Run Claude Code with free-tier models from OpenRouter or DeepSeek instead of paying Anthropic per request.
Host your own local AI models with Ollama or llama.cpp and use them through Claude Code without cloud costs.
Access Claude Code's coding assistant through a Discord or Telegram bot for team collaboration.
Transcribe voice notes and send them as coding instructions to your AI assistant.
Requires configuring at least one AI provider (DeepSeek API key, Ollama instance, or local model setup) and Discord/Telegram credentials to see functional routing.
Free Claude Code is a proxy server that lets you use Anthropic's Claude Code developer tool (a command-line AI coding assistant) while routing its AI requests to free or cheaper alternative AI model providers instead of directly to Anthropic's paid API. Here is the problem it solves: Claude Code is built to talk to Anthropic's API, which charges per use. This tool sits between Claude Code and the internet, intercepting those API calls and translating them to work with other services such as NVIDIA NIM, OpenRouter, DeepSeek, LM Studio (a local model runner), llama.cpp (another local runner), or Ollama. This means you can run Claude Code using free-tier models from those services, locally-hosted open models, or whatever alternative you prefer. The proxy handles the protocol translation automatically. From Claude Code's perspective, nothing has changed, it still sends its normal messages. The proxy rewrites those messages into the format each backend expects and translates the responses back. It supports streaming responses, tool use (the ability for the AI to call external functions), and reasoning/thinking blocks. Beyond the terminal, it also offers optional wrappers to use the same AI access through a Discord or Telegram bot, and supports voice note transcription so you can speak your coding instructions. Setup involves running the proxy locally on your machine, setting an environment variable to point Claude Code at localhost instead of Anthropic's servers, and configuring which provider and model to use. The project is written in Python 3.14, uses the uv package manager, and is MIT licensed. The full README is longer than what was provided.
Generated 2026-05-21 · Model: sonnet-4-6 · Verify against the repo before relying on details.