Creating an AI agent

Use Agent Studio to build an AI agent without code

Agent Studio is the quickest way to start building an AI agent. Below is a step-by-step guide to creating an AI agent in Agent Studio. Once complete, you'll have a working AI agent, which can also be extended and customized further with Squid SDKs if needed.

Create a New Agent

1. Navigate to the Agent Studio section from the left sidebar menu.
2. Click on the Create New Agent button on the right side of the page.
3. Provide a a descriptive name for your agent in the Agent ID field. Note: This ID cannot be changed afterward.
   • Example: "ACME business analyst".
4. Add a detailed description of what your agent does, for future reference.
   • Example: "Queries and emails my ACME home goods business data."
5. Click Create to initialize your new agent.

Agent Configuration

The next steps take place in the Overview tab where you'll start configuring your agent.

1. LLM Model: Use the default LLM (GPT-4o) selected, or choose another LLM to power your agent.

2. Instructions: Explain to your agent how it should respond and interact with the user, in addition to explaining to the agent what it needs to do:

   • It can be helpful to define what kind of role the agent will be taking on (e.g., "You're an expert assistant for an internal team...")
   • You can include tone, personality, and wordiness guidelines (e.g., "Keep responses concise" or "Explain difficult concepts in plain English and using analogies")
   • Add in any optional special instructions for handling data (e.g., "include source citations")

3. Agent Abilities: Enhance your agent by giving it one or more capabilities by clicking on the "Add Abilities" button:

   • Knowledge Base: Upload PDFs or text documents that the agent should reference.
   • Communication: Set up email, Slack, and other messaging channels the agent should use to send answers or responses.
   • Databases: Allow the agent to connect to SQL and NoSQL database tables and collections that hold relevant data.
   • SaaS: Connect the agent to services like Jira, ServiceNow, or Zendesk. More services are coming soon, or reach out if you have a specific service that you'd like to work with.
   • AI Functions: Add specialized capabilities, including built-in functions to summarize a URL or custom AI functions you may have defined through Squid's Backend SDK.

   Upon connecting to a database, SaaS, communication service, or another AI agent, a window will pop up asking for more information.

   Provide an explanation on when to use the connector (e.g., "Call this connector whenever the user asks to send an email") along with any specific instructions that you'd like to include (e.g., "Send emails in HTML format").

Testing Your Agent

1. Use the "Test Agent" tab to interact with your agent directly and test it out.
2. As you interact with your agent, steps your agent is taking and what information it's accessing is shown side-by-side.
3. You can check to see if the agent is accessing the right resources or responding as you'd expect.
4. Experiment and adjust your instructions and/or connectors as needed based on your agent's responses. If you need, help, feel free to contact us.

Audit Log

If you have "Report Audit Logs" enabled in the Agent Settings tab, then you can go to "Audit Log" tab to see a log of all of your agent's interactions.

These include a timestamp of the interaction, the user ID, and the prompt. Expanding each row will also show the exact prompt and the agent's answer. More fields will be coming soon.

Agent Settings and Guardrails

You can access the Agent Settings tab to configure your agent's description, privacy settings, toggle audit logging on and off, or set up guardrails. Learn more about Agent Settings and guardrails.

Next Steps

Once you're comfortable creating AI agents in Agent Studio, learn how to build more complex agents using AI functions and the Squid SDKs.