Choosing memory for an AI agent
Every entry checked against that project’s own documentation on . Prices and limits change; the shape of the design rarely does.
The products in this space are not variations on one idea. They disagree about what remembering means, and that disagreement shows up in the code you end up writing. Picking by score gets you the wrong answer. Picking by which disagreement matches your product gets you the right one.
One page each, written to be useful even if you choose that one rather than us. We build WOS, so none of this is neutral.
One page for each
- Mem0 alternatives Extract facts with a model as you write. Right for you if you want conversations turned into clean facts, and the widest ecosystem.
- Zep alternatives Build a temporal knowledge graph. Right for you if your questions are about when something was true, not just whether it is true.
- Letta alternatives Let the agent manage its own memory. Right for you if you want the agent to own and edit its own memory, inside their runtime.
- Cognee alternatives Run a pipeline over documents into a graph and vectors. Right for you if what you are remembering is a document corpus, and you want to own the pipeline.
- Supermemory alternatives Bundle memory with retrieval over connected sources. Right for you if you want one API over connectors, documents and chat memory together.
- Building agent memory on a vector database Build it yourself on a vector database. Right for you if you are prototyping, or you want no vendor in the path and will build the rest.
How to choose
The shape of your question, not a feature list. WOS is the last row, on the same terms as the rest.
| Look at | If this is your situation |
|---|---|
| Mem0 | You want conversations turned into clean facts, and the widest ecosystem. |
| Zep | Your questions are about when something was true, not just whether it is true. |
| Letta | You want the agent to own and edit its own memory, inside their runtime. |
| Cognee | What you are remembering is a document corpus, and you want to own the pipeline. |
| Supermemory | You want one API over connectors, documents and chat memory together. |
| Vector DB | You are prototyping, or you want no vendor in the path and will build the rest. |
| WOS | You need the same store to work in many languages, and a per-query cost that does not drift as history grows. |
Where WOS fits
Nothing you send is rewritten or summarised, and the work happens at read time. A question finds the memory whatever language either one is written in, and no language model ever runs over what is stored. A query returns a bounded context: 2,484 tokens at the median over 2.2 million memories, measured with a tokenizer rather than estimated. Median engine round trip is about 320ms.
Where it is the wrong choice. It is closed source and hosted, so running it on your own hardware is a conversation rather than a download. There is no knowledge graph, so a question that is really about validity intervals is better served elsewhere. And it is the newest name on this page, with a smaller ecosystem than the ones above it.
Full detail in the documentation, the measured numbers on how we benchmark, and the rates on pricing.
- Every entry checked against that project’s own documentation on September 1, 2026.
- We quote nobody else’s prices, and use only figures each project published itself.
- Every page says what you would give up by leaving, as specifically as the rest of it.
- Out of date or unfair? Tell us and it gets corrected.