Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.zencoder.ai/llms.txt

Use this file to discover all available pages before exploring further.

SubAgents allow you to run a different agent as a sub-process of the current (master) agent. This enables powerful orchestration patterns where tasks are delegated to specialized agents with isolated contexts, different models, or even different CLIs.

​
What is a SubAgent?

A SubAgent is technically an MCP tool call that returns its result back to the master agent. Each SubAgent can have:
  • Different context: The master agent decides what information to pass downstream, keeping the SubAgent focused on its specific task
  • Different tools: Tool availability can vary depending on task requirements (e.g., SubAgents don’t have the Ask tool since UI interaction happens through the master agent)
  • Different model or CLI: Run SubAgents with a different LLM model or even a completely different CLI runtime

​
Why SubAgents?

Specialization

SubAgents don’t carry the “garbage” from the current session, resulting in higher quality outputs for focused tasks

Cost Efficiency

SubAgents can run with simpler, cheaper models for tasks that don’t require the full power of a frontier model

Orchestration

Run different models in parallel for “best of N” scenarios out of the box with ZenCLI

​
How to Use SubAgents

SubAgents are available at three levels:

​
1. General SubAgent Tool

The SubAgent capability is always available with ZenCLI. You can invoke SubAgents directly in chat:
  • "Run /skillX as subagent"
  • "Research Y with subagent using model Z"

​
2. Automatic Usage via System Prompts and Skills

ZenCLI may automatically decide to use SubAgents for certain tasks, such as planning, exploration, or code review.
Check out the /comprehensive-review skill for an example of SubAgent-powered code review that provides thorough, multi-perspective feedback.

​
3. Native UI Features

SubAgents power built-in plugin features like Get Second Opinion, which works similarly to Zenflow but uses SubAgents under the hood.

​
Using Different CLIs for SubAgents

To enable different CLI runtimes for SubAgents, add the following to ~/.zencoder/settings.json: 
{
  "availableRuntimes": ["zencoder", "claude", "codex"]
}
When configured:
  • Adding claude to available runtimes routes all SubAgents using Anthropic models through Claude Code
  • Adding codex enables OpenAI Codex as a SubAgent runtime
  • The root agent always uses the CLI selected in the UI
SubAgents are supported for all CLIs when used through the ZenCLI adapter. One limitation is that parallel SubAgents are not yet supported for Claude Code.