free-claude-code setup: Ollama and cloud providers

This tutorial configures free-claude-code, a local proxy that routes Claude Code requests to alternative models while retaining its terminal interface. The presenter identifies the repository and its author before following the quick start. He clones the project, edits its hidden .env file, selects a provider and model, and runs the proxy and Claude Code in separate terminals. The demonstration uses macOS; Windows users are pointed toward PowerShell instructions.

The proxy listens on port 8082. OpenRouter and NVIDIA NIM requests still go to external services. The OpenRouter example uses DeepSeek V4 Flash to create and revise a habit tracker. The NVIDIA example configures GLM 4.7; an API error is resolved by turning off Claude Code's fast mode. Those model editions are distinct from older DeepSeek and GLM releases.

For local inference, the presenter installs Ollama, downloads Gemma 4 and selects its local provider configuration at localhost:11434 without an API key. A response through the proxy is slow and heats up his MacBook; these are observations from that machine, rather than general limits of local inference.

Provider logs help verify routing because replies and a manually edited status line can misidentify the backend. The final example uses an Opus model to direct DeepSeek through the proxy while building a calorie tracker. The app demonstrations illustrate the coding workflow; they are not controlled comparisons of model quality.