Building an Agent
The agent builder has seven tabs. This guide walks through each one so you can confidently configure any kind of agent.

Opening the builder
- To create a new agent: click Agents → New agent, enter a name, then click Create.
- To edit an existing agent: click the agent name in the Agents list, then click Edit agent (pencil icon).
The builder opens with the Overview tab active and a live Test pane on the right side of the screen.
Tab 1 — Overview
General settings for the agent:
- Name — shown in the Agents list and in chat headers.
- Description — a short sentence explaining what the agent does. Shown to teammates in the agent list.
- Welcome message — the first message the agent sends when a user opens it. Use this to guide users on what to ask.
- Suggested prompts — up to 4 example questions shown as clickable chips below the welcome message. Good suggested prompts reduce the learning curve for new users.
Tab 2 — System Prompt
The system prompt defines the agent's persona, behaviour, and constraints. It is always sent at the top of every conversation, invisible to the end user.
See System Prompts for writing tips. The character limit is 8,000 characters.
Tab 3 — Tools
Toggle which tools the agent can use:
| Tool | What it does |
|---|---|
| Web Search | Live web queries via Serper |
| Code Execution | Run Python/JS code in a sandbox |
| Image Generation | Generate images from text prompts |
| TTS | Convert responses to spoken audio |
| RAG | Search the agent's document knowledge base |
Each tool has per-tool configuration — for example, you can restrict Code Execution to Python only, or set the image generation style.
Code Execution runs in an isolated sandbox but still consumes credits. Enable it only for agents where code execution is genuinely needed.
Tab 4 — Knowledge
Attach documents from your workspace library to give the agent a knowledge base. The RAG tool (enabled on the Tools tab) is required for the agent to query these documents.
See Knowledge Base for details.
Tab 5 — Appearance
Customise the agent's visual identity:
- Emoji avatar — pick any emoji as the agent's icon.
- Gradient avatar — choose a colour gradient background.
- Image avatar — upload a custom image (PNG or JPG, max 2 MB).
Tab 6 — Sharing
Control who can see and edit the agent. See Sharing Agents for full details.
Tab 7 — Analytics
View usage statistics for the agent:
- Total messages and conversations
- Active users over time
- Average messages per session
- Top users by message count
Analytics data is available from the moment the agent is first used.
Using the test pane
The test pane on the right is a live chat with your agent. Every time you save a change (for example updating the system prompt), the test pane reflects it immediately. Use it to verify the agent's tone and tool usage before sharing it with the team.
Changes are not automatically saved as you type. Click Save in each tab to persist your changes before switching tabs or leaving the builder.
Was this helpful?