Memtro vs Letta
A framework for stateful agents and their own memory, versus a company memory every agent can point at.
This comparison describes Letta (formerly MemGPT) as we understand it at the time of writing from its public materials. Products change; check the current details before deciding.
Letta is an open-source framework and service for stateful agents. It grew out of the MemGPT research on managing an agent's context — what stays in the window, what is paged out to long-term storage, how an agent edits its own memory — and gives you agents that persist across sessions.
Memtro is not an agent framework. It is a hosted memory and tool layer that whatever agent you already run can point at: an OpenAI-compatible endpoint and an MCP server carrying the organisation's memory, connectors that read by default, routing and jobs.
Where they differ
| Memtro | Letta | |
|---|---|---|
| What it is | A memory and connector layer behind an endpoint | A framework for building stateful agents |
| Whose memory | The organisation's, shared across people and clients, classified per fact | The agent's own, managed by the agent |
| Connectors | Tools for business systems, reading by default and acting where you grant it, with hourly sync and anchoring | Tools you define |
| Running it | Hosted | Open source, self-host or cloud |
| Audience | A whole company, through tools it already uses | Engineers building agents |
| Time | Bitemporal: as-of reads and a timeline of changes | Agent memory as it currently stands |
| Models | Your keys, routed by difficulty | Your choice, in your code |
Honestly, where Letta is stronger
If you are designing an agent and want control over how its context is managed — what it remembers, when it writes to itself, how its window is composed — Letta gives you that at the level where it is decided. It is open source, and you can read and change the mechanism. Memtro deliberately does not expose that: context assembly is ours, and you steer it with scope and permissions rather than code.
When to choose which
Choose Letta when you are building agents and the memory architecture is part of what you are building.
Choose Memtro when you want a company's memory to exist once and be reachable from everything — chat, editors, jobs, MCP clients and whatever agents you build, Letta included. An agent built on Letta can call Memtro's tools for company context like any other MCP client.