shrew

README

Shrew: Ultra-lightweight Agent

Shrew is a powerful, zero-dependency TUI agent that connects to any API and automates your command line. It's a single static binary written in Go that gives Large Language Models a rich, interactive terminal to perform real-world tasks.

Philosophy

• Lightweight & Fast: A single static binary written in Go with minimal dependencies, ensuring maximum portability and performance.
• Model Agnostic: Connect to any model provider in the cloud (Gemini, OpenAI, Anthropic, etc.) or run locally using Ollama.
• Agentic Execution: The model can execute shell commands directly in your terminal via <run> tags, allowing it to write code, manage files, and run tests.
• Hackable & Extensible: Designed with a clean, easy-to-understand codebase so it can be easily forked, modified, and extended with new skills.
• Extensible Skills System: Teach Shrew new workflows and give it long-term knowledge by adding simple Markdown files to the skills/ directory.

Installation

The easiest way to install Shrew on a new machine is with the one-command installer. This script will automatically detect your OS and architecture, download the latest binary from GitHub Releases, and install it.

curl -fsSL https://raw.githubusercontent.com/Masmedeam/shrew/main/install.sh | sh

Other Methods

From Source

git clone https://github.com/Masmedeam/shrew.git
cd shrew
go build -o shrew main.go
sudo mv shrew /usr/local/bin/

Go Install

go install github.com/Masmedeam/shrew@latest

Configuration

Configure Shrew by creating a .env file in your project root or by setting environment variables.

Google Gemini (Default)

GEMINI_API_KEY=your_api_key

OpenAI or any Compatible API

Use this for providers like DeepSeek, MiniMax, Groq, etc.

SHREW_PROVIDER=openai
OPENAI_API_KEY=your_api_key
SHREW_API_URL=https://api.example.com/v1/chat/completions # The provider's URL
SHREW_MODEL=model-name

Custom Command Bridge

For any other API, you can provide a shell command that takes the conversation history as JSON via stdin and returns the model's response via stdout.

SHREW_PROVIDER=cmd
SHREW_COMMAND="python3 ./my_api_adapter.py"

Usage

Simply run the agent to start an interactive session:

shrew

You can then issue high-level instructions, and Shrew will use its skills, tools, and reasoning to accomplish the task.

License

This project is licensed under the MIT License. See the LICENSE file for details.