After six months in private beta with 300 developers across 18 countries, woozcode is now available to everyone.
What we built
Woozcode is a local plugin that compresses AI prompts before they leave your machine. It reduces token usage by 40-60% on typical development workloads with no change to model, workflow, or output quality.
The v1.0 release includes: - wooz compress: core prompt compression - wooz context: intelligent file context selection - wooz diff: incremental prompt diffing for multi-turn sessions - Local daemon with REST API and Node.js SDK - Usage dashboard with per-session analytics - Support for OpenAI, Anthropic, Mistral, and any OpenAI-compatible endpoint
What we learned in beta
The beta taught us two things we had not expected.
First, the quality concern is smaller than developers expect. We heard from beta users that they expected compression to degrade output quality noticeably. In practice, at balanced compression level, the difference is imperceptible on the vast majority of tasks. The model is robust to missing whitespace, duplicate imports, and commented-out dead code.
Second, the savings are larger than the headline numbers suggest. A 60% reduction in prompt tokens reduces both input and output costs, since shorter prompts correlate with shorter outputs in many task types. Several beta users reported total bill reductions of 70-75%.
Where we go next
The roadmap for v1.1 includes the semantic cache (currently in Pro beta), a VSCode extension for inline usage visibility, and support for streaming compression so the daemon can work on tokens as they arrive rather than waiting for the full prompt.
Woozcode is free for individuals. Pro plans are available for teams with higher volume and advanced tooling needs. We are grateful to every developer who tested the beta, filed issues, and pushed the tool to its limits.