Kwaak is an open-source tool designed to automate software development tasks using AI agents, targeting developers who want to offload routine coding, testing, and documentation improvements. It allows users to run multiple autonomous agents concurrently from their terminal, aiming to reduce tech debt while developers focus on core feature development.

How It Works

Kwaak indexes the codebase using DuckDB for context, then provides a chat-like interface for users to interact with AI agents. Agents operate within Docker containers, executing code and commands in a sandboxed environment to ensure system security. The tool supports various LLM providers (OpenAI, Ollama, Anthropic, Azure) and can integrate custom tools via the Model Context Protocol (MCP). It handles code changes via patch-based editing by default and can automatically lint, commit, push, and create pull requests.

Quick Start & Requirements

• Installation: Pre-built binaries via Homebrew (brew install bosun-ai/tap/kwaak), curl installer, cargo binstall, or pacman on Arch Linux.
• Prerequisites: Docker installed and configured with an Ubuntu-based Dockerfile in the project root that includes git. API keys for LLM providers and optionally a GitHub token for pull requests.
• Setup: Run kwaak init in the project directory to create a configuration file.
• Running: Execute kwaak in the project root. Initial indexing may take time.
• Docs: Getting Started, Configuration

Highlighted Details

• Supports parallel execution of multiple autonomous AI agents.
• Offers a chat-like terminal interface for interaction.
• Integrates with OpenAI, Ollama, Anthropic, Azure, and OpenRouter for LLMs.
• Supports Python, TypeScript/JavaScript, Go, Java, Ruby, Solidity, C, C++, and Rust.

Maintenance & Community

• Project is part of the bosun.ai initiative.
• Community support available via Discord.
• Actively seeking contributions.

Licensing & Compatibility

• MIT License.
• Permits commercial use and linking with closed-source projects.

Limitations & Caveats

Kwaak is considered alpha software with ongoing development and potential for breaking changes. Performance may be impacted by LLM tier limits (Tier 1 accounts are not recommended) or insufficient Docker resource allocation. Index corruption can occur, requiring re-indexing.