About

About FreeRide

A local OpenAI-compatible gateway for running AI coding agents on free-tier inference providers. No subscriptions, no vendor keys, no lock-in.

What is FreeRide?

FreeRide is a lightweight Python daemon that runs on your machine and acts as a smart router for AI inference requests. When you point your coding agent (Claude Code, OpenAI Codex, Gemini CLI, Aider, Continue, etc.) at localhost:11343, FreeRide fans out each request across your configured free-tier provider keys: OpenRouter, Groq, NVIDIA NIM, Cloudflare Workers AI, HuggingFace, Cerebras, and local Ollama.

If a provider hits a rate limit or quota, FreeRide automatically fails over to the next available route. Your agent never sees the error; it just gets a response from a different backend. This lets you maximize free-tier coverage without managing keys or retry logic in each tool.

Why FreeRide?

Modern AI coding agents lock you into their vendor's inference backend. If you want to use free-tier models from Groq or OpenRouter, you need a separate tool or manual key swapping. If one provider rate-limits you, your agent stops working until the quota resets.

FreeRide solves this by giving every agent a single OpenAI-compatible endpoint that handles failover, key rotation, and provider selection automatically. You bring your own keys (BYO), configure them once, and every agent you run gets instant access to all of them with zero code changes.

Core principles

Local-first

Requests go from your machine directly to the provider you configured. FreeRide does not sit in the middle, proxy your traffic, or see your prompts. It's a local daemon that lives on your machine and exits when you close it.

BYO keys

FreeRide does not issue keys, create accounts, or host a shared key pool. You bring your own free-tier API keys from the providers you trust. Add one key to start, add more when you want better coverage. Your keys stay local and never leave your machine.

Free-only

The router stays inside the free paths you gave it. Exhausted routes fail clearly instead of billing quietly. If all providers are rate-limited or out of quota, you get an error, not a surprise charge. Free-tier limits are real constraints, and FreeRide respects them.

How it works

FreeRide implements the OpenAI Chat Completions API and Anthropic Messages API. When you make a request, the gateway:

  1. Receives your request at localhost:11343/v1/chat/completions
  2. Selects the first available provider based on model and quota state
  3. If the provider returns 429 (rate limit) or 403 (quota), tries the next key/provider
  4. If the model isn't found, advances to the next provider that supports it
  5. If all providers fail, returns 503 Service Unavailable with a Retry-After header
  6. On success, returns the response with an X-FreeRide-Provider header showing which backend fulfilled it

The failover logic is transparent to your agent. From the agent's perspective, it's just talking to a single OpenAI-compatible endpoint that never rate-limits (until all providers are exhausted).

Open source

FreeRide is MIT-licensed and developed in the open. The gateway daemon lives in the FreeRideV3 repository on GitHub. You can read the code, open issues, submit PRs, or fork it for your own use case. The Python package is published to PyPI as freeride-gateway.

This site (free-ride.xyz) is a static Next.js 14 export hosted on Cloudflare Pages. Its source is in freeride-web. The live traffic stats and installer are served by a Cloudflare Worker at api.free-ride.xyz.

When to use FreeRide

FreeRide is designed for developers who:

  • Run coding agents (Claude Code, OpenAI Codex, Gemini CLI, Aider, Continue, etc.) and want to maximize free-tier inference without switching tools
  • Have free-tier API keys from multiple providers (Groq, OpenRouter, NVIDIA NIM, etc.) and want to use them all seamlessly
  • Need automatic failover when one provider hits rate limits or quota caps, without manual key rotation
  • Want a local-first solution that doesn't require trusting a third-party proxy or hosted key service
  • Prefer open-source tools where you can read the code, run it yourself, and understand exactly what it does

If you need production-grade inference with SLAs, dedicated quota, or guaranteed uptime, use a paid service. FreeRide is optimized for free-tier development workflows where occasional rate limits are acceptable and cost is a hard constraint.