Deploying agents
When you finish building your agent, deploying transforms your draft into a live experience that users can interact with. Deployment makes your agent available across your chosen channels and helps ensure your audience benefits from your latest instructions, features, and updates.
Think of deployment as publishing a snapshot of your agent's current state. You can continue to refine your agent in draft mode and redeploy whenever you're ready to push updates to your live users.
Deploying gives you:
Stable delivery: Publish reliable versions of your agent that users can depend on.
Real-world validation: Confirm that your tools, connections, and behaviors work correctly in production scenarios.
Continuous improvement: Update your draft, test changes, and redeploy to roll out improvements as your agent evolves.
How deploying agents works
Deployment follows a clear workflow that separates development from production:
Build in draft: You create and refine your agent in the draft environment, where you can experiment freely without affecting live users.
Test your changes: Use the preview page to verify your agent behaves as expected. The preview always reflects your latest draft version.
Deploy to live: When ready, deployment captures your agent's complete configuration, including instructions, knowledge files, tools, and collaborators, and publishes that snapshot to the live environment where users access it.
Update and redeploy: After deployment, you can continue editing your agent. When you're ready to push updates, redeploy to replace the live version with your latest changes.
For more information about draft and live environments, see Environments.
Each deployment uses the latest draft version, including any recent changes made by collaborators. Always review and test your agent before you deploy it. If teammates are contributing, coordinate to help ensure accuracy and consistency.
If the system detects unsupported models or invalid configurations during deployment, you see an error on the summary page. For help with resolving these issues, see Connecting to channels.
Before you begin
Before you deploy, make sure you:
Provide a clear and valid name and description so users know what your agent does.
Complete all required behaviors, tools, and connections to avoid deployment blockers.
Verify that your draft behaves as expected so your users get the best experience.
These details can ensure a clean and successful deployment.
Deploy the agent
When you are ready to publish the most recent version of your agent, you can deploy it from the configuration page.
To deploy the agent:
Open your agent configuration page.
Select Deploy to open the Pre-deployment Summary.
Review your agent’s components, connection status, and deployment settings.
Review pre‑deployment summary
Use the summary to confirm that everything is ready for production.
Section | Information displayed |
|---|---|
Profile | Agent description, Selected model, and Voice configuration settings |
Knowledge | Uploaded files and knowledge assets |
Tools | All associated tools and their connection status |
Collaborators | List of Collaborators to the agent (if any) |
Behavior | Expected Agent’s behaviour and responses |
A green checkmark 
indicates that a tool’s security schema matches its connection.
Configure connections
If your agent uses APIs, databases, or third‑party services to run certain tasks, deployment is your chance to make sure that these connections are ready. Confirming authentication and connection settings helps your agent work smoothly when users depend on it.
Heading | Description |
|---|---|
Server | The linked external server or endpoint |
Associated Tools | Tools that use this connection |
Live | Whether the connection is active |
Connection Status | Whether authentication succeeded |
Validate and set up connections
During deployment, you can see the connections your agent actively uses. If anything required is missing or incomplete, deployment is paused to prevent you from publishing a broken experience. The disabled Deploy button helps you detect and fix issues early.
To continue, you need to configure or reconnect the missing connection by using the following steps.
To create a new connection:
Click Add connection +.
Enter a Connection ID and Display name.
Click Save and continue.
(Optional) Enable Single sign-on (SSO).
Select an Authentication type:
API Key
Basic Auth
Bearer Token
Key Value Pair (Team credentials only)
Enter the required details for the chosen authentication type.
Choose a credential type:
Member credentials: Each user uses their own credentials.
Team credentials: All users share team credentials.
Choose team credentials for shared system access. Choose member credentials when each user needs individual authorization.
Click Add connection. The connection appears in the table. To edit it, click the pencil icon
.Click Deploy to complete the setup. After deployment, the agent displays a live icon
that indicates it's active in production. Hover over the icon to view the deployment timestamp.
Updating a deployed agent
Your agent doesn't remain static after deployment. You can continuously improve it by changing your draft and redeploying:
Change your draft agent, for example, update instructions, add tools, and modify knowledge.
Test your changes by using the preview page to make sure they work as expected.
When ready, click Deploy again to push your updates to the live environment.
The new deployment replaces the previous live version, making your latest changes available to users immediately.
With this workflow, you can iterate on your agent while you maintain a stable production version. Users always interact with the last deployed version, while you work safely in draft mode.
Undeploy the agent
If your agent is live and you need to stop it, you can undeploy it anytime. Undeployment gives you full control over when users can interact with your agent. When you undeploy:
All production instances stop immediately.
The agent becomes unavailable in live channels.
Users can no longer access it.
After undeployment, your agent returns to draft state, and it will no longer appear in live search results.
To undeploy:
Open the agent configuration page.
Click the Options icon (⋮), select Undeploy, and confirm.
What to do next
After deployment, your agent becomes available across all configured channels. To learn how to use agents in the chat experience, see Using Orchestrate Chat.
Known issues
For known issues related to deploying agents, see Known issues for deploying and channels.