edouard-claude/snip

CLI proxy that reduces LLM token usage by 60-90%. Declarative YAML filters for Claude Code, Cursor, Copilot, Gemini. rtk alternative in Go.

What it solves

AI coding assistants often waste significant amounts of context window tokens on verbose shell output (e.g., hundreds of lines of passing tests or full git commit metadata) that provides little to no useful signal for the LLM. snip reduces this noise by filtering and condensing shell output before it reaches the AI tool.

How it works

snip acts as a CLI proxy that intercepts shell commands run by AI agents. It uses declarative YAML pipelines to match specific commands and apply transformations—such as removing redundant lines, truncating output, or formatting results into a concise summary. If no filter matches, the command passes through unchanged. It integrates with major AI tools via native hooks (like Claude Code and Cursor) or prompt injection (like Gemini and Windsurf).

Who it’s for

Developers using AI coding agents (such as Claude Code, Cursor, Copilot, Aider, or Windsurf) who want to reduce token usage, lower costs, and prevent the LLM's context window from being filled with irrelevant logs.

Highlights

  • Broad Integration: Supports a wide array of AI tools including Claude Code, Cursor, GitHub Copilot, Gemini CLI, and others.
  • Extensive Built-in Library: Ships with 132 YAML filters covering common tools like Git, Go, Rust, Python, Docker, and Kubernetes.
  • Declarative Pipelines: Uses YAML data files rather than compiled code for filters, making them easy to extend and the engine extensible.
  • Token Tracking: Includes a dashboard (snip gain) to track and report total tokens saved and efficiency gains.
  • Flexible Actions: Provides 20 different pipeline actions, including regex extraction, JSON processing, and Go template formatting.

Related

  • Project
    rtk-ai/rtk

    A high-performance CLI proxy that compresses bash output for AI agents, reducing input tokens by up to 90% to lower costs and save context space.

  • Project
    fkirc/attranslate

    A CLI tool that helps AI agents translate localization files efficiently by syncing only missing strings and preserving manual corrections.

  • Project
    gglucass/headroom-desktop

    A cross‑platform desktop tray app that intercepts Claude Code, Codex, OpenCode and Grok Build requests, compresses prompts and tool output locally, and cuts token usage by about 50 % while keeping the original results.

  • Project
    Fast-Editor/Lynkr

    Lynkr is a Node‑JS gateway that sits between AI coding tools and LLM providers. It compresses JSON tool results, caches similar queries (sub‑300 ms hits with zero token cost), and automatically routes each request to the cheapest‑enough model across 14+ providers. No code changes are needed—just point your tool at `http://localhost:8081/v1` or use the `lynkr wrap` command. The project includes a dashboard, real‑time streaming, optional Rust‑based code analysis, and extensive configuration via a `.env` wizard.

  • Project
    Green-PT/honey-for-devs

    Honey is a cross-tool coding skill that optimizes AI agents by reducing unnecessary code and verbose prose, resulting in lower token usage and costs while maintaining code quality.