Create an Agent
Creating a new agent
Section titled “Creating a new agent”- Go to Console → Agents and click New agent — or, if you have no agents yet, use one of the three starting points on the Console home
- Pick a template, describe what you want, or start blank
- Fill in the configuration (described below)
- Click Save
Basic info
Section titled “Basic info”| Field | Description |
|---|---|
| Name | What your agent is called. Visible to users if deployed publicly. |
| Description | Internal note about what this agent does. Not shown to end users. |
| Agent Type | Internal or Customer-Facing. Internal agents can optionally Allow external guests. See Agent Types →. |
| Avatar / Icon | Optional image shown in chat interfaces. |
Instructions
Section titled “Instructions”The instructions define the agent’s behavior. Write them in plain English.
See Writing Agent Instructions → for detailed guidance and examples.
Capabilities tab
Section titled “Capabilities tab”Capabilities are abilities built into the agent runtime. Enable only the ones this agent needs:
| Capability | What it enables |
|---|---|
| Run code / shell in a sandbox | File operations, git, and terminal commands in an isolated container |
| Drive a real browser (Chromium) | Navigate sites, extract data, and drive web apps |
| Web search | Look up up-to-date information from the web |
| Computer use | Vision-based clicks, typing, and scrolling for web UIs (delegated to a sub-agent) |
| Voice | Real-time voice conversations |
| Media generation | Create images, video, audio, and diagrams |
Internal agents include the sandbox and browser capabilities by default. Customer-facing agents do not receive sandbox or browser tools. See Sandboxes → and Computer Use & Browser →.
Integrations tab
Section titled “Integrations tab”Select which integrations the agent can use. You must first create a connection in Integrations before you can attach it here.
On a team org, each attached connection shows who it acts as (“Acts as: you / <name> / team”). You can only attach a teammate’s personal connection if they’ve shared it with you.
Acts as each user — for sign-in based integrations like Notion, Jira, or Google Calendar on an internal agent, you can flip the binding to act as each teammate’s own account: everyone connects their own login from the integration page (“Connect your own…”), and the agent uses the runner’s account. Choose what happens for teammates who haven’t connected theirs: quietly use the team connection (default) or ask them to connect their own first. Computer-side integrations (GitHub, Google Workspace, cloud CLIs) always act as one team identity, and customer conversations always use the team connection.
Knowledge tab
Section titled “Knowledge tab”Attach one or more knowledge collections. When attached, the agent will automatically search them during conversations.
See Create Knowledge →.
Skills tab
Section titled “Skills tab”Attach skills from the system library or your org’s custom skills. Skills are workflow templates the agent can follow for specific task types.
See Using Skills → and Skill Library →.
Settings tab
Section titled “Settings tab”| Setting | Description |
|---|---|
| Execution Mode | Auto, Fast, or Deep. Internal agents default to Auto (SmartRouter). |
| Monthly Credit Limit | Cap the maximum credits this agent can spend per month. |
| Memory | Enable persistent memory — the agent remembers context across separate conversations. |
| Language | Response language override. Defaults to the user’s input language. |
Deleting an agent
Section titled “Deleting an agent”Deleting is an owner/admin action — members can’t delete agents (see Roles & permissions).
- Open the agent’s Settings tab
- Scroll to the bottom → Delete Agent
- Confirm deletion