JayPokale/Chisle

Cut your AI coding agent's token bill on three axes: terse prose, YAGNI-first code, and tool-output compression. Claude Code, Pi, Cursor, Codex, Gemini + 4 more. Zero deps, published benchmarks including the runs it loses.

What it solves

Chisle reduces the token usage and costs associated with AI coding agents like Claude Code, Cursor, and others. It minimizes both the output generated by the model (by removing filler and boilerplate) and the input context (by compressing oversized tool outputs that are otherwise re-billed in every turn of a conversation).

How it works

Chisle operates on two axes:

  1. Output Compression: It injects a ruleset into the AI session that enforces a "YAGNI" (You Ain't Gonna Need It) efficiency ladder, encouraging the model to skip unnecessary abstractions, use standard libraries, and write the minimum code required to solve a problem.
  2. Input Compression: It uses a PostToolUse hook and an MCP proxy to scrub, elide, and deduplicate tool outputs before they re-enter the context window. This prevents the model from being re-billed for massive, redundant logs or data dumps.

Who it’s for

It is designed for developers who use AI coding agents and want to lower their API costs and reduce "noise" in their chat sessions without sacrificing the quality or functionality of the generated code.

Highlights

  • Zero dependencies: A lightweight plugin that installs with a single command.
  • Broad compatibility: Works with a wide range of agents including Claude Code, Cursor, Windsurf, Cline, Gemini, and the Copilot CLI.
  • Transparent benchmarks: The project publishes performance data, including "loss" cases, showing significant reductions in token counts compared to bare models.
  • Configurable: Supports easy activation/deactivation via natural language commands or CLI flags.

Related

  • Project
    Gentleman-Programming/gentle-shell

    gentle‑shell is a terminal‑based coding‑agent built on the Pi LLM platform. It lets you describe a task in natural language, orchestrates the work through a structured “Organic Driven Development” workflow, can delegate to specialised sub‑agents, shows live Git diffs and agent status, and requires you to approve the exact change before committing. Installers exist for macOS, Windows and Linux, and the UI is customisable via a command palette and a `/gentle:customize` panel.

  • Project
    edouard-claude/snip

    A CLI proxy that filters and condenses verbose shell output to reduce LLM token usage by 60-90% for AI coding assistants.

  • Dispatch
    Whistle 16.9 MB Speech‑to‑Text Model: Tiny, Fast, Multilingual On‑Device ASR

    Whistle is a 16.9 MB CPU‑only speech‑to‑text model that runs on the same C++ engine as Needle, supports seven languages, and outperforms Whisper‑base and Moonshine‑tiny v2 on several benchmarks while delivering first‑token latency under 12 ms.

  • Project
    Houseofmvps/codesight

    A zero-dependency tool that pre-compiles codebase structure and documentation into compact context maps and wikis to drastically reduce token consumption for AI coding assistants.

  • Project
    vercel-labs/skills

    A CLI that discovers, installs, updates, and runs reusable prompt bundles (“skills”) for many AI coding assistants (Claude Code, Cursor, OpenCode, etc.), pulling them from any Git repository or direct URL.