Runtype
Agents

Creating and configuring Agents

Build agentic workflows by creating Agents and configuring their tools, instructions, and behavior.

Create an Agent

Configuration lives in tabs (Settings, Behavior, Capabilities) within the Agent editor.

  1. Click Agents in the sidebar (under Products) to open the Agents page

  2. Click New Agent (header button)

  3. In the slide-over, choose how to create the Agent (from a template or "Create Empty Agent") and configure:

    • Name: Descriptive name (e.g., "Customer research Agent")

    • Description: What the Agent does

    • Model: Select AI model (GPT-4o or Claude recommended for Agents)

  4. Click Create Agent (or "Create Empty Agent" / "Create [template name]") to create the Agent. You can then open the Agent to continue configuration in the editor.

You can also create an Agent directly from a Product by clicking Add Capability → New Agent. The Agent will be created and added as a Capability in one flow.

System prompt

Provide high-level guidance in the System prompt (Behavior tab):

You are a customer research Agent. Your goal is to gather comprehensive
information about customer requests by searching knowledge bases, looking up
order history, and analyzing previous interactions.

Always be thorough but concise. Prioritize recent information over historical data.

Keep the system prompt focused on goals and behavior, not specific steps (the Agent decides those).

Add tools

Tools are functions the Agent can call:

  1. In the Agent editor, open the Capabilities tab

  2. In the Tools section, click Configure Tools

  3. In the Tool Selection sheet, select the tools you want (search/filter as needed):

  4. Click Apply Changes to save tool selection

You can add multiple tools—the Agent will choose which to use based on the task.

Give tools descriptive names and clear descriptions. The Agent uses this information to decide when to call each tool.

Tool descriptions

Each tool needs a description explaining when to use it:

Good description:

Searches the knowledge base for articles matching a query. Use this to find
documentation, FAQs, or policy information. Returns top 5 matching articles.

Poor description:

Searches stuff

Agent parameters

Configure Agent behavior in the Behavior tab:

  • Temperature: Controls randomness (0–2 scale; 0.5–0.7 recommended for balanced reasoning)

  • Max Tool Calls: Maximum tool calls before stopping (default: 10)

If you enable Agent Loop, a separate Max Iterations setting controls the maximum reasoning turns (iterations) the Agent can take.

Test the Agent

  1. In the Agent editor, click Test Agent (play icon)

  2. In the Test Agent panel, enter a test goal or query in the chat input

  3. Send the message (Send button or Enter)

  4. Watch the Agent's reasoning and tool calls in the test panel

The panel shows each decision the Agent makes and tool it invokes.

Adding to Products

Once configured, add the Agent to a Product as a Capability:

  1. Go to your Product

  2. Click Add Capability

  3. Select your Agent

  4. Click Add Capability

The Agent is now available through the Product's Surfaces.

Best practices

  • Start small: Begin with 2-3 tools, add more as needed

  • Clear instructions: Define goals and constraints clearly

  • Test iteratively: Run with various inputs to see how Agent reasons

  • Monitor usage: Agents can make many tool calls—watch costs

  • Set iteration limits: Prevent infinite loops

Agents are more expensive than Flows because they make multiple LLM calls during reasoning. Use them when autonomous decision-making justifies the cost.

Next steps

Was this helpful?