This project addresses the limitation of using multiple AI models by creating a structured, heterogeneous team workflow. Instead of comparing identical outputs, AI-Pair assigns distinct roles to different AI models (Claude, Codex, Gemini) to leverage their unique strengths in creation and review. This enables a more comprehensive and nuanced evaluation of code or content, benefiting developers and content creators seeking deeper insights beyond simple answer comparison.

How It Works

AI-Pair functions as a "Claude Code Skill," orchestrating a semi-automatic workflow. A user initiates a task, which a "creator" agent (Claude Code) executes. The output is then passed to two parallel "reviewer" agents: a Codex CLI agent for analytical review (bugs, security, performance) and a Gemini CLI agent for editorial review (readability, style, audience fit). The user retains control at each step, deciding whether to iterate or finalize the work based on consolidated, non-overlapping feedback. This approach maximizes the utility of diverse AI capabilities.

Quick Start & Requirements

• Primary Install/Run: Install the project as a global Claude Code skill:

  git clone https://github.com/axtonliu/ai-pair.git ~/.claude/skills/ai-pair
  

  Restart Claude Code.
• Prerequisites:
  • Claude Code CLI: npm install -g @anthropic-ai/claude-code
  • Codex CLI: npm install -g @openai/codex
  • Gemini CLI: npm install -g @google/gemini-cli
  • All CLIs require authentication configuration.
  • Execution occurs within a terminal environment.
• Links: Repository: https://github.com/axtonliu/ai-pair

Highlighted Details

• Heterogeneous Review: A real-world example demonstrated distinct feedback from each AI: Claude identified source interpretation issues, Codex (GPT) challenged a logical leap, and Gemini suggested audience-fit improvements for an article's opening—none of which overlapped.
• Workflow Evolution: Evolved from the AI Roundtable Chrome extension, adapting the concept of multi-AI discussion to a command-line interface with structured role assignments for practical daily workflows.

Maintenance & Community

The project is currently in an Experimental status, functioning as a public prototype. The primary focus is on demonstrating tool and system integration rather than active codebase maintenance. Contributions are welcome for reproducible bug reports, documentation improvements, and small fixes. Feature requests may not be addressed due to limited maintenance capacity. The author is Axton Liu.

Licensing & Compatibility

• License: MIT.
• Compatibility: The MIT license is permissive, generally allowing for commercial use and integration within closed-source projects without significant restrictions.

Limitations & Caveats

This open-source version exclusively features the Agent Teams mode. Advanced features like manual mode (CLI instances communicating via shared files) and iTerm2 orchestration are part of a separate, privately maintained version requiring specific local setups. The project is experimental and does not yet cover all edge cases. Maintenance capacity is limited, potentially impacting the responsiveness to feature requests.