A multi-agent orchestration system, TAKT defines AI agent workflows via YAML, enabling structured, visible, and replayable execution. It targets developers and researchers leveraging AI agents like Claude Code and Codex, offering a robust framework for coordinating complex tasks and integrating AI into CI/CD pipelines.

How It Works

TAKT employs a music metaphor: "pieces" are workflow definitions, and "movements" are individual steps. YAML files declare agent assignments, permissions, and failure handling for each movement, ensuring explicit control over AI agent interactions. This approach makes non-deterministic AI decisions transparent and reproducible, providing essential structure for multi-agent coordination and guardrails for automated pipelines.

Quick Start & Requirements

• Installation: npm install -g takt
• Prerequisites:
  • Provider CLIs (Claude Code or Codex) OR direct API access (Anthropic or OpenAI API Key).
  • GitHub CLI (gh) is required for GitHub Issue integration.
• Costs: Direct API usage incurs costs based on provider pricing; monitor usage, especially in CI/CD.

Highlighted Details

• Supports interactive mode for task refinement via AI conversation or direct execution via --task or --piece flags.
• Seamless integration with GitHub Issues for task definition and execution.
• --pipeline mode enables non-interactive, automated execution suitable for CI/CD, with options for auto-creating Pull Requests.
• Task management features (add, run, watch, list) facilitate batch processing using .takt/tasks/ files.
• Highly extensible with custom YAML "pieces" and Markdown agent prompts.
• Supports parallel execution of sub-movements within a larger movement.

Licensing & Compatibility

• License: MIT.
• Compatibility: The MIT license generally permits commercial use and integration with closed-source projects.

Limitations & Caveats

TAKT coordinates AI agents but does not autonomously decide on project goals; users must provide the task. It is designed for workflow structure across agents, not as a direct replacement for agent-specific extensions (Skills) or parallel agent execution (Swarm). Full automation requires explicit opt-in via pipeline mode.