Pick by the job. Chat-app memory suits one person in one app. Rules files suit fixed project conventions. RAG suits finding passages. Mem0, Zep, and Letta suit developers building memory into their own app. Archilas gives AI a memory of the team, served to any MCP client.
The options
- Built-in memory in a chat app: personal, inside one product.
- CLAUDE.md, Cursor rules, AGENTS.md: fixed rules you write by hand, per project.
- RAG over documents: good for passages, weak on which decision is newer. See memory vs RAG.
- Mem0, Zep, Letta: libraries and services for developers. See how they store memory.
- Archilas: hosted team memory from your tools and AI chats, read by any MCP client, with cited, dated answers.
Questions to ask any tool
- Does it keep the date and source of each fact?
- Does it say "I don't know" when nothing matches?
- Can every AI tool you use read it, or only one?
- Can you delete one fact, one source, and everything?
A longer list is the MCP memory server checklist.
Questions
- Is Archilas a Mem0 alternative?
- Only if you want hosted team memory for MCP clients. If you are building memory into your own app, Mem0, Zep, or Letta fit better.
- Do I need Archilas if I have rules files?
- Rules files hold fixed conventions. Archilas holds decisions that change, from tools where your team talks.