DocumentationAgents

Configure an agent with a clear role.

Define the agent's identity, instructions, knowledge, model behavior, and execution permissions from one configuration form.

Configuration at a glance

5 configuration sections

Only the name, system prompt, and AI model are required. Integrations and autonomous capabilities can be added when the agent needs them.

Overview

Build the operating profile

An agent combines persistent instructions with a selected language model and optional sources of knowledge or tools. This configuration is reused in the Playground, API calls, workflows, and widgets that target the agent.

1. Define

Name the agent and establish its role, boundaries, and response style.

2. Connect

Attach private knowledge and external tools only when the use case requires them.

3. Tune

Choose the model, reasoning effort, creativity, and native capabilities.

Start with the smallest set of permissions and integrations that completes the task. Add broader access after validating the agent in the Playground.

Section 01

Agent Identity

Give the agent a recognizable name and a concise description so it can be identified in directories and target selectors throughout the application.

Agent NameRequired

The display name used in the Agents directory, target selectors, and execution views.

Use a role-oriented name such as Customer Support Assistant or Lead Researcher.

DescriptionOptional

A short summary of the agent specialty, primary tasks, and intended audience.

This text helps collaborators choose the correct agent; it does not replace the system prompt.

Section 02

Instructions & Personality

Set the communication style and write the persistent system prompt that governs every run. The live counters in the card track prompt words and characters as you type.

Instructions and Personality section with tone options and system prompt — Create Agent · Behavioral configuration

Professional

Clear, polite, and structured responses.

Friendly

Warm, empathetic, and approachable.

Technical

Precise, detailed, and mathematically rigorous.

Creative

Imaginative, expressive, and engaging.

Formal

Strict, authoritative, and protocol-driven.

Concise

Short, direct, and straight to the point.

Custom

Flexible style configured through the system prompt.

Communication ToneRequired

Provides a reusable style preset for the agent's responses. The default is Professional.

Choose Custom when the system prompt defines a style that does not fit a preset.

System PromptRequired

Defines the agent's role, goals, constraints, preferred output format, and behavior when information is missing.

Lead with the persona, then list responsibilities, boundaries, source rules, and response format in explicit language.

Section 03

Knowledge & External Integrations

Ground the agent in indexed documents with one RAG knowledge base and connect any number of remote MCP servers for external tools and operations.

Knowledge and External Integrations section with RAG and MCP selectors — Create Agent · Knowledge and tool connections

Knowledge BaseOptional

Attaches one RAG workflow that retrieves relevant chunks from indexed documents for each query.

Use the searchable selector or create a RAG first. Clear the selection to disable retrieval.

MCP ServersOptional

Attaches one or more remote Model Context Protocol servers that expose external operations.

Select multiple servers, then use Configure Tools to control which discovered tools remain enabled.

When an MCP server is attached, all currently discovered tools are enabled initially. Open Configure Tools to keep only the operations this agent should be allowed to call.

Section 04

AI Model & Engine

Select the language model that executes the agent, then tune model-specific reasoning and creativity controls. Available models are grouped by provider in the selector.

AI Model and Engine section with model, reasoning, and temperature controls — Create Agent · Model configuration

AI ModelRequired

The provider model used for every agent execution. The selected model card shows its provider and description.

Models are grouped by provider to make the available model sources easier to compare.

Reasoning IntensityConditional

Controls how much explicit reasoning effort a compatible model applies before returning an answer.

Available values depend on the model. None uses the provider default; unsupported models hide this control.

TemperatureConditional

Controls response variation from 0.0 for focused output to 2.0 for more creative output.

The default is 0.2 with 0.1 increments. Models without temperature support hide this control.

Section 05

Autonomous Capabilities

Grant native platform permissions independently. Both switches are disabled by default and can be enabled only for agents that need live information or sandboxed computation.

Web Search

Allows the agent to search the public web for current information that is not present in its model or attached knowledge base.

Default: Off. Enable for time-sensitive research and verify sources in the requested output.

Code Execution

Allows the agent to run Python in a sandbox for calculations, data processing, and programmatic logic.

Default: Off. Enable for computation-heavy tasks; do not treat the sandbox as persistent storage.

Complete configuration

Validate and create

The Create Agent action becomes available after all required fields are valid. Missing values are highlighted next to their fields, and the bottom action bar provides the same validation state as the header action.

Required before save

Agent Name contains a recognizable value.
System Prompt contains the operating instructions.
An AI Model is selected.

After creation

The application opens the new agent in edit mode, where its generated ID is shown and every setting can be revised. Test the saved behavior in the Playground before connecting it to production workflows or widgets.