paoloanzn/free-code

free-code is a modified, buildable version of Claude Code, the terminal-based AI coding assistant made by Anthropic. The original source code became publicly accessible in March 2026 when it was found embedded in the npm distribution, and this fork takes that code and makes three specific changes before packaging it as a single installable binary. The first change is removing all tracking. The original tool sends data back to Anthropic through several channels: usage analytics, error reports, and session information. This build cuts all of those out. You get the same tool, but nothing gets reported back to anyone. The second change is removing extra behavioral restrictions that Anthropic injects into every conversation at the system level. The underlying AI model still has its own built-in safety behaviors, but the CLI itself adds another layer of restrictions on top of those. This fork removes that extra layer, leaving only what the model itself enforces. The third change is unlocking experimental features. The official release ships with 88 feature flags, most of them turned off. This fork enables 54 of those flags that compile cleanly. The full list is in a separate FEATURES.md file in the repository. To use it, you need the Bun JavaScript runtime installed (or the install script will handle that for you). You run one command to install, then use the /login command to connect your account with whichever AI provider you prefer. The tool supports five providers out of the box: Anthropic's own API, OpenAI's Codex models, AWS Bedrock, Google Cloud Vertex AI, and Anthropic Foundry. Switching between them is done by setting an environment variable before you run the tool. You can also build the binary yourself from source if you want to control which feature flags get compiled in.