Trace Commons asks devs to donate coding sessions for open AI training

A Reddit user has launched Trace Commons, an initiative to crowdsource coding agent traces into a public dataset under the permissive CC-BY-4.0 license. The motivation stems from concerns that Anthropic (Claude Code) and OpenAI (Codex) are amassing enormous amounts of proprietary coding session data, which could create an insurmountable advantage for their models in code generation and understanding. Without an open alternative, the user warns, open-weight and open-source models will fall further behind.

The project invites developers to donate their coding agent sessions—recordings of how they interact with AI coding assistants—to a central repository hosted on Hugging Face. The collected data is intended to be used for training open models, helping level the playing field. Contributors retain attribution rights per the CC-BY-4.0 license. The initiative is in its early stages, and the creator is soliciting feedback from the community to refine the dataset format and collection process.