Overview
Open-source, mem0-compatible memory for AI agents.
FishMem is a memory engine for AI agents. It extracts the durable facts worth remembering from a conversation, stores them on a hybrid graph + vector layer, and recalls them when they're relevant — across sessions and millions of memories.
One engine, two ways to use it:
- Open source — self-host the
fishmemengine (Apache-2.0). - Hosted cloud — FishMem Cloud runs the same engine behind a mem0-compatible REST API, with a dashboard, API keys, usage metering, and billing.
Get started
Quickstart
Write and recall your first memory in minutes.
Authentication
Create an API key and call the API securely.
How memory works
Extraction, graph + vector storage, reconciliation.
API reference
Every endpoint, with request and response shapes.
Explore
FishMem Cloud
Hosted, mem0-compatible API with dashboard, metering, webhooks, and billing.
FishMem Open Source
Self-host the Apache-2.0 engine and dashboard.
Cookbooks
End-to-end recipes for support, coding, and companion agents.
Integrations
Vercel AI SDK, Mastra, OpenAI Agents SDK, Cloudflare Agents SDK.
Agent Plugins
Persistent memory for Claude Code, Codex, OpenClaw, and Hermes.
API Reference
Every endpoint, with request and response shapes.
Why memory
LLMs are stateless — they forget everything between sessions, and RAG over static documents isn't the same as memory that changes over time. FishMem keeps bi-temporal facts (it knows what's true now and what used to be true) and recalls them with hybrid vector, keyword, temporal, and graph-diffusion signals. The API is mem0-compatible, so adopting it is a base-URL change, not a rewrite.