Specialist sub-agents for AI tasks
Let an AI feature ask a configured specialist to complete a limited task and return its result privately.
What it is
A Sub-agent function lets one AI feature ask another configured feature for help. The specialist uses its own instructions, functions and skills, then returns a result to the calling AI. The caller remains responsible for the reply in the conversation.
Who uses it and why
Users with Settings and permission to manage AI, or Developer, configure specialists for tasks that need different instructions or tools. For example, the main assistant can ask a product specialist to research a request before composing its answer.
Where to find it
Create or edit a function at Settings › AI › Functions and select Sub-agent as its Type. Assign it to the caller through Functions at Settings › AI › Features.
How it works
- The calling AI supplies an objective and, when configured, selected context.
- INSIDE starts a separate specialist session using Target AI Feature.
- Context Policy controls what background information is included. None adds no background; Summary adds a summary; Conversation adds conversation context; Selected context uses supplied context.
- The specialist runs within its configured limits and returns a completed result, a request for input or a failure.
- The caller uses that result to continue its work or ask the required question.
The specialist is instructed to return privately rather than send messages or transfer the conversation directly. Its configured functions can still perform actions, so select them deliberately.
Configure it
Use Settings with permission to manage AI, or Developer.
- Configure the specialist as an Ask AI, Copilot or Autopilot feature on the same website.
- Create a Sub-agent function and select Target AI Feature.
- Choose Context Policy explicitly and set the execution limits.
- Add Result Schema (JSON) if the caller needs a structured result.
- Save and assign the function to the calling feature.
The specialist uses the target's tools; it does not simply inherit all the caller's functions. The originating feature's applicable restrictions still govern the delegated work.
Settings
| Setting | Where | Applies to | Default | Who can change it | What it does |
|---|---|---|---|---|---|
| Context Policy | Settings › AI › Functions | website | None in the new-function form; Summary if omitted | Settings role, Developer role | Controls whether the specialist receives no context, a summary, conversation context or selected context. |
| Maximum Iterations | Settings › AI › Functions | website | 6 | Settings role, Developer role | Limits the specialist's model iterations to a value from 1 to 20. |
| Maximum Tool Calls | Settings › AI › Functions | website | 12 | Settings role, Developer role | Limits the specialist to 0 to 100 tool calls. |
| Maximum Tokens | Settings › AI › Functions | website | Not set: no additional specialist token limit | Settings role, Developer role | Adds a specialist token limit when supplied, from 1 to 2000000. |
| Timeout (seconds) | Settings › AI › Functions | website | 45 seconds | Settings role, Developer role | Limits the specialist request to between 1 and 600 seconds. |
| Target AI Feature | Settings › AI › Functions | website | Not set: select a specialist feature before saving | Settings role, Developer role | Selects the specialist feature on the same website. |
| Result Schema (JSON) | Settings › AI › Functions | website | Not set: no structured result schema is required | Settings role, Developer role | Optionally requires a structured result matching your schema. |
Fields
| Field | Where | Required | What it is for |
|---|---|---|---|
| Type | Settings › AI › Functions | Yes | Select Sub-agent to configure a specialist invocation. |
Permissions
| Permission | Lets you |
|---|---|
| Settings | Configure specialist functions when AI is allowed under Settings Permissions. |
| AI | Allows AI configuration within Settings Permissions. |
| Developer | Configure functions and their feature assignments. |
Limits and good to know
- A feature cannot invoke itself. A specialist cannot invoke another specialist.
- The objective is required and cannot exceed 8,000 characters. Returned reply text cannot exceed 100,000 characters.
- A structured result that fails the configured schema is returned as a failure.
- An identical failed request is suppressed for the rest of the calling execution.
- A specialist can be invoked while its Enabled switch is off. That switch is not a reliable way to prevent delegated use; remove its assignment from the caller.
- Requests needing input return a question to the caller rather than opening a separate customer conversation.
Depends on and works with
+1 more · this feature uses
- AI accounts and shared defaultsRequires
The specialist needs a configured AI account to run its work.
View all connections as a list
This feature uses
- Roles and permissionsConfigured by
Roles control function and feature configuration.
- AI feature settings and change historyConfigured by
The target is a configured AI feature on the same website.
- AI instructions and templatesConfigured by
The specialist uses its target feature's instructions.
- AI run limitsRequires
Specialist work runs within execution limits.
- AI functions and required inputsRequires
Specialist invocation is configured as a function available to the calling feature.
- Reusable AI skills and instructionsEnhances
Target-owned skills can supply specialist instructions and functions.
- AI conversation plans and task progressEnhances
A specialist plan requesting input produces a question for the caller.
- AI answer citations and summariesEnhances
Specialist results can include source citations.
- AI accounts and shared defaultsRequires
The specialist needs a configured AI account to run its work.
Used by
- AI functions and required inputsEnhances
Sub-agent functions request work from another configured AI feature.
- AI usage and estimated costEnhances
Specialist invocations can retain originating and executing feature context with their usage.
- AI platform execution limitsConfigured by
Specialist invocations can impose their own smaller execution allowances.
FAQ and troubleshooting
Why is the target unavailable?
It must be an available Ask AI, Copilot or Autopilot feature on the same website. Check the selected target and its account.
Why did the specialist fail without an answer?
Check the objective, timeout, execution limits and result schema. The caller receives a failure if the specialist cannot complete the request.
Why is no second specialist called?
Delegation is limited to one specialist level.
Why did disabling the target not stop its use?
A specialist can run independently of automatic feature selection. Remove the Sub-agent function from the caller or the skill that supplies it.