Recall and Mem0 compared
Last updated: 13 October 2026
Mem0 is a memory layer for AI apps. It distills conversations into memories and returns the most relevant ones when your agent asks. It comes as a managed platform and as an Apache-2.0 library and server.
Every fact about Mem0 on this page is from Mem0'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 | Mem0 | |
|---|---|---|
| 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 memory layer that turns conversations into stored memories (facts) and returns the relevant ones on search, scoped by user, agent, app and run.1 |
| Hosted service | Yes. Data is stored in Frankfurt (EU). | Yes, the Mem0 Platform at app.mem0.ai.2 |
| 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 library (pip install mem0ai, npm install mem0ai) and a self-hosted server (Docker Compose) are Apache 2.0.2 On-prem deployment of the platform is an Enterprise option.3 |
| 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. | Monthly plans by request count: Hobby free, Starter $19, Pro $249, Enterprise custom.3 |
| Free to start | $5 of credit once, no card needed. | Hobby: 10,000 add requests and 1,000 retrieval requests a month, no card needed.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. | Mem0's own API: add, search and update memories. Python and TypeScript SDKs and a CLI.21 Not compatible with Recall's API. |
Pricing
The two price different things. Mem0 sells monthly plans with an allowance of add requests (storing a new memory) and retrieval requests (fetching memories for a context).3 Recall has no monthly fee: it charges per message token learned and per question, and context and search are included.
| Mem0 plan3 | Price a month | Add requests a month | Retrieval requests a month |
|---|---|---|---|
| Hobby | Free | 10,000 | 1,000 |
| Starter | $19 | 50,000 | 5,000 |
| Pro (adds graph memory, Dream consolidation, analytics) | $249 | 500,000 | 50,000 |
| Enterprise (adds on-prem deployment, SSO, audit logs) | Custom | Unlimited | Unlimited |
Recall: learning $1.75 per 1M message tokens (a message counts as at least 50 tokens), a question $0.0008 at minimal, $0.005 at low (the default) and up to $0.25 at max; context, search, storage and the night shift included; $5 of credit to start. A retrieval in Mem0 returns memories for your own model to read; the closest Recall calls, context and search, are included in the price, and a Recall question also writes the answer. Which costs less depends on how long your messages are and how often you ask.
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 | Mem0 | How each was graded | Compare directly? |
|---|---|---|---|---|
| LongMemEval | 91.0% (455 of 500) on LongMemEval-S | 94.4% (472 of 500) in the README; the per-question results file gives 93.4% (467 of 500)4 | Recall: LongMemEval's official judge, gpt-4o-2024-08-06 with the benchmark's own prompts. Mem0: Mem0's managed platform, top-200 retrieval, answers and PASS/FAIL verdicts by gpt-5 in Mem0's own harness; the README does not say which LongMemEval variant.4 | No: a different judge model and grading prompt, and the variant is not named. |
| LoCoMo | Not run in full. On a 75-question sample: 60.0% with a strict judge, 79.2% with a lenient one | 92.5% (1,425 of 1,540)4 | Mem0's harness, top-200 retrieval.4 | No: Recall has no full LoCoMo run. |
| BEAM 1M | Not run yet | 0.641 average score, 700 questions4 | Mem0's harness, top-200 retrieval.4 | Recall has no figure to compare. |
| BEAM 10M | Not run yet | 0.486 average score, 200 questions4 | Mem0's harness, top-200 retrieval.4 | Recall has no figure to compare. |
Mem0's README says these scores reflect its managed platform, which includes optimizations not available in the open-source SDK.2 Mem0 publishes no BEAM 100K or 500K figure that we found, and Recall has not run BEAM at 1M or 10M, so the two have no BEAM split in common.
Hosting and licences
Hosted. Mem0 runs the Mem0 Platform; on-prem deployment of it is part of Enterprise.3 Recall runs a hosted service with data stored in Frankfurt (EU).
Self-hosted. Mem0's library and self-hosted server are Apache 2.0, an OSI licence; you bring the vector database, the LLM and the hosting.23 Recall's server is source-available under the Functional Source License (each version Apache 2.0 after two years) and needs only Postgres with pgvector, or an embedded database, plus your model provider keys.
API
Mem0 has its own API: you add memories, search them and update them, scoped by user, agent, app and run.1 It is not compatible with Recall's API, so moving means changing your calls. A Mem0 user maps to a Recall peer and a run or conversation to a session. If you kept the conversations, send them to Recall's messages endpoint and Recall learns from them; memories you want kept word for word can go in as conclusions (POST …/conclusions).
When to choose Mem0
- You want an Apache-2.0 library that runs inside your own process, or a fully open-source server.
- A free monthly allowance fits your app: Mem0's Hobby plan includes 10,000 add requests and 1,000 retrieval requests a month.
- You prefer a fixed monthly price by request count to paying per token and per question.
- You want graph memory with entity linking (Pro and up), or you weigh Mem0's higher published LongMemEval figure, measured on its own harness and judge.
When Recall fits better
- You want Recall to write the answer, with citations, a confidence and an explicit "I don't know", in one model call.
- You want facts with dates: when a fact stops being true it is closed, and the history stays visible.
- You want no monthly fee, data stored in the EU, or the v3 memory API that Honcho's clients also speak.
See all comparisons, or try Recall with $5 of free credit: quickstart.
Sources
Read on 13 October 2026.