Agent settings
The agent settings page is where you build and configure your agent. That covers everything from its name and avatar to the AI model it uses and the tools it can access.
Agent Creator
The Agent Creator is an AI assistant built into the editor. It helps you set up your agent through conversation. Open it by clicking the circular avatar button in the top-right corner of the editor.
When you ask it to build a new agent from scratch, the Agent Creator now takes your available skills and shell environment into account. The agent it creates is already set up to use them.
You can also trigger it contextually from within specific sections:
- Let AI write your instructions — available in the instructions section. Describe what your agent should do, its tone, and any constraints. The AI will draft a system prompt for you. It can now also make inline, precise edits to your agent’s existing system prompt, name, or welcome message. You don’t need it to rewrite the whole field for a small change.
- Let AI add Integrations — available in the Add integrations and agents section. Tell the AI which apps your agent needs to work with. It will suggest and add the right integrations, including which actions to enable and whether they should need approval.
- Let AI design your avatar — available in the avatar section. Describe your agent’s personality or visual style and the AI will generate a custom avatar.
- Let AI add knowledge — available in the avatar section. Ask the Agent Creator to add knowledge files and links directly to your agent. It can also turn the knowledge and web access toggles on or off. It can limit access to specific files, folders, or sites.
The Agent Creator can also manage integrations and sub-agents your agent already has, not just add new ones. Ask it to:
- Remove an integration from your agent.
- Enable or disable specific actions on an integration the agent already has. This is the same as using the checkboxes on the integration card.
- Turn “Requires approval” on or off for individual actions. This is the same as flipping the switch on the integration card — see Requiring approval before an action runs.
- Remove a sub-agent (another agent connected to this one as a capability). This is the same as choosing Remove from agent on the sub-agent’s card.
- Turn “Requires approval” on or off for a sub-agent. This is the same as flipping the switch on the sub-agent’s card. It controls whether your agent must pause and ask your permission before calling on that sub-agent during a conversation.
Asking the Agent Creator to add an integration or sub-agent you already have won’t remove an approval requirement you’ve set on it by hand. To turn approval off instead, ask the Agent Creator to update that integration or sub-agent directly. You can also use the switch on its card.
The Agent Creator panel appears as a slide-out on the right side of the editor. It stays in context with the section you opened it from. It also stays in sync with the form as you work. This includes edits you make yourself and, while in edit mode, the agent’s initial avatar. So your next message to it already reflects your latest changes.
Asking for a change to specific text
You can highlight text in the name, welcome message, or instructions fields and add it to the chat. This lets you ask the Agent Creator for a change to just that passage, instead of the whole field.
Give your agent a name and avatar
Give your agent a name that reflects what it does. The name appears in the workspace agent list and in any conversation started with the agent.
For the avatar, choose from one of the seven default icons or upload your own image. The avatar is shown alongside the agent’s name throughout the platform.
Give your agent instructions (system prompt)
The system prompt is the core instruction set for your agent. It defines how the agent thinks, behaves, and responds across every conversation.
Write instructions in plain language. Be specific about:
- The agent’s role and purpose
- Tone and communication style
- Any rules or constraints it should follow
- How it should handle topics outside its scope
You can expand the instruction field into a full-screen editor for longer prompts. An AI writing assistant is also available to help you draft or refine instructions.
Optional advanced settings
Click Optional advanced settings to expand the model and efficiency options.
Choose a model
Select the AI provider and model that powers your agent. First select a Model type (the provider), then select the specific Model name. The model list filters automatically based on the provider you choose.
For a full breakdown of available models and guidance on which to pick, see the model configuration guide.
Temperature
Temperature controls how creative or consistent your agent’s responses are.
| Value | Effect |
|---|---|
| Low (e.g. 0.2) | Predictable, focused responses — suited to factual or process-driven agents |
| High (e.g. 1.5) | More varied, creative responses — suited to brainstorming or content agents |
The temperature slider is only available for models that support it. If the selected model does not allow temperature configuration, the slider is shown but disabled.
Estimate human hours to measure AI efficiency
Enter an estimate of how many hours this agent saves per conversation. This feeds into workspace-level efficiency tracking and ROI reporting — it does not affect how the agent behaves.
Add integrations and agents
Connect integrations and other agents so your agent has tools to act with. See Agent tools & capabilities for the full reference on adding integrations, enabling actions, and requiring approval.
Help your agent learn
Give your agent access to information so it can answer questions and complete tasks accurately.
- Give full access to workspace content folder — the agent can read any file stored in your workspace’s Content area
- Give your agent full web search access — the agent can search the internet for up-to-date information
- Add content sources — add specific files (
.csv,.pdf,.md,.txt,.docx, and more). You can also enter website links, or browse and select files from your workspace content
Create your agent’s welcome message
Write a short intro that appears when someone starts a new chat with your agent. Explain what the agent does and how it can help — write it in the agent’s voice.
Ready to add tools and integrations? See Agent tools & capabilities.