Free Claude Code setup with local and API models

Free Claude Code (FCC) is presented as a local proxy that connects the Claude Code coding assistant to other model providers. The speaker says it translates Anthropic and OpenAI message formats, including tool calls and thinking blocks, and also supports Codex and Pi. The distinction matters: this setup uses the Claude Code harness with replacement models; it does not provide free access to Claude models.

The tutorial covers installation on Mac and Linux through a shell script, plus a PowerShell installer for Windows. According to the walkthrough, the installer can provision Claude Code and Codex. The proxy listens on port 8082 by default, and its browser admin UI at localhost:8082/admin lets users enter provider keys, select a model and validate settings. Per-tier routing assigns different models and reasoning effort to Opus, Sonnet and Haiku-style requests.

For users who want to run models locally, the speaker names Ollama, LM Studio and llama.cpp, with offline use dependent on the machine's capacity. Models need enough context for the agent's system prompt and tool definitions, as well as reliable tool calling. The speaker warns that very small models can struggle and that output quality depends on the chosen model.

The video also covers editor configuration, Discord and Telegram control, and voice notes through local Whisper or NVIDIA NIM. A sponsored TestSprite segment discusses testing a live app and returning screenshots of failures.