Really Artificial Explore GitHub
Menu
← ALL WORK
01 / Connect · Available software

One gateway.
Your infrastructure.

Freeport gives your application a common interface to model providers, with routing, spending controls, and an audit trail in the environment you run.

Run Freeport Read the source
MIT LICENSEDSELF-HOSTEDNODE.JS + TYPESCRIPTSQLITE BACKED
The architecture

A place to control
what happens between.

Your application talks to Freeport. Freeport applies your configured controls and routes the request to a model provider.

Cloud or local: the request goes to the provider you choose.

Built into the gateway
01

Choose your providers.

Route to OpenAI, Anthropic, Google, or a configured OpenAI-compatible endpoint, including local model servers.

02

Plan for failure.

Configure fallback chains and circuit breakers. Distribute requests across multiple keys for a provider.

03

Set spending boundaries.

Track usage by project, set daily or monthly budgets, and use a kill switch when traffic needs to stop.

04

See what happened.

Inspect request logs, usage, and cost in the admin UI. Export Prometheus metrics into your own monitoring.

05

Add controls at the boundary.

Configure PII detection, content filtering, and rate limits. Provider keys are encrypted at rest.

06

Keep familiar interfaces.

Accept OpenAI and Anthropic message formats, with translation for streaming, tools, vision, and JSON mode.

Quickstart

Start local.
See the whole path.

You will need a supported Node.js version and either provider credentials or a local model endpoint. Start with the gateway, then connect your application.

Full setup & configuration
  1. 01

    Start the gateway.

    Terminal
    npx @reallyartificial/freeport

    Freeport starts at http://localhost:4000.

  2. 02

    Add a provider.

    Open http://localhost:4000/ui/ in your browser. Add your provider credentials or configure a local endpoint.

  3. 03

    Connect your application.

    Set your OpenAI SDK base URL to the gateway’s /v1 endpoint. For Anthropic clients, use the gateway’s /v1/messages endpoint. Send a request and inspect its log in the dashboard.

Before you build

A few things
worth knowing.

Where does my data go?

The gateway and its SQLite database run in your environment. Requests are forwarded to the model providers you configure. Using a cloud provider means sending prompts to that provider; using a local endpoint keeps model traffic within that setup.

Does Freeport host the models?

No. Freeport sits between your application and model providers. You supply provider credentials or an endpoint for a model server you operate.

Can I use it with my existing application?

Freeport accepts OpenAI-style requests at /v1 and Anthropic Messages requests at /v1/messages. Start by changing the client base URL, then test the request shapes and provider features your application uses. Compatibility varies with upstream features.

What should I configure before deployment?

Set NODE_ENV=production and configure FREEPORT_API_KEY and FREEPORT_ADMIN_API_KEY. Review provider credentials, network access, budgets, guardrails, and log retention for your environment. The repository documents the configuration options.

What does it cost?

Freeport is MIT-licensed software you can run and modify. You pay for your own infrastructure and any upstream model usage.

Open source, all the way through

Read it. Run it. Improve it.

View on GitHub Report an issue