All docs

Quick Start

Get Squeezr running in under two minutes. Two commands and you're done.

Step 1: Install

npm install -g squeezr-ai

Step 2: Run setup

squeezr setup

That's it. squeezr setup handles everything automatically:

  • Sets ANTHROPIC_BASE_URL and GEMINI_API_BASE_URL to point at the proxy.
  • Installs a shell wrapper in your PowerShell profile (Windows) or ~/.bashrc / ~/.zshrc (Linux/macOS/WSL) so env vars refresh automatically after each squeezr command — no need to restart your terminal.
  • Registers auto-start so the proxy comes back up after a reboot (Task Scheduler on Windows, systemd on Linux, launchd on macOS).
  • Imports the MITM CA certificate so Codex trusts the proxy's TLS (Windows Certificate Store on Windows; ~/.squeezr/mitm-ca/bundle.crt on macOS/Linux/WSL).

Step 3: Start the proxy

squeezr start

Use your coding tool exactly as before — Squeezr compresses transparently.

What happens behind the scenes

Every request passes through a three-layer compression pipeline:

  1. System prompt compression — Claude Code's ~13KB system prompt is compressed once and cached. Subsequent requests reuse the cached version, saving ~3,000 tokens per request.
  2. Deterministic preprocessing — Zero-latency rule-based transforms: ANSI escape codes stripped, repeated stack frames deduplicated, JSON whitespace collapsed, progress bars removed.
  3. Tool-specific patterns — 30+ rules matched against git, test runners, build output, package managers, infra tools, and more. Errors and actionable information are always preserved.

Typical savings

  • Per tool result: 70–95% reduction depending on tool
  • Per session (2 hours): ~200K tokens → ~80K tokens (60% savings)
  • System prompt: ~13KB → ~600 tokens (cached)

Next steps