This is a rewrite of the Claw Code project aimed at significantly reducing token usage. In our initial experience, we observed up to 74% token savings, with around 30% on average, without sacrificing quality.

It was adapted from ultraworkers/claw-code.

It is still an early project and under active iteration.

The optimization work was achieved through a combination of Codex, Grok, and the IterX code optimization tool. Together, they were used to iterate on prompt reduction, tool-surface minimization, replay compaction, and measurement-driven refinement across the harness.

[dead]