squeez

Overview

squeez is a sophisticated token optimization tool that reduces context usage in AI CLI environments through multiple compression techniques. It implements hooks for major AI hosts including Claude Code, Copilot CLI, OpenCode, Gemini CLI, and Codex CLI, offering up to 95% bash compression, cross-call deduplication, and adaptive intensity based on session pressure. The project includes a built-in MCP server with 13 read-only tools that allow LLMs to query session memory directly, providing a rich context engine that tracks file access patterns and tool usage.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• squeez_filter

  Compress input from stdin using squeez's filtering algorithm

  Try:Filter and compress this output for meTry:Apply squeez compression to this data
• squeez_compress_md

  Compress markdown files in-place using squeez's algorithm

  Try:Compress this markdown file to reduce token usageTry:Make this CLAUDE.md file more concise
• squeez_benchmark

  Run benchmarks to test squeez's compression effectiveness

  Try:Run squeez benchmarks to see compression performanceTry:Test how well squeez handles different scenarios
• squeez_protocol

  Print the auto-teach payload that helps LLMs learn squeez's protocol

  Try:Show me squeez's protocol markersTry:Get the auto-teach payload for squeez
• squeez_calibrate

  Auto-tune squeez configuration based on benchmark results

  Try:Optimize squeez configuration for my systemTry:Run auto-calibration for squeez
• squeez_budget_params

  Output JSON budget parameters for a specific tool

  Try:Get budget parameters for the read toolTry:Show token budget for grep operations
• squeez_agent_costs

  Track and analyze sub-agent token costs

  Try:Show me my sub-agent token usageTry:Analyze the cost of my agent calls
• squeez_session_summary

  Get a structured summary of the previous session

  Try:What did I work on in the last session?Try:Show me the previous session summary
• squeez_token_tracking

  Track token usage across tool calls

  Try:Show me my current token budgetTry:How many tokens have I used this session?
• squeez_cross_call_dedupe

  Check for and remove redundant content across tool calls

  Try:Find duplicate content in my recent commandsTry:Remove redundant outputs across calls
• squeez_memory_file_stats

  Analyze memory files for compression opportunities

  Try:Show me compression stats for my CLAUDE.mdTry:Analyze my memory files for token savings
• squeez_config_analyzer

  Analyze squeez configuration for optimization opportunities

  Try:Review my squeez configuration for improvementsTry:Optimize my squeez settings

Note: Tool names inferred from command examples and documentation. While the README mentions exposing 13 read-only tools via MCP server, specific tool names aren't clearly listed in a dedicated tools section. Tools were created based on the comma

Comparable tools

Installation

curl -fsSL https://raw.githubusercontent.com/claudioemmanuel/squeez/main/install.sh | sh

npm install -g squeez
npx squeez

cargo install squeez

squeez mcp

{
  "mcpServers": {
    "squeez": {
      "command": "squeez",
      "args": ["mcp"]
    }
  }
}

FAQ

Which AI CLI hosts does squeez support?

squeez supports Claude Code, GitHub Copilot CLI, OpenCode, Gemini CLI, and OpenAI Codex CLI, with additional support for Pi through a TypeScript extension.

How much token compression can I expect?

squeez can reduce bash output by up to 95%, with compression rates varying by scenario. Benchmarks show reductions between 99% for repetitive outputs to 22% for markdown files.

Compare squeez with