Letta and Mem0 get compared a lot, but they're different categories. Letta (formerly MemGPT) is a stateful agent framework — it runs agents with self-editing memory blocks and a whole runtime. Mem0 is a memory layer you add to an agent you already have. So the real question is whether you want a framework to build on, or a memory component to drop in. (Looking at the whole field? See our roundup of the best memory for AI agents.)
Letta vs Mem0 at a glance
| Letta (MemGPT) | Mem0 | |
|---|---|---|
| Category | Stateful agent framework | Memory layer / API |
| Scope | Runs the agent + its memory | Adds memory to your agent |
| Adopt when | Building on a memory-first framework | Keeping your own agent, adding memory |
| Documents / RAG | Via tools you add | Not the focus |
When to choose Letta
Choose Letta when you want the framework itself to own agent state and memory management, and you're comfortable building your agent inside its model.
When to choose Mem0
Choose Mem0 when you already have an agent and just need a memory store to plug in, without adopting a whole framework.
Where Kit for AI fits
If what you actually need is memory plus document retrieval — not a new agent runtime — Kit for AI is a drop-in API: remember/recall memory and document knowledge bases with hybrid retrieval, callable from whatever agent or framework you're already using, or natively over MCP.
Related reading
Go deeper: our Letta alternative take · our MemGPT alternative take — or start with the complete guide to persistent memory for AI.
Want memory that reads your documents too, across every tool you use? Start free and connect over MCP or the REST API.