Your first agent

This walkthrough creates a small research agent, gives it a task in chat and puts it on a daily schedule. You need the creator or admin role and at least one connected model provider (Platform keys).

1. Create the agent

1. On the Dashboard, click Create Agent.
2. Choose a starting point:
   • Blank agent — an empty agent with default instructions;
   • a template such as scout (research), sage (analysis) or scribe (writing);
   • Import a GitHub repository that already contains an agent definition (see Creating agents).
3. Give it a name, for example market-scout. Names are lowercase and become part of URLs and API paths.
4. Pick a compute profile if you are on ExoHuman Cloud. Standard is enough for research and writing agents.
5. Click Create. The agent’s container is created and started.

2. Tell it what it is for

Open the agent and go to Settings → General to review its runtime, then open Files and edit AGENTS.md. This file is the agent’s standing instructions — its role, what good output looks like, which tools to use and when to ask a person. For example:

# Market scout

You monitor public news about our three main competitors and summarise what changed.

- Sources: company blogs, press releases, reputable trade press.
- Output: a short report with links. Flag anything about pricing as high priority.
- If you are unsure whether something is relevant, ask the operator instead of guessing.

3. Chat with it

Open the Chat tab and send a message: “What did our competitors announce this week?”. The agent works in its container — browsing, running tools, writing files — and answers. Each message is an execution you can inspect in Tasks, with its duration, cost and tool calls.

The agent page chat is stateless per turn (it uses the recent messages as context). For an ongoing conversation, use the Workspace.

4. Put it on a schedule

1. Open Schedules → New schedule.
2. Choose a preset or write a 5-field cron expression, for example 0 8 * * 1-5 (08:00 on weekdays), and your timezone.
3. Enter the task: “Write today’s competitor briefing and publish it as a report.”
4. Save, then click Run now to test it.

Scheduled runs only execute when the agent’s Autonomy is switched on (the Autonomy chip in the agent header or its Dashboard tile). Autonomy is off by default so nothing runs unattended until you decide it should. See Scheduling & loops.

5. Keep it safe

Before giving the agent credentials or write access to important systems:

• review its guardrails;
• add only the credentials it needs;
• decide who it may delegate to on the Permissions tab.

Next

• Invite your team and share the agent.
• Learn every configuration option.