Crystal is a desktop app for managing multiple AI coding sessions in parallel. It creates isolated git worktrees for Claude Code and Codex so you can explore different solutions simultaneously, review diffs, and squash commits cleanly.
The app provides build/test hooks, a diff viewer, and multi-platform support (macOS DMG, Windows source build, Linux). It’s aimed at developers who want to compare agent outputs, keep histories tidy, and ship with confidence.
Use Cases:
Get the Latest Release Here
Build (https://github.com/stravu/crystal/actions/workflows/build.yml/badge.svg) Quality (https://github.com/stravu/crystal/actions/workflows/quality.yml/badge.svg) Join our Discord (https://img.shields.io/badge/Join%20our-Discord-5865F2?style=for-the-badge&logo=discord&logoColor=white)
Crystal lets you use AI on isolated copies of your code so you can work on multiple tasks instead of waiting for your agents to finish.
Run one or more sessions with Claude Code, Codex, or both
Rich output for your agents
Run your code and test before merging
https://github.com/user-attachments/assets/5ca66e5b-8d05-4570-8417-5e8dcd7726ef
@openai/codex or Homebrew) with ChatGPT account or API keyCreate a new project if you haven't already. This can be an empty folder or an existing git repository. Crystal will initialize git if needed.
For any feature you're working on, create one or multiple new sessions:
As sessions complete:
When everything looks good:
macOS: Download Crystal-{version}.dmg from the latest release (https://github.com/stravu/crystal/releases/latest)
Windows: Windows is supported but requires building from source
brew install --cask stravu-crystal
# Clone the repository
git clone https://github.com/stravu/crystal.git
cd crystal
# One-time setup
pnpm run setup
# Run in development
pnpm run electron-dev
# Build for macOS
pnpm build:mac
# Build for Windows
pnpm build:win
# Build for Linux
pnpm build:linux
We welcome contributions! Please see our Contributing Guidelines for details.
If you're using Crystal to develop Crystal itself, you need to use a separate data directory to avoid conflicts with your main Crystal instance:
# Set the run script in your Crystal project settings to:
pnpm run setup && pnpm run build:main && CRYSTAL_DIR=~/.crystal_test pnpm electron-dev
This ensures:
~/.crystal_test for its data~/.crystal To use Crystal with cloud providers or via corporate infrastructure, you should create a settings (https://docs.anthropic.com/en/docs/claude-code/settings) file with ENV values to correctly connect to the provider.
For example, here is a minimal configuration to use Amazon Bedrock via an AWS Profile:
{
"env": {
"CLAUDE_CODE_USE_BEDROCK": "1",
"AWS_REGION": "us-east-2", // Replace with your AWS region
"AWS_PROFILE": "my-aws-profile" // Replace with your profile
},
}
Check the deployment documentation (https://docs.anthropic.com/en/docs/claude-code/third-party-integrations) for more information on getting setup with your particular deployment.
For a full project overview, see CLAUDE.md. Additional diagrams, database schema details, release instructions, and license notes can be found in the docs directory.
Crystal is open source software licensed under the MIT License.
Crystal includes third-party software components. All third-party licenses are documented in the NOTICES file. This file is automatically generated and kept up-to-date with our dependencies.
To regenerate the NOTICES file after updating dependencies:
pnpm run generate-notices
Crystal is an independent project created by Stravu (https://stravu.com/?utm_source=Crystal&utm_medium=OS&utm_campaign=Crystal&utm_id=1). Claude™ is a trademark of Anthropic, PBC. Codex™ is a trademark of OpenAI, Inc. Crystal is not affiliated with, endorsed by, or sponsored by Anthropic or OpenAI. This tool is designed to work with Claude Code and Codex, which must be installed separately.
Made with ❤️ by Stravu
Official Eden emulator release mirror providing multi-arch Debian/Ubuntu packages and RC builds with direct download links and community support references.
Local GUI and WebUI for multi-agent AI (Gemini/Claude/Codex/Qwen), with MCP tool management, remote access, persistent chats, and bundled file/image/Excel helpers.
MongoDB-compatible document database on PostgreSQL with BSON support, gateway translation, and full-text/geospatial/vector features under an MIT license.
Cross-platform Kotlin/JVM terminal and SSH client with SFTP GUI, host tagging, X11 forwarding, plugin ecosystem, and installs via brew/winget or direct downloads.
Local-first AI notebook for files, web pages, PDFs, and videos with smart notes, citations, and Surflet applets—runs on macOS/Windows/Linux with your choice of local or cloud LLMs.
Lightweight self-hosted server monitoring with web dashboards, low-overhead agents, Docker/one-click/binary deploys, and host tagging with simple credential setup.