Integrations

Add FishMem memory to any agent framework via the REST API, the fishmem SDK, or a first-class adapter.

FishMem plugs into any agent framework two ways:

Directly — call the mem0-compatible REST API or the fishmem SDK from wherever your agent runs. This works with every framework, no adapter required.
A first-class adapter — a small package that wires FishMem into your framework's own memory or tool abstractions, so recall and persistence happen on the framework's lifecycle instead of by hand.

Because the API is mem0-compatible, many existing mem0 integrations can be repointed at FishMem by changing the base URL to https://fishmem.com and the key to fm_.... (mem0 compatibility is a migration convenience — it does not imply any endorsement.)

Adapters

Vercel AI SDK

Retrieve-and-inject or expose memory as a tool with the ai package.

Mastra

Attach a FishMem-backed memory to a Mastra Agent.

OpenAI Agents SDK

Recall before each run, persist after, as session memory.

Cloudflare Agents SDK

Per-user memory inside an Agent Durable Object.

The integration pattern

Every adapter is a thin wrapper around the same four steps. If your framework isn't listed, implement these directly with the SDK:

Search FishMem for memories relevant to the incoming message, scoped by user_id (and optionally agent_id / run_id).
Inject the results into the model's context or system prompt.
Run the model.
Add the new exchange back to FishMem so the next turn can recall it.

Reads (search and get) are free; an inferred add costs 2 credits. See scoping for how user_id, agent_id, and run_id isolate memories.