Subagents
When a request is complex, Blaze’s AI can decide on its own to spin up a “subagent” — a separate instance of itself — and hand off part of the investigation or work to it. This page explains what’s happening when you see a subagent running during a conversation.
1. What is it?
A subagent is a mechanism where Blaze’s main AI delegates part of a complex request to a separate AI instance.
For example, for a request like “look into how several objects are configured, then implement the change”, Blaze may split the work like this:
-
Hand off the investigation to a subagent, which checks the relevant files and metadata
-
Take the results back and use them in the main AI’s implementation or response
Whether a subagent is used is decided automatically by the AI — there’s nothing you need to trigger or operate yourself.
Multiple subagents can work in parallel. Investigations and tasks that can proceed independently can run at the same time. For example, subagents can investigate different objects while the main AI brings their findings together.
Up to 20 subagents can work at once in a conversation. Additional work waits for an available slot before starting. No settings change is needed.
Subagents can also continue working in the background independently of the main AI. Their results are delivered to the main AI when they finish and can be used for further work.
2. What you’ll see in the conversation
While a subagent is running, it appears as a block in the timeline, just like any other tool execution (file edits, running a command, and so on).
-
The block’s title shows a short description of what it’s investigating or doing
-
Press the subagent’s name or progress preview to open its messages and tool activity in a tab on the right.
If you notice extra blocks you don’t recognize, there’s no need to worry — Blaze is simply splitting the work behind the scenes.
3. Safety
-
Subagents follow the mode selected in the main conversation. See Approval Mode for details about operation confirmations.
-
A subagent can call further subagents, but there’s a limit to how deep this nesting can go — it can’t grow without bound
4. Settings
There’s no setting to turn subagents on or off individually. The AI decides automatically whether to use one, based on the request.
If the behavior isn’t what you expected, or the investigation or changes seem to be going further than intended, you can add follow-up instructions mid-conversation to narrow the scope of the work.
5. Stop subagents
The stop button in the main conversation stops the main task. Subagents continue working.
To stop one subagent, press its name or progress preview in the conversation to open its tab, then press the stop button at the bottom-right of that tab’s input. Work delegated by that subagent is stopped too.
To stop all subagents in a conversation, press Esc twice in quick succession in the conversation view. Press “Stop all” in the confirmation to stop running and waiting subagents in that conversation, including work they delegated. Work in other conversations is unaffected.

6. Review progress after restarting
Subagent progress and results are saved. After restarting the app, press the subagent in the conversation to review them.
Work interrupted by a restart does not keep running automatically. To continue, send a follow-up instruction from the input in the subagent’s tab.