Crystal is an open-source AI coding harness and desktop-oriented agent platform maintained by Stravu. Designed as an Electron-based application with a dedicated frontend, backend main process, and shared tooling, Crystal enables developers to run autonomous and semi-autonomous coding workflows directly within their local development environment. In daily use, Crystal operates across cross-platform desktop environments (macOS and Windows). It orchestrates tasks across local repositories, utilizing automated test suites, Git operations, and integration with Anthropic Claude models (including Claude Code and Claude Opus) to execute code changes, analyze repository context, and handle multi-step development tasks. Unlike purely terminal-based agents, Crystal integrates a structured frontend interface powered by Electron alongside Node.js-based repository management. It includes custom build and packaging pipelines for distributing standalone desktop packages (DMG for macOS, NSIS installers for Windows) and managing git state during autonomous editing sessions. Crystal is released as open-source software under its GitHub repository (`stravu/crystal`). The project is actively maintained with releases spanning version 0.3.x, relying on external LLM provider API credentials for generative code operations.
Electron-Based Architecture — Combines a desktop frontend UI with a Node.js main process backend for managing coding workflows. - Cross-Platform Support — Provides native build configurations and installers for macOS (DMG) and Windows (NSIS, Visual Studio 2022 build support). - Integrated Claude Model Execution — Integrates with Anthropic LLM backends (such as Claude Code and Claude Opus) for autonomous task generation and diff creation. - Automated Git Integration — Manages branch manipulation, patch creation, and multi-file commit flows with automated error handling for expected Git states. - Local Test Execution — Features an integrated test harness to execute and validate test suites as changes are generated.
Use cases
Use Case: Desktop-Driven AI Coding Workflows Developers can run complex coding sessions through a dedicated Electron application rather than raw CLI interfaces. The tool manages context, executes changes across files, and displays execution steps visually.
Use Case: Automated Bug Fixing and Refactoring Crystal automates code modifications by utilizing LLMs like Claude Opus to inspect local code, generate fixes, run verification tests, and stage clean Git commits automatically.
Use Case: Cross-Platform Development on macOS and Windows Teams working across mixed OS environments can build and package Crystal natively using platform-specific scripts, including macOS DMG creation and Windows NSIS installer compilation.
Who it is for
Crystal is built for developers and teams who prefer a dedicated desktop client environment over pure CLI terminal interfaces for managing AI coding tasks. It is less suitable for headless CI/CD automation pipelines or developers seeking minimal, single-binary terminal-only agents.
Install guide
Installation Clone the repository and install dependencies using Node.js/npm: bash git clone https://github.com/stravu/crystal.git cd crystal npm install To build the desktop application for your platform: bash # For macOS npm run build # For Windows (requires Visual Studio 2022 Spectre-mitigated libraries) npm run build:win:x64
First Run Launch the development application: bash npm start Configure your Anthropic API credentials when prompted in the application settings to enable Claude-backed code generation.
Example Session Open a local repository inside the Crystal interface and issue a task prompt: text "Refactor the authentication module to use async/await and update the corresponding test suite." Crystal will parse the repository context, run existing tests, make file changes, and present a diff for review.
AgentPack — AgentPack is an open-source orchestration harness and configuration management framework developed by Vishal (vishal2612200) for multi-agent coding environments. It standardizes…
LoopTroop — LoopTroop is an open-source AI coding agent harness and orchestration framework maintained by the looptroop-ai organization on GitHub. Built with…
SWE-agent — SWE-agent is an open-source autonomous software engineering agent harness built and maintained by researchers from Princeton University and Stanford University.…
Continue — Continue is an open-source AI code assistant and agent framework originally developed by the Continue team (continuedev) and recently acquired…
Vibe Kanban — Vibe Kanban, originally developed by Bloop AI, is an open-source orchestration harness and visual task board designed to manage and…
Hivelore — Hivelore is an open-source agent harness and contextual memory orchestrator created by Doucs91. It provides persistent project memory, decision tracking,…