Captain Code

Local agent orchestration tool that selects and manages multiple AI coding agents

Coding & Development Coding AssistantsContact for Pricing

About Captain Code

Captain Code is a local agent orchestration tool designed for users who already operate multiple AI coding agents. It acts as a central coordinator that classifies tasks and routes them to the most suitable model or team from existing subscriptions, API keys, and connected open-source models. The tool eliminates the need for manual model selection by automatically choosing between providers such as Claude, Codex, Cursor, Grok, and others based on task requirements and constraints. Captain Code supports both routine and demanding tasks, allowing users to set preferences for cost efficiency, speed, or frontier-level performance. It includes features for quota management, workflow automation, and mid-turn steering to adjust running agents. The system maintains unified memory and ledgers to prevent context loss and supports inspectable decision-making through commands like /captain why. Users can also enforce policies such as open-weight model constraints or deterministic serving tuples.

Key features

Agent routing and model selection
Quota and budget management
Mid-turn steering with /btw
Workflow automation with /workflow
Cost-saving preferences with /cheap
Team-based task execution with /team
Frontier-level performance with /frontier
Open-weight model constraints with /oss

Use cases

Managing multiple AI coding agents for large projects
Automating routine coding tasks with cost-efficient models
Coordinating specialized teams of agents for complex tasks

Pros

Centralizes routing across multiple AI coding agents
Supports both subscription-based and open-source models
Includes quota management and automatic handoffs when limits are hit
Allows mid-turn steering and policy-based model selection
Provides inspectable decision logs for transparency

Cons

Requires existing subscriptions or API keys for full functionality
Limited to local agent orchestration without cloud-based coordination
Command-based interface may require learning specific syntax

Frequently asked questions about Captain Code

What is Captain Code and who is it designed for?

Captain Code is a local agent orchestration tool designed for builders who already run multiple AI coding agents. It acts as a central coordinator that classifies tasks and routes them to the most suitable model or team from existing subscriptions, API keys, and connected open-source models.

How does Captain Code select the best model for a task?

Captain Code automatically classifies every prompt and routes it to the right model or team based on task requirements, constraints, and user preferences for cost efficiency, speed, or frontier-level performance. Users can also enforce policies like open-weight model constraints or deterministic serving.

Can Captain Code work with open-source models?

Yes, Captain Code supports open-source models and allows users to connect them alongside proprietary models. Open-source models can be added through providers like OpenRouter, Hugging Face, or NVIDIA NIM.

What features does Captain Code offer for managing quotas and workflows?

Captain Code includes features for quota management, workflow automation, and mid-turn steering to adjust running agents. It records reset times and hands off tasks to another available worker when limits are hit, preserving partial output and context.

How can users inspect or steer the model selection process in Captain Code?

Users can inspect every model selection using the /captain why command. They can also steer the process mid-turn with commands like /frontier, /team, /repeat, /workflow, /cheap, /parallel, /oss, /deterministic, and /interrupt.

What types of models and providers does Captain Code support?

Captain Code supports a variety of models and providers, including Claude, Codex, Cursor, Grok, OpenRouter, NVIDIA NIM, Hugging Face, and others. It also includes built-in legs for local subscription CLIs and remote models through OpenCode.