Skip to main content

Create Your First Agent

This guide walks through the first successful setup: describe the job, create the agent, review its first plan, add useful context, and keep approval boundaries clear.

Choose the workspace and guided setup​

Open /app/agents on your deployment and select My workspace or the intended Team workspace. Agents and connections created here inherit that context. See Workspaces And Navigation.

If Getting started is available, start or resume the guided checklist: choose or create an agent, give it a first task, and review its work. Use a starting point and Use example fill in a draft; review it and use the normal create or send control to submit it.

Guide progress belongs to your account in that workspace and can continue across browser sessions. Pausing or dismissing the guide does not delete the agent or cancel work. Restart guide resets the walkthrough, not your agents or conversation history.

1. Start from a goal​

Open the platform and describe the role you want filled. Use plain business language.

Strong prompts include:

  • "Create an agent that prepares a weekly competitor brief for our product team."
  • "Set up an agent to monitor vendor renewal risks and summarize open questions."
  • "Create an agent that turns customer feedback files into themes and leadership notes."
  • "Set up a lightweight sales research agent for target-account prep."

Include constraints if they matter:

  • One-time setup or ongoing work.
  • Internal-only work or allowed external communication.
  • Preferred output format.
  • Files, apps, or websites it should use.
  • Usage expectations.

2. Choose blank setup or a template​

You can create an agent from your own description, or start from the Company Templates.

Use a template when the goal matches an existing pattern. The template gives you a starting charter, suggested tools, and a faster path to useful work. You can edit the agent after creating it.

3. Review the new agent​

After creation, open the agent's chat and review:

  • Name: the display name you and your team will use.
  • Charter: the role and boundaries.
  • Status: whether the agent is active or paused.
  • Plan: any proposed steps or deliverables.
  • Tools and apps: connected capabilities the agent can use.
  • Channels: chat, email, contacts, or app channels.
  • Files: attached or reusable workspace files.
  • Usage: where to review what the agent consumes.

If the agent starts with a plan, check whether the steps match your intent before adding more instructions.

4. Add context​

Useful context makes the agent safer and more effective:

  • Upload background files to the chat or filespace.
  • Tell the agent which sources, apps, or systems matter.
  • Connect only the apps or MCP servers it needs.
  • Add allowed contacts only when communication is part of the job.
  • Store credentials through Secrets And Credentials, not in chat.

5. Configure approval boundaries​

Configure the product controls required by the workflow: contact permissions for communication, Review Before Send for external email, and approval-required API operations for supported integration writes. Check who can approve each request before starting work.

Also tell the agent which decisions it should bring to you. For example:

Ask me before sending external messages, contacting new people, changing another agent, spending heavily, or using credentials for a new destination.

This instruction guides agent behavior; it does not configure mandatory approval controls or an exact spending cap. Approvals can appear as pending requests in chat or Outbox. Use the request's approval control when you are ready.

See Approvals And Requests.

6. Send the first work request​

Give the agent one clear request. For example:

Use the uploaded customer feedback file to identify the five strongest themes. Create a concise summary and list the exact file sections you used.

Watch the timeline for:

  • Acknowledgement or clarifying question.
  • Plan or deliverable updates.
  • File references.
  • Tool activity.
  • Pending requests.
  • Final response or generated files.

7. Refine after the first result​

Use feedback to shape future work:

  • "Make the summaries shorter."
  • "Use this table format next time."
  • "Do not include external links unless they are essential."
  • "Run this every Monday at 9:00 AM America/New_York and confirm the next run."
  • "Pause this agent until I restart it."

The goal is not a perfect first prompt. The goal is a useful first agent with visible work, clear boundaries, and enough context to improve over time.

For recurring work, follow Schedules And Reminders. To suspend the agent or interrupt a run, use Stop And Pause.

Next steps​

Last updated on