
- Building a long-running agent that manages its own context: Letta's model fits.
- Wanting the coding agents and chat assistants you already use to know what your team decided: that is what Archilas is for.
Related posts
Questions
- Is Archilas an alternative to Mem0?
- Partly. Both can be reached over MCP. Mem0 is mainly a memory API for developers building apps. Archilas builds a team's memory from connected tools for the agents the team already runs.
- Which is most accurate?
- This post does not say. There is no head-to-head benchmark here, and claims like that should come with a published method.