Self-hosting

FishMem's memory engine is open source under Apache-2.0 at github.com/fishmem-labs/fishmem. The hosted cloud runs this same engine on the same code path, so self-hosting gives you identical memory behavior — you just operate the infrastructure yourself.

What you'll run

The engine needs a few things wired up. At a high level:

• A graph store for relationships between facts.
• A vector index for semantic recall.
• An LLM provider for fact extraction and reconciliation.
• An embedding provider to turn text into vectors.

The engine sits on top of these and handles extraction, storage, and recall.

High-level steps

1. Clone the repo from github.com/fishmem-labs/fishmem.
2. Configure your stores — point the engine at a graph store and a vector index you operate.
3. Configure model providers — set your LLM and embedding providers and their credentials.
4. Run it and start adding and searching memories against your instance.

See the repository README for the exact configuration keys, supported store and provider options, and version requirements — those live with the code so they stay current.

When self-hosting makes sense

Self-hosting is the right call when you need:

• Data residency or control — memories stay inside your own infrastructure.
• Network isolation — run it inside your VPC with no external API calls.
• Custom providers — bring your own models or stores.
• Local evaluation — try the engine end to end before committing.

It costs you operational work: you run and scale the graph store, the vector index, and the model integrations, and you handle upgrades.

When the cloud is easier

If you'd rather not operate that stack, FishMem Cloud runs the same engine for you and adds a dashboard, API keys, usage metering, and billing. For a full comparison and guidance on choosing, see Platform vs. open source.