Freeway

A local control plane for free LLM APIs.

Freeway is an open-source gateway that aggregates the fast-moving free LLM ecosystem behind a unified local API surface. Bring your own keys — Freeway normalizes OpenAI/Anthropic protocols, routes requests, and falls back across providers. All from localhost.

简体中文 · Contributing (EN) · 贡献指南 (中文)

Mission

Freeway is a local control plane for free LLM APIs. It normalizes protocol differences, resolves models, checks route availability, and falls back when a provider fails — all from localhost.

The goal is not to wrap one provider. The goal is to offer one gateway layer that can keep absorbing the providers, models, and compatibility quirks that matter across the free-model ecosystem.

Why this exists

The free-model ecosystem is expanding quickly, but the developer experience is still fragmented:

• provider APIs differ in behavior and response shape
• model availability changes quickly
• free tiers appear, move, rate-limit, or disappear
• clients and coding agents still want one predictable local endpoint

Freeway compresses that fragmentation into a single local gateway that is easier to operate, easier to integrate, and easier to extend.

What Freeway provides

• Protocol normalization — OpenAI and Anthropic compatible endpoints from one server
• Fallback routing — when a provider is rate-limited or unavailable, Freeway tries another
• Model discovery — fetch available models from supported providers and keep a unified free-tier catalog updated
• Runtime API key management — configure provider keys through the web UI or REST API, no restart required
• Health checks — monitor provider availability and latency from the console
• Local web console — browse providers and models, check health, configure keys, test requests
• Works with Claude Code, Cursor, Continue.dev, OpenCode, and any OpenAI/Anthropic-compatible client

Coverage philosophy

Freeway is not positioned as a thin wrapper for one API vendor.

It is an aggregation layer designed to keep up with the free LLM landscape over time. That means tracking useful providers, normalizing compatibility gaps, and making the resulting surface more stable for local tools, scripts, and agent workflows.

The ambition is broad coverage. The implementation stays pragmatic: integrate what matters, keep the gateway reliable, and improve compatibility as the ecosystem shifts.

Ecosystem references

Freeway tracks the broader free-model ecosystem through public resource collections, including:

• awesome-free-llm-apis
• free-llm-api-resources

These are ecosystem references, not hard dependencies. They help guide ongoing provider coverage and compatibility work.

Current capabilities

Compatibility layer

• OpenAI-compatible chat completions
• OpenAI-compatible model listing
• Anthropic-compatible messages API bridging
• Stable non-stream usage normalization across OpenAI-compatible and Anthropic-compatible responses
• Conservative Anthropic streaming behavior without fake zero-usage placeholders

Gateway operations

• Provider health checks and status summaries
• Model catalog refresh and cache fallback
• Local runtime key management
• Optional gateway auth with FREEWAY_API_KEY
• Optional outbound proxy support with HTTP_PROXY

Local console

• Browse providers and models
• Check provider health and latency
• Configure provider keys
• Refresh model catalogs
• Test local requests from the browser

Supported Providers

Currently wired through src/providers/index.ts:

openrouter, groq, github, cloudflare, siliconflow, cerebras, mistral, cohere, nvidia, llm7, kilo, zhipu, opencode

Quick Start

1. Prerequisites

• Node.js 18+
• npm

2. Install and launch

npm install
npm run build
npm start

Default server address:

• http://localhost:8787

3. Open the console

Visit:

• http://localhost:8787/

Then configure provider keys in the API Keys tab, or provide them with environment variables.

Configuration

Configure your agent

Freeway exposes both OpenAI and Anthropic compatible endpoints, so most coding agents and LLM clients can connect directly.

Detailed per-agent setup guides are available in docs/agents/.

• OpenCode setup guide

Claude Code

Set the base URL to Freeway:

export ANTHROPIC_BASE_URL=http://localhost:8787
export ANTHROPIC_API_KEY=<your FREEWAY_API_KEY or any non-empty string>

Then run claude normally. Freeway routes Claude Code's Anthropic API calls to the best available free provider.

Cursor

In Cursor Settings → Models → OpenAI API Key:

• Base URL: http://localhost:8787/v1
• API Key: your FREEWAY_API_KEY (or leave empty if gateway auth is off)

Continue.dev

In config.json:

{
  "models": [
    {
      "title": "Freeway",
      "provider": "openai",
      "model": "llama-3.3-70b",
      "apiBase": "http://localhost:8787/v1",
      "apiKey": "your FREEWAY_API_KEY"
    }
  ]
}

OpenCode

Set environment variables before running:

export OPENAI_BASE_URL=http://localhost:8787/v1
export OPENAI_API_KEY=<your FREEWAY_API_KEY>

Any other OpenAI/Anthropic client

Point the base URL to http://localhost:8787 (Anthropic) or http://localhost:8787/v1 (OpenAI) and provide your gateway key if configured.

API key precedence

Effective key precedence is:

1. Runtime key set via UI/API
2. Environment variable
3. Persisted .freeway/config.json

Common environment variables

API Examples

OpenAI-compatible chat completion

curl http://localhost:8787/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $FREEWAY_API_KEY" \
  -d '{
    "model": "llama-3.3-70b",
    "messages": [{"role": "user", "content": "Say hello from Freeway"}],
    "stream": false
  }'

Force a provider explicitly

{
  "model": "groq/llama-3.3-70b"
}

Anthropic-compatible messages request

curl http://localhost:8787/v1/messages \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $FREEWAY_API_KEY" \
  -d '{
    "model": "llama-3.3-70b",
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Claude-style local base URL usage

For Anthropic-compatible clients that let you override the base URL, point them at:

• http://localhost:8787

Freeway serves the compatibility routes under that origin.

HTTP Endpoints

Project Structure

src/
  index.ts                # Entry point
  server.ts               # HTTP server + routes + static hosting
  router.ts               # Provider routing and retry logic
  providers/              # Provider definitions and model sync orchestration
  models/                 # Canonical model registry + sync/cache adapters
  web/                    # Console UI (HTML/CSS/JS)
  config*.ts              # Runtime + persisted key config
  health.ts               # Provider health checks and summary
  anthropic-bridge.ts     # Anthropic <-> OpenAI request/response bridge
  usage.ts                # Gateway-level usage normalization helpers

Development

npm run dev
npm run build
npm start
npm run test:usage

Contributing

• English: CONTRIBUTING.md
• 中文： contribution.md

License

MIT