SyftHub

The home for AI/ML endpoints.

Discover, share, and run AI models and data sources — like GitHub, but for endpoints.

Documentation · Python SDK · Contributing

---

What is SyftHub?

SyftHub is a registry and discovery platform for AI/ML endpoints. Publish a model or a data source, point friends or teammates at it, and let anyone build on top of it — through a web UI, an SDK, or a chat interface backed by retrieval-augmented generation.

If you've ever wished there were a single place to find the model your team's been talking about, share your own, or wire one into a chat app without standing up new infrastructure — that's what SyftHub is for.

Highlights

• Browse and discover public models and data sources, sorted by popularity and freshness.
• Publish your own with public, internal, or private visibility, full markdown READMEs, and versioning.
• Chat with endpoints — built-in RAG aggregator orchestrates retrieval and generation across endpoints.
• Collaborate in organizations with role-based access for teams.
• Federated identity — SyftHub issues short-lived signed tokens so satellite services can verify users without phoning home.
• A Python SDK that searches sources, shows what each one costs and needs before anything is sent, and pays metered sources on the way.

Quick Start

Clone, copy the env template, and run:

git clone https://github.com/IonesioJunior/syfthub.git
cd syfthub
cp .env.example .env
make dev

The full stack starts behind nginx at http://localhost:8080. API docs live at http://localhost:8080/docs.

To stop everything: make stop.

Using the Python SDK

# Python (syfthub-sdk 2.x: pip install --pre syfthub-sdk)
import syfthub

with syfthub.Hub(token="syft_pat_...", budget=1.00) as hub:   # optional cap, per currency
    for source in hub.sources.list():
        print(source.path, source.pricing)

    plan = hub.search("adverse events in phase 3", ["erin/trials", "carol/papers"])
    pre = plan.preflight()                 # what each source costs and needs; nothing sent
    results = plan.skip(*(row.path for row in pre.held)).execute()
    for path, doc in results.documents:
        print(path, doc.score, doc.content[:80])

See the Python SDK guide for the full reference.

The TypeScript SDK, the Go SDK and the syft CLI are retired. Their last state is kept on the legacy-sdks branch, and the published packages (@syfthub/sdk 0.3.1 on npm, syfthub-sdk 0.3.1 on PyPI) stay available for existing users. The TypeScript client now lives in components/frontend/src/lib/sdk as part of the web app.

Documentation

Full docs live in docs/:

• Architecture overview — services, data flow, tokens.
• Local setup — clone to running.
• Publishing endpoints — share your first model.
• API reference — backend, aggregator, MCP.
• Runbooks — deploy, rollback, incident response.

Project layout

syfthub/
├── components/        services (backend, frontend, aggregator, mcp)
├── sdk/               the Python SDK
├── deploy/            deployment configs
└── docs/              documentation

Contributing

Issues and pull requests are welcome — see CONTRIBUTING.md. For release process and versioning, see RELEASING.md.

License

Apache 2.0