Overview
Socrates Builder guides you through creating an AI Agent by asking a series of questions about the agent's role, instructions, tools, and output. As you respond, Socrates tracks progress in a to-do list and updates the configuration artifact alongside the chat in real time. Once the agent is complete, Socrates saves it to the AI Agents hub, publishes it, and replaces the placeholder step on the canvas with the live agent, if the session was started from the canvas.
This path is recommended when you want AI assistance designing the agent's behavior, or when starting from scratch without a clear configuration in mind.
Prerequisites
Any integrations the agent will use as tools must be configured in Torq before setup.
How to use
Start Socrates Builder
Socrates Builder is available from two entry points.
From the canvas
Open the workflow: Go to Build > Workflows and open the workflow.
Place a new agent step: In the step picker, select AI Agents, then drag New AI Agent onto the canvas.
Open the Builder: Click the New AI Agent placeholder step to select it, then click Build with Socrates above it.
From the AI Agents page
Open the AI Agents page: Go to Build > AI Agents.
Open an agent: Click an existing Draft agent, or click Create to open a new one.
Open the Builder: Click Build with Socrates in the top bar.
The breadcrumb at the top shows the workflow and agent name. Click it at any point to go back.
Answer the initial questions
Socrates opens with a series of questions about the agent's role and goal. Provide a clear description of the following:
The job the agent owns end-to-end.
What a good result looks like, and what the agent should return.
The systems the agent may call, and which integration instance each should use.
What the agent must not do: actions to leave to an analyst, thresholds it should not act below, data it should not touch.
The inputs the agent receives, and where they come from.
The more detail you provide upfront, the fewer follow-up rounds are needed. For guidance on structuring your input, see Socrates Builder: Prompting and Iteration Guide.
Follow the to-do list
Once Socrates understands the agent's purpose, it creates a to-do list and works through it step by step: capturing the agent's purpose, selecting tools, choosing a model, running a test, and publishing. The to-do list is visible on the right side of the chat and updates as each step completes. Socrates prompts you when input is needed.
Review and refine the configuration
As the conversation progresses, a configuration artifact appears alongside the chat. Use the tabs to review what Socrates has built:
Properties: The agent's name, description, model, and availability.
Instructions: The agent's behavior guidelines and mission.
Params: Dynamic input parameters passed to the agent at runtime.
Tools: The integrations and actions the agent can call during execution.
Output: The agent's response format and JSON schema.
To refine the configuration, give feedback directly in the chat. Target the specific area that needs adjustment:
Configuration: "The agent is using the wrong Slack instance. Switch it to soc_slack."
Boundaries: "The agent closed the case itself. It should recommend closing and leave the action to an analyst."
Output: "Include the enrichment results that the verdict is based on."
Test the agent's behavior
Before finishing, test the agent's boundaries rather than the happy path:
"Run it against an alert with no source IP. Does it handle the missing field or fail?"
"Run it against a benign alert. Confirm it returns benign and takes no action."
"Confirm it never calls the account-disable tool, whatever the verdict."
Finish and return
When the agent is complete, two options appear:
Back to canvas: Returns to the workflow. The placeholder step is replaced with the live agent.
Open the AI Agents page: Opens the agent in the AI Agents page for further review or manual adjustments.
The agent is saved to the AI Agents page and published automatically. To chat with it, open Socrates and select it from the agent picker at the top of the page. For details, see Socrates Builder: Getting Started.
After the agent is published, its configuration panel in Socrates is read-only. To make changes, click Open in AI Agents in the top-right corner to open the agent in the AI Agents page.
