This project provides a real-time, cross-platform collaboration framework for multiple AI models, including Claude, Codex, and Gemini. It addresses the inefficiency and high token costs associated with traditional AI interaction models by establishing persistent, lightweight communication channels. This allows AIs to maintain their own context, enabling parallel workflows and significantly reducing token overhead, making advanced multi-AI assistance more accessible and cost-effective for developers and researchers.

How It Works

ccb employs a "Persistent Dual-Pane" approach, contrasting with traditional stateless AI execution models. Instead of Claude sending full context repeatedly, ccb creates a persistent, low-overhead channel for small message exchanges. This design allows individual AI models like Codex and Gemini to manage their own context, leading to drastically reduced token consumption (50-200 tokens/call compared to 5,000-20,000 tokens/call) and enabling a parallel, rather than master-slave, work mode. Session recovery is also supported.

Quick Start & Requirements

• Install: Clone the repository and run ./install.sh install. Native Windows users can use install.cmd or install.ps1, selecting a BackendEnv (WSL or Windows).
• Prerequisites: Python 3.10+, and either tmux or WezTerm (WezTerm is recommended for its native split-pane support, especially on Windows). Windows users must install WezTerm via its native .exe installer and configure it to connect to WSL if applicable.
• Run: Use ccb up <model> (e.g., ccb up codex gemini) to start AI sessions. Resume previous sessions with ccb up <model> -r.
• Update: Run ccb update for the latest version.
• Links: Full demo video available on GitHub Releases.

Highlighted Details

• Significantly reduced token costs per call (50-200 vs 5k-20k).
• Optimized send/receive latency for faster AI responses.
• Persistent AI context and session recovery (-r flag).
• Supports collaborative workflows with defined roles for Claude, Codex, and Gemini.
• Includes unique usage examples like "AI Poker Night!" and parallel code reviews.

Maintenance & Community

The project is maintained by bfly123. Users are encouraged to debug issues locally and share git diff with the maintainer. A call for testers ("欢迎大佬们") is present, indicating an active interest in community feedback and contribution. No specific community channels (like Discord/Slack) or formal roadmap are detailed.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Cross-platform support for Windows, macOS, and Linux.

Limitations & Caveats

WSL1 is not supported due to FIFO limitations. Native Windows users require careful configuration of WezTerm and the BackendEnv setting to ensure proper execution and session management.