Chisle
Cut your AI coding agent's token bill by half with terse output and compression
Visit chisle.jaypokale.me ↗What is Chisle?
Chisle is a plugin that reduces token usage for AI coding agents across three dimensions: making the model write in terse, senior-developer style; compressing tool output before the model reads it; and teaching the model to fetch only what it needs from files. It works with Claude Code, Pi, OpenCode, Cursor, Windsurf, Cline, and six other agents.
The tool achieves 44% of baseline token usage on coding tasks and 41% on longer work where costs compound. All benchmarks are published with raw transcripts, including the rare cases where it costs more than the bare model. Installation is a single command with zero dependencies, network calls, or LLM calls.
Momentum
Daily activity over the last 30 days: upvotes, unique visits, new GitHub stars and Hacker News points.
How do we know it's vibe coded?
vibetrend discovered Chisle automatically. Evidence: GitHub JayPokale/Chisle; repo has .cursor, .windsurf, AGENTS.md. View source.
Similar vibe-coded apps
Browser and desktop 2.5D + 3D CAD/CAM workspace for CNC machining
In-browser behavior tree execution engine for AI agents at ~60Hz
Encrypted secret manager CLI that hides API keys from AI agents
Local plugin that makes AI coding agents provide evidence and resist flattery
Zero-config Docker ingress router with automatic TLS provisioning
