Deploying Your First Agent

Deploying an agent means making it reachable outside your setup flow. Your team can access the agent across communication channels, and external events can trigger the agent to start work automatically.

What deployment means

A deployed agent has a clear place to run. That might be chat, Slack, phone, meetings, or an orchestration that starts from a schedule or app event. Before deploying, make sure the agent has been tested against the examples and evaluations that matter most.

Step 1: Add a custom surface

Use Surfaces to choose where people can reach the agent. Start with the channel your team will actually use first, then expand once the behavior is stable.

• Chat: Let teammates start direct conversations with the agent.
• Slack: Let the agent respond where team conversations already happen.
• Phone: Let the agent receive calls from approved callers.
• Meetings: Let the agent join calls, take notes, and follow up.

For chat surfaces, configure the available models, template prompts, and computer panel tabs so the first user experience is focused.

Step 2: Add a trigger and set it live

Add a trigger when the agent should start automatically. A trigger can come from a third-party app event or a schedule. Wire the trigger to the agent that should handle the first step.

• Third-party app: Start from an event like a new ticket, lead, form submission, or message.
• Schedule: Run the orchestration on a recurring cadence.
• Webhook: Use a direct event source when webhook triggers are available.

Test the trigger with a low-risk event first. Once the run looks clean, turn the trigger on so the agent can run on future matching events.

Step 3: Add the agent to an orchestration

Add the agent to an orchestration when its work should be part of a larger process. Orchestrations connect agents, triggers, and handoffs into a directed flow.

• Use one agent for each clear responsibility in the process.
• Connect agents when one agent's output should become another agent's input.
• Describe each handoff so the next agent knows what context it receives.
• Keep the first orchestration small enough to test end to end.