Every turn, coding agents resend files, diffs, instructions, and logs they have already seen. Halv removes that wasted context locally before each request reaches the model. You can keep the same CLI, login, model, and subscriptions for Claude Code, Codex, Kimi, and GLM while getting far more work from your plan. Halv also includes split terminals, multi-agent workflows, local sandboxing, a Crux code index, command filtering, and a live savings meter with a receipt for every answer. View startup
Want to discover more AI signals like this?
Explore Steek