MiskinSave AI tokens.

Local. No server. Compresses command output + injects caveman-mode prompts.

⚡

Command Filter

Run any command through format-specific compressors. git, cargo, npm, docker — 40+ commands supported. 80%+ token reduction.

🗿

Caveman Mode

Inject "be concise" prompts into your AI tool. Same answers, 65% fewer output tokens. Four compression levels.

🔌

Multi-Agent

Hooks for Claude Code, Copilot, Cursor, Gemini, Codex, OpenCode, Windsurf. One install command.

📊

Analytics

Track token savings per command, per session, per day. ASCII graphs, JSON export. No telemetry, all local.

🦀

Rust

Single ~7MB binary. Zero runtime dependencies. <10ms overhead. Cross-platform Linux/macOS/Windows.

🔒

Privacy First

No telemetry. No accounts. No network calls. Everything stays on your machine.

Quick Start

bash

# Install
npm install -g miskin      # recommended
# or: cargo install miskin

# Set up hooks for your AI tool
miskin init               # Claude Code
miskin init --agent cursor
miskin init --agent copilot

# Restart your AI tool — commands are now auto-filtered
miskin git status          # 90% fewer tokens
miskin cargo test          # failures only
miskin docker ps           # compact table

Approximate Savings

Note: Rough projections, not benchmarks. Your numbers will vary. Run miskin stats to see real savings.

Command	Typical raw	With Miskin	Approx.
`git status`	up to ~2,000	~200	~90%
`cargo test`	up to ~5,000	~500	~90%
`git diff`	up to ~1,000	~300	~70%
`eslint`	up to ~2,000	~400	~80%
`docker ps`	~500	~100	~80%
`npm install`	~1,000	~50	~95%

How It Works

Without Miskin:                     With Miskin:

git status ──▶ shell ──▶ LLM        git status ──▶ miskin ──▶ git ──▶ filter ──▶ LLM
  (~2000 tokens)                      (~200 tokens, 90% saved)

The hook rewrites git status to miskin git status. Miskin runs the real command, compresses output through format-aware filters, and returns compact results to the LLM.