Brings advanced multi-model orchestration to the Codex AI framework, enabling users to assign specialized roles (Planner, Advisor, Executor) to different AI models for enhanced task execution. It targets Codex users seeking to improve planning quality, accelerate implementation through parallelism, and optimize the use of premium AI models by leveraging distinct models for distinct stages of a workflow.

How It Works

Codex Orchestration introduces three distinct roles within a Codex task: a Planner to create and refine task plans, an optional Advisor to review and approve plans, and a required Executor to implement the approved plan. Codex acts as the central coordinator, passing work between these roles and verifying results. This approach allows for specialized model application—e.g., using a powerful model for planning and a different one for execution—leading to more robust plans, potential for up to 2x faster parallel implementation, and reduced reliance on single, potentially rate-limited, premium models. The Planner and Advisor can iterate up to five times before execution begins or stops if approval is not reached.

Quick Start & Requirements

Installation involves adding the plugin via the Codex marketplace: codex plugin marketplace add Cjbuilds/Codex-Orchestration codex plugin add codex-orchestration@codex-orchestration

Prerequisites include Python 3.11 or newer. Fable 5 integration requires the official Claude Code CLI and a compatible Claude login; no separate Anthropic API key is needed for Codex. Other models require existing authenticated, compatible providers configured within Codex.

Example setup commands: /codex-orchestration setup planner: Claude Fable 5 High, advisor: GPT-5.6 Sol High, executor: GPT-5.6 Luna Extra High /codex-orchestration setup advisor: Claude Fable 5 High, executor: GPT-5.6 Luna Extra High (uses current Codex model as Planner)

Highlighted Details

• Role Specialization: Assign specific AI models to Planner, Advisor, and Executor roles for tailored task management.
• Parallel Execution: Enables independent implementation tasks to run concurrently, potentially doubling execution speed for suitable workloads.
• Model Limit Optimization: Aims to reduce premium model usage by approximately 40% by distributing tasks across specialized roles.
• Custom Role Creation: Supports defining project or personal roles for models from various providers via /codex-orchestration create project role:.
• Fable 5 Integration: Bundled support for Fable 5 as a Planner or Advisor without requiring direct Anthropic API key configuration in Codex.

Licensing & Compatibility

The project is licensed under the MIT license, which permits permissive use, modification, and distribution, including for commercial purposes. It is designed to integrate seamlessly with the Codex framework.

Limitations & Caveats

Codex remains the ultimate orchestrator, with Planner and Advisor roles operating under its direct instruction and not communicating with each other or Executors independently. Advisor approval serves as a planning gate and does not guarantee successful implementation. Model routing relies on pre-configured providers within Codex; the plugin does not manage credentials or bypass existing permissions. Version 0.5.1 or newer is recommended for reliable Planner assignment.