free-code is a modified build of Claude Code, a terminal-native AI coding agent.
The project addresses the desire for an unrestricted version of Claude Code by removing telemetry, stripping security-prompt guardrails, and enabling experimental features. It works by taking the publicly available source of Claude Code and applying three categories of modifications: eliminating outbound telemetry endpoints including OpenTelemetry, GrowthBook, and Sentry reporting; removing system-level instruction injections that constrain Claude's behavior beyond the model's own safety training; and unlocking feature flags that ship disabled in the public release. The result is a single binary with no external callbacks, installable through a one-liner that checks system requirements, installs dependencies if needed, and builds with all experimental features enabled.
Developers should choose this tool if they want to use Claude Code without telemetry collection or additional prompt-level restrictions, and if they are willing to build from source rather than use a prebuilt binary. It suits projects where developers prefer to authenticate directly with their chosen model provider through the `/login` command rather than rely on Anthropic's infrastructure. The tool is distributed with an IPFS mirror option for alternative access.
The project maintains a clean, buildable fork structure with clear separation of changes from the upstream snapshot. Development activity shows active maintenance of the build system and feature flag audit, with documentation of all modifications applied and a structured approach to tracking which experimental features compile cleanly.
