Skip to main content

Agent Builder

Use the AI-powered Agent Builder to configure your agents through natural language conversation​

Why use the Agent Builder​

Configuring an AI agent typically means navigating multiple tabs and forms to set up instructions, connectors, functions, guardrails, and more. The Agent Builder simplifies this by letting you describe what you want in plain language and making the configuration changes for you.

The Agent Builder is a conversational AI assistant embedded in Agent Studio. It appears as a collapsible panel alongside your agent's configuration, so you can chat with it while seeing your agent update in real time.

Overview​

When you open the Agent Builder for the first time on a new agent, it prompts you with three onboarding questions to get started:

  1. What should the agent help with?
  2. Who is the agent for?
  3. What information or tools should it use?

These questions help the Builder set up a solid foundation for your agent's configuration.

What the Builder can configure​

  • Agent instructions and description
  • LLM model selection
  • Guardrails
  • Connected connectors (databases, APIs, SaaS services, etc.)
  • AI functions
  • Connected agents (multi-agent workflows)
  • Knowledge bases

What the Builder cannot do​

The Builder works with existing resources in your application. It cannot:

  • Create new connectors, knowledge bases, or AI functions (it will guide you to create them manually first)
  • Configure security rules or access control
  • Set up audit logging
  • Delete agents

For these tasks, use the Agent Settings page or the relevant configuration sections in the Squid Console.

Conversation memory​

The Builder remembers context within a session, so you can build on previous messages. Clearing the chat starts a fresh conversation.

Using the Agent Builder​

  1. Open the Builder - In Agent Studio, click the Agent Builder icon on your agent's page. For newly created agents, a "Start Here" banner highlights the Builder.
  2. Describe what you want - Type a natural language request in the chat input. The Builder interprets your intent and makes the appropriate configuration changes.
  3. Review changes - The Builder responds with a summary of what it changed. Changes are immediately reflected in the agent's configuration.
  4. Iterate - Continue the conversation to refine the agent. The Builder remembers prior context within the session.
  5. Test your agent - Click Test to open the test chat widget and verify the agent behaves as expected.
  6. Clear chat - Use the Clear Chat button to start a fresh Builder conversation if needed.

What you can ask the Agent Builder to do​

Agent instructions and personality​

  • "Make this a customer support agent that responds in a professional, helpful tone"
  • "Update the instructions to always cite sources when providing information"
  • "Add a step-by-step process for handling refund requests"

Model selection​

  • "Switch the model to Claude Sonnet"
  • "Use GPT-4o for this agent"

Guardrails​

  • "Enable the profanity filter and professional tone guardrails"
  • "Add a custom guardrail that prevents the agent from discussing competitor products"

Connectors​

  • "Connect the Salesforce connector and use it when the user asks about customer accounts"
  • "Disconnect the Jira connector"
  • "Update the Slack connector's usage to only send notifications about completed tasks"

AI functions​

  • "Add the generateReport function"
  • "Remove the sendEmail function"

Sub-agents (multi-agent workflows)​

  • "Connect the analysis-agent and delegate complex data analysis tasks to it"
  • "Disconnect the billing-agent"

Knowledge bases​

  • "Connect the product-docs knowledge base for answering product questions"
  • "Update when the FAQ knowledge base should be used"

Agent description​

  • "Update the description to 'Internal HR assistant for employee onboarding questions'"

Tips​

  • The Builder can only connect resources that already exist in your application. If a connector, knowledge base, function, or agent doesn't exist yet, the Builder will guide you on how to create it.
  • For database connectors, ensure the Schema tab has generated descriptions for all tables and columns before connecting them through the Builder. This helps the agent understand the data structure.
  • Start with the three onboarding questions to let the Builder set up a solid foundation, then refine from there.
  • You can make multiple changes in a single message (e.g., "Connect the Salesforce connector and update the instructions to handle account inquiries").

Next steps​

  • Abilities - Learn about each ability type you can add to an agent
  • Agent Settings - Configure settings the Builder doesn't manage, such as access control and audit logging
  • AI Agent SDK - Configure agents programmatically using the Squid SDK
  • Getting Started tutorials - Follow step-by-step tutorials for building agents