Sub-Agents You Can Open and Read

When the Mythex agent splits a complex task into nested runs, each shows up in chat as a row you can open and read. How it works and what you control.

Mythex Team · 2026-09-29 · 3 min read

For complex tasks, the Mythex agent can hand parts of the work to nested agent runs — subagents — that investigate or build in parallel. Those runs aren't hidden. Each one appears in your chat as an Agent row showing what it's doing, and you can click it to read its full transcript in a read-only tab.

What it does

A subagent is a second agent run that the main agent starts to take on one piece of a larger job. Say you ask for a change that touches several parts of an app: the main agent might send one subagent to look into how something works today, while it carries on with the rest.

While that nested work runs, an Agent row appears in the transcript. It tells you what the subagent is doing and how many tools it has used so far. When you want the detail, click the row and the run's transcript opens in its own tab.

A subagent's job is either to investigate — look into something and report back — or to implement part of the change. Either way it runs inside the task you gave the main agent, not as a separate conversation you have to manage.

If you're new to the idea of an agent that uses tools on its own, what an AI agent is is a good short primer.

Why we built it this way

An agent that quietly spins off helpers can feel like a black box: something happened, and you're asked to trust the summary. We'd rather you could check.

So every nested run is readable. If the result surprises you, you can open the subagent's transcript and see what it looked at and what it did, instead of guessing. And because the transcript sits in its own tab, the main conversation stays uncluttered — you read the detail only when you want it.

How to use it

There's nothing to set up. Work with the main chat the way you normally would, and subagents appear when the agent decides a task needs them.

Watch the Agent row. While it's running, it shows what the subagent is doing and its tool count.
Click the row to open the run's transcript in a tab.
Read it. Subagent tabs are read-only — you don't type into them.
Steer from the main chat. If you want to change direction, do it in the parent conversation: send a follow-up, switch between Build, Plan and Ask, or press Stop.
Close the tab with × when you're done. The main thread and your project files stay exactly as they were.

Plan tabs work the same way

Plans get the same treatment. When the agent finishes a plan, a plan card appears in chat with the todos already listed. View Plan opens the plan in a read-only tab, so you can read the whole approach through without it scrolling away. Build, on the card or in that tab, starts implementing it. You don't need to find a file path; the card and the tab have everything. There's more on that flow in plan before you build.

Limits

You can't start a subagent yourself. The main agent decides when to use them. If you want a task split up, describe the task; the decision is still the agent's.
Subagent tabs are read-only. You watch and read them; all steering happens in the parent chat.
Closing a tab doesn't undo anything. It just stops you watching. To roll back work, use the checkpoint on a reply in the main chat — see Chat.

Subagents — the docs page for this feature
Chat — threads, tabs, Stop, and the plan card
Iterate without rebuilding — how to steer a project in small steps