Building an Agent

The Agent builder walks you through everything an agent needs in a guided, step-by-step flow, with a live preview beside you the whole way.

Opening the Agent builder

1. From the sidebar, go to Agents.
2. Click New Agent.
3. The builder opens with a step-by-step layout on the left and a live preview on the right.

[Screenshot: Agent builder showing the step list and live preview]

The Agent builder steps

The builder is organized into a series of steps. You can move between them freely - nothing is locked. Your progress saves automatically.

1. Purpose

Describe what this Agent is for and who it serves. This step sets the foundation - everything else (personality, knowledge, services) flows from it.

• Give your Agent a clear, descriptive name (e.g., "Acme Onboarding Concierge").
• Write a short purpose statement - one or two sentences on what success looks like for this Agent.

Tip - Be specific. "Helps customers" is too broad. "Answers pre-sale questions about Acme Pro, qualifies leads, and books demos" is much better.

2. Personality

This step has four sub-tabs that shape how your Agent talks.

• Vibe - Pick a high-level tone (playful, professional, warm, no-nonsense, etc.). Vibes influence both writing style and, when voice is enabled, the speaking style.
• Persona - Write a short character description. Who is this Agent? How do they introduce themselves? What do they sound like?
• Guardrails - Define what the Agent should and should not do. Examples: "Never discuss pricing - always defer to sales." "Always confirm the user's intent before acting." Guardrails become explicit instructions to the underlying model.
• Voice - Toggle Conversational voice on if you want the Agent to be talkable. When voice is on, pick a voice from your Your voices (cloned or generated) or the Library. See Voice Mode for the full guide.

[Screenshot: Personality step with vibe, persona, guardrails, and voice sub-tabs]

3. Knowledge

Choose which parts of your Notebooks this Agent can read from at runtime.

• Attach one or more knowledge sources (documents, websites, connectors).
• The Agent will only search the sources you attach here, not your entire Notebook library. This is how you keep an HR Agent focused on HR docs and a Sales Agent focused on the catalog.
• You can attach the same source to multiple Agents.

See How Notebook Search Works for the mental model behind retrieval.

4. Services

Give your Agent tools it can call during a conversation - for example, search the web, look up a customer, draft an email, or hand off to another agent.

• Toggle on the services you want available. Available options depend on what you've connected (Composio, MCP servers, web tools).
• For each service, you can usually pick which specific actions to expose - e.g., from Gmail you might allow "draft" but not "send."
• You can also add sub-agents the Agent can hand off to when a request falls outside its scope.

Tip - Start with one or two services. Too many tools makes the agent less reliable, not more.

5. Clarify

Decide how aggressively your Agent should ask clarifying questions before acting.

• A concierge should clarify a lot ("Is this for Acme Pro or Acme Free?").
• A quick-lookup agent should mostly just answer.

6. Model

Choose which AI model powers the Agent. Larger models reason better but cost more and run slower. Defaults are usually a good starting point.

• See LLM Providers to add new model providers.
• You can change the model later without losing any of the rest of your setup.

7. Publish & Share

Once everything looks right, publish your Agent.

• Use the live preview on the right to chat with it as you go. The preview reflects unsaved changes.
• When you're happy, hit Publish. The Agent becomes available to share or embed (see Deploying for sharing options).

Live preview

Every step has a live preview pane on the right. As you tweak the persona or attach knowledge, send a test message - the response updates against your latest configuration. Use this to:

• Sanity-check guardrails ("Is it refusing the right things?")
• Verify retrieval ("Did it find the right doc?")
• Try voice ("Does the picked voice match the vibe?")

Tips

• Iterate, don't perfect. Ship a first version, watch real conversations in Execution History, then tune persona and guardrails based on what people actually ask.
• Keep the purpose narrow. Narrow Agents beat broad ones. If your Agent is trying to do five jobs, split it into five Agents with handoffs.
• Don't over-attach knowledge. More sources mean noisier retrieval. Attach only what's relevant.

Next steps