> ## Documentation Index
> Fetch the complete documentation index at: https://docs.sketricgen.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Agents

> Configure SketricGen agents with clear roles, instructions, models, tools, knowledge, structured outputs, and sub-agent behavior.

Agents are the working units inside SketricGen. A Brand Agent may be a single customer-facing agent with curated tools. An AI Workforce may contain multiple agents that collaborate on internal work.

<span id="overview" />

## Agent Configuration

Each agent can include:

* Name and role
* Instructions
* Model selection
* Tool access
* Knowledge Base access
* Structured output
* Main-agent or sub-agent behavior

## Instructions

Write instructions like a job description:

* What is the agent responsible for?
* What should it avoid?
* Which tools can it use?
* When should it ask a question?
* When should it call a sub agent?
* What does a successful answer look like?

## Models

Agents use the supported model catalog in the product. Available models can include OpenAI, Claude, Gemini, DeepSeek, Qwen, and other approved models. The in-product model picker is the current source of truth.

Start with the default model and settings. Lower randomness for customer-facing Brand Agents, and use stronger reasoning models for AI Workforce workflows that involve research, analysis, code, or larger files.

Some models support settings such as temperature, top P, reasoning controls, verbosity, structured outputs, tool calling, file input, or image input. Other models do not. If a setting is not visible for the selected model, treat it as unsupported for that model.

You can change the model per agent without rebuilding the workflow structure. Change one model setting at a time, then re-test with the same prompt and inspect traces before treating the result as production-ready.

## Tools

Only give agents the tools they need. Brand Agents should get a smaller, safer tool set. AI Workforce agents can use broader internal connectors when the workflow requires it.

For AI Workforce, add marketplace skills when the agent needs an optimized task process. Add connectors when the agent needs to take action in a business app. The best workflows usually combine clear instructions, the right model, a small set of tools, and one or two relevant skills.

## Structured Outputs

When an application needs predictable JSON, connect the dedicated [Structured Output tool](/dev-guide/structured-output-tools) to the agent that should emit it. Structured Output is a tool, not an agent-level response-schema setting.

## Sub-Agents

<span id="agent-as-a-tool" />

Older interfaces described a specialist agent as an **agent as a tool**. In the current product, add it as a **sub-agent** in Agent Build. The main agent delegates a bounded task to the sub-agent, receives its result, and remains responsible for the user conversation and final answer.

Sub-agents cannot contain their own sub-agents. Give each one a distinct responsibility, focused instructions, and only the tools it needs. See [Agent Build](/ai-workforce/agent-build#main-agents-and-sub-agents) for the current configuration surface.

## Related

* [Brand Agents Overview](/brand-agents/overview)
* [Agent Space and Agent Build](/ai-workforce/agent-spaces)
* [Multi-Agent Orchestration and Handoffs](/user-guides/orchestration-and-handoffs)
* [Tools](/user-guides/tools)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.