> For the complete documentation index, see [llms.txt](https://doc.aissist.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.aissist.io/capabilities/sub-agents.md).

# Sub Agents

Sub agents give the AI focused instructions, knowledge and actions for one scenario at a time.

*Last updated: October 25, 2026*

A sub agent handles one scenario, such as returns, order tracking or pre-sales questions. This page explains what sub agents are, why they work better than one long set of instructions, and how they behave in a conversation. For the steps, see [Add Sub Agents](https://doc.aissist.io/tutorial/add-sub-agents.md).

### What a sub agent is

A sub agent is a scenario plus what the AI needs for it:

* a **Scenario Description** - when it applies, in plain words, such as "the customer wants to return an item"
* an **Instruction** - how to handle that scenario, step by step
* linked **Assets** and **Actions** - the knowledge and system access the scenario needs
* optional **Context**, **Task**, **Summary** and **Acknowledgement** under **Options**
* conversation **Control**: no reply, stop after this reply, or close

You find them under **Workspace → Sub Agents**.

### Why use sub agents

* **Focused instructions.** Return steps apply only to returns. They do not leak into billing or shipping conversations.
* **Less noise, better answers.** A linked asset is used only while its sub agent is active, so the AI is not distracted by unrelated content. Actions work the same way.
* **Tags for reporting and automation.** The sub agent's **Name** is added as a tag in your help desk when it is triggered. Your team can build views, triggers and routing on it, and Aissist's reports break results down by sub agent.
* **Handover to the right team.** In a gateway's **Handover Rules**, map a sub agent to the team that should receive its conversations.
* **Improve one workflow at a time** without touching the rest.
* **Structured notes for your team.** A **Summary** sends what the AI collected to the help desk as a note.

<figure><img src="https://301624521-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FyKHM836KGDAfAWvw6XL6%2Fuploads%2FdJcHUsv3isJ4pbqRPDjK%2Fagent-platform-tag.png?alt=media&amp;token=f46ced2f-eed5-4e22-a68c-636b3604b42b" alt="A sub agent tag in the agent platform"><figcaption></figcaption></figure>

### How it works

#### Detection

Each message is checked against your sub agents' scenarios. The **Detection Trigger** decides what is checked:

* **Inbound** - triggered by user messages. Use it for what customers ask for: returns, order status.
* **Outbound** - triggered by assistant responses. Use it for outcomes of the AI's reply, such as a transfer to a person.

When a sub agent is detected, its instruction and context join the workspace's for the rest of the conversation. See [Instructions and Information](https://doc.aissist.io/capabilities/instructions-and-information.md) for how the levels combine.

#### Assets and actions

An asset or action linked to a sub agent is used only when that sub agent is active. One linked to no sub agent is available to the whole workspace. Aissist also reads each sub agent's scenario to work out which assets and look-up actions it is likely to need, and adds them. Actions that change data are never added this way; they must be linked by you. See [Actions](https://doc.aissist.io/capabilities/actions.md).

#### Entry and exit actions

On the **Actions** tab, **Entry Actions** run when an inbound sub agent is activated. **Exit Actions** run when an outbound sub agent is activated or an inbound one exits.

#### Controls

The **Control** tab has three rarely used switches: **No Response for Current Interaction** (for spam or out-of-office replies), **No Response After Current Interaction** (reply once, then stop), and **Close** (closes the ticket or conversation in your help desk).

With **Respond Only to Tagged (recognized by sub-agent)** on the workspace **Setting** tab, the AI replies only when a sub agent matches (recommended only for email).

### With Ask Engine

Click **Add Sub Agent** and describe what the sub agent is for, in your own words:

```
Items that arrive damaged - take the order number and photos, explain the replacement window, hand over if they want a refund instead.
```

```
Refund requests - check the order and the refund policy, create a refund ticket in the help desk, and say when to expect the money.
```

Ask Engine searches your assets for what the sub agent will need to tell customers, checks which actions the workspace has and which sub agents already exist, then writes the scenario and instruction and picks the assets and actions to link. You review it and click **Create sub-agent**. Nothing is created before that.

### Best practice

* **One workflow per sub agent.** Split a broad `order_support` into `order_tracking`, `return_or_refund` and `order_cancellation`.
* **Start with one high-volume, low-risk scenario,** such as order tracking. Test it, watch it in live traffic, then add the next.
* **Link only what the workflow needs.** Order tracking needs the shipping action, not the refund policy.
* **Start the instruction with a workflow title, then numbered steps,** with exact handover triggers.
* **Hand over when the workflow needs approval or access the AI does not have,** after collecting the details.

### Related

* [Add Sub Agents](https://doc.aissist.io/tutorial/add-sub-agents.md)
* [Escalation and Handover](https://doc.aissist.io/capabilities/escalation-and-handover.md)
* [Evaluation](https://doc.aissist.io/capabilities/evaluation.md)
