Agent Builder

The Agent Builder brings your model, prompt, knowledge, and actions together in Build. Use Auto for help from the AI assistant, the gear button for Settings, and Preview to test your agent.

The video uses the previous builder layout. The controls below describe the current layout.

Auto

Describe what you want to build or change. The AI assistant suggests changes for you to review. When it suggests an action, accepting the suggestion opens that action's connection dialog so you can finish setup without leaving the conversation.

Build

Choose your Tool Type and Model, then write the System prompt that tells the agent how to behave. Forms also have a prompt frame for their inputs. Use the microphone in the system prompt to dictate additional instructions in supported browsers, then review the transcription.

Open the model picker to compare models and set supported reasoning or search options. See Models for those controls.

Knowledge

In Build → Knowledge Base, choose Connect files, or Add more files when sources are already connected. Search the picker or browse its Files, Web, and Apps groups. You can add documents, images, audio, video, web pages, YouTube sources, RSS feeds, and connected apps.

Sources added here connect to this agent. The workspace's Agents → Knowledge Base page manages sources shared across agents. See Knowledge Base for supported sources and sync behavior.

Actions

In Build → Actions, choose Browse actions to connect tools, APIs, MCP servers, or other agents. Connected actions appear below the library button. See Actions for setup and account connections.

Settings

The gear button beside Build opens Settings:

• Capabilities controls file uploads, voice input, spoken responses, and visible reasoning steps, with additional capabilities where supported.
• Customize contains your profile image, chat icon, placeholder text, intro message, and ice breakers.
• Token Lengths controls the space available for responses, memory, user documents, and knowledge.
• Advanced contains additional model controls and Model Reminder, an instruction included with every user message.
• Knowledge Settings contains knowledge context and search mode.

Choose voice, beside Allow users to hear responses, offers Pickaxe voices and Custom voice. The latter connects your ElevenLabs account so you can choose one of your voices or enter a Voice ID. Its API key needs Text to Speech, Voices, and Voice Generation permissions.

Preview

Test the agent in the preview beside your configuration. Select Sender lets you test as a particular user. When your changes are ready, choose Update and close.