Recall and Letta compared
Last updated: 13 October 2026
Letta (formerly MemGPT) is a platform for stateful agents: the agent itself runs on Letta and keeps and edits its own memory. Recall is a memory service that an agent you already have calls. They solve different parts of the problem, so this page is mostly about which one fits.
Every fact about Letta on this page is from Letta's own public pages, linked under Sources, as we read them on 13 October 2026. What we could not check there is left out. Tell us at support@recallmem.dev if something has changed.
At a glance
| Recall | Letta | |
|---|---|---|
| What it is | A memory API. It learns what is true about each person from your agent's conversations and answers questions about them, with sources, in one model call. | A platform for building stateful agents that learn from experience. Letta Code is its agent harness.12 |
| Hosted service | Yes. Data is stored in Frankfurt (EU). | Yes, Letta Cloud and the Letta API.13 |
| Self-hosting and licence | Yes. The server is source-available under the FSL-1.1-ALv2 (each version becomes Apache 2.0 after two years); the SDKs and integrations are MIT. Self-hosting, source. | Yes. The local runtime or a self-hosted App Server. Apache 2.0.452 |
| Pricing | Pay as you go from prepaid credit: learning $1.75 per 1M message tokens, questions $0.0008 to $0.25 by level. Context, search, storage and the night shift are included. Pricing. | Free; Pro $20 a month; API Plan $20 a month plus $0.10 per active agent a month, $0.00015 per second of tool execution, and model usage at its token cost; Teams Pro $20 per seat a month; Enterprise custom.3 |
| Free to start | $5 of credit once, no card needed. | Free plan with up to 3 stateful agents and limited usage.3 |
| API | The v3 memory REST API, which Honcho's clients also speak; TypeScript and Python SDKs, an MCP server, and plugins for Claude Code and Codex. | The Letta API and the Letta Agent SDK for TypeScript; Python apps use the App Server's WebSocket protocol.6 Not compatible with Recall's API. |
Pricing
Letta prices an agent platform: plans, active agents, tool execution time and model usage.3 Recall prices a memory service: $1.75 per 1M message tokens learned and $0.0008 to $0.25 per question, with context, search and storage included and no monthly fee. The two do not compare line by line; with Recall your agent and its model run, and are paid for, wherever you run them today.
| Letta plan3 | Price | What it includes |
|---|---|---|
| Free | $0 | Up to 3 stateful agents, limited usage, your own API keys |
| Pro | $20 a month | A usage quota, pay-as-you-go overage, up to 20 stateful agents |
| API Plan | $20 a month | Unlimited agents at $0.10 per active agent a month, tool execution at $0.00015 a second, model usage at its token cost |
| Teams Pro | $20 a seat a month | Shared agents and access control |
| Enterprise | Custom | Volume pricing, SSO, dedicated support |
Published benchmark results
Each figure is as its maker published it, with its own answer model, judge and harness. None is a rerun, and a different judge alone can move a score by several points, so read the last column before comparing. Recall's runs and method are on Evals.
| Benchmark | Recall | Letta | How each was graded | Compare directly? |
|---|---|---|---|---|
| LoCoMo | Not run in full. On a 75-question sample: 60.0% with a strict judge, 79.2% with a lenient one | 74.0%, an agent with a filesystem, GPT-4o mini7 | Letta: judge and number of questions not stated in the post.7 | No: Recall has no full LoCoMo run. |
| LongMemEval-S | 91.0% (455 of 500) | No result found | — | Nothing to compare. |
| BEAM | 0.683 (all 400 questions) at 100K, 0.670 (all 700 questions) at 500K | No result found | — | Nothing to compare. |
Letta's post measures how well a simple agent with file tools does on LoCoMo, not Letta's own memory system as a whole.7
Hosting and licences
Hosted. Letta Cloud runs your agents. Recall runs only the memory, with data stored in Frankfurt (EU); your agent runs wherever it runs now.
Self-hosted. Letta's local runtime and App Server are Apache 2.0; self-hosted agents are not reachable through Letta's web app and are not backed up automatically.45 Recall's server is source-available under the Functional Source License (each version Apache 2.0 after two years).
API
In Letta you create an agent, then send it messages and stream its replies.6 Its memory lives in MemFS, a git-backed memory filesystem the agent inspects and edits, and a background "dreaming" pass consolidates it.8 Recall has no agent runtime: your agent sends it conversations, reads context for its prompt and asks it questions. Moving from one to the other is a change of architecture, not of settings.
When to choose Letta
- You want a platform that runs the agent itself, with its state, tools and memory managed for you.
- You want an agent that edits its own memory, or you build on Letta Code.
- You want an Apache-2.0 runtime you can host.
When Recall fits better
- You already have an agent, on any framework or model, and want it to remember the people it talks to.
- Many people, many conversations: Recall keeps a separate, dated record per person and answers questions about them with sources.
- You want memory in Claude, ChatGPT, Cursor, Claude Code or Codex through MCP or a plugin, without moving your agent.
See all comparisons, or try Recall with $5 of free credit: quickstart.
Sources
Read on 13 October 2026.