Deploy an agent

You build on a draft. Real conversations run a version: a saved copy of the agent that Deploy publishes. So you can keep editing a live agent, and customers only see your changes when you deploy them. An agent’s status is Draft, Live or Archived. Only Deploy and Undeploy move it between Draft and Live.

​

Draft and live versions

• Edits change the draft. Everything you change in the editor saves to the draft: actors, steps, routes, settings and channels.
• Deploy publishes a version. It saves the draft as the next numbered version (v1, v2, and so on) and makes it the one real conversations run.
• Conversations already running keep the version they started with, to the end. New conversations get the new version.
• Tests run the draft. The Test drawer always runs your latest edits, deployed or not.

​

The status in the header

Next to the agent’s name, the header shows where it stands: Beside it, Saving… and then Saved show each edit reaching the draft. If a save fails, the header says Couldn’t save, with Retry.

​

Fix issues first

The Issues button in the rail on the left carries a badge: red with the number of problems that block a deploy, amber when there are only warnings. With nothing to report, it shows a check. Click it to open the Issues panel, which lists them in two groups:

• Must fix before deploy. Deploy stays blocked until these are fixed. For example, an Inbound agent with no channel, a channel with a required field left empty, or a route condition that reads a parameter the agent doesn’t have.
• Worth a look. These never block a deploy. For example, an actor with no instructions, an actor that no route from Start reaches, an Ask Question step with no slot name, or a Transfer step with no actor to hand to.

Each issue says what’s wrong and where. Click it to jump there: the canvas brings the actor or route into view and opens it, or the channels or settings open. The list updates as you edit. Actors with issues also show a dot on the canvas.

​

Deploy

What Deploy does depends on the execution mode:

• Inbound (Responder): it publishes a version, then starts the channels from it. Inboxes are polled, Slack and Teams listeners subscribe, phone numbers go live, and the website widget accepts visitors.
• Outbound (Initiator): it publishes a version. There’s nothing to listen on: conversations you start from Start Conversation, a workflow or the API run that version.

The server checks the agent again when you deploy. If it still finds a problem, or a channel fails after the version is published, the message says what went wrong. Fix it and deploy again.

​

Edit a live agent

Keep editing while the agent is live. Your edits stay in the draft:

• The header shows Undeployed changes, and the Deploy button gets an amber dot (Deploy changes).
• Customers keep talking to the live version.
• The Test drawer runs your edits, so you can try them first.

When you’re ready, click Deploy to publish them as the next version. When the draft already matches the live version, Deploy stays available (Redeploy: live is up to date). It publishes nothing new, and for an Inbound agent it checks the channels again: Already up to date · channels re-checked. That’s useful after you fix a connection. To change the agent’s execution mode, undeploy it first.

​

Version history

Open the ⋯ menu next to the status and choose Version history, or click Undeployed changes. Each deploy saves a version here, newest first, with who deployed it and when. The version customers talk to is marked Live, or Last deployed while the agent is undeployed.

​

Roll back

Click Roll back to this version on any other version and confirm. Customers talk to that version again, and new conversations start on it. Your draft isn’t touched. It shows as undeployed changes, so you can fix what went wrong and deploy again. Rolling back doesn’t copy the old version into your draft. On an agent that isn’t live, the button reads Deploy this version: it makes the agent live again on that version. An archived agent can’t roll back. Unarchive it first.

​

Agents that went live before versions

An agent that was live before versions existed shows Live with a No saved version pill. Customers talk to whatever is saved, so every edit reaches them at once. To keep that from happening by accident, its panels keep a Save Changes button (Save on a route) and a warning instead of saving automatically. An actor’s Tools, Knowledge, Integrations and Hooks sections still save each change at once, as they always did, so those changes reach customers straight away. Click No saved version, then Save live version. What customers talk to now becomes version 1, and from then on your edits go to the draft until you deploy. You can also save it from the empty Version history, or by deploying.

​

Undeploy

Click Undeploy (the square, shown while the agent is live) in the rail and confirm. For an Inbound agent, the platform:

1. Stops all channel listeners, so no more inbound messages are picked up. This includes the WhatsApp subscription and phone numbers.
2. Shows the website widget as offline.
3. Changes the status back to Draft.

No conversation is ended, but conversations in progress on email, chat apps and WhatsApp stop getting replies, because nothing picks up their new messages any more. For an Outbound agent, Undeploy only takes it back to Draft, for example to change its mode. Conversations in progress aren’t ended, and conversations you start still run the last deployed version until you deploy again. Undeploying keeps the versions, and you can deploy again at any time.

​

Archive

Archive agent puts an agent aside. It’s in the ⋯ menu next to the status, and as Archive in the agent’s ⋯ menu on the Chat & Voice Agents list. It’s only for an agent that isn’t live: undeploy first. Unarchive, in the same places, returns it to Draft.

​

What happens to running conversations

​

Mode-specific notes

​

Inbound (Responder)

Listeners poll continuously while deployed. The polling cadence is per-channel:

• Email: every ~minute.
• Slack, Teams: event-driven (push from Slack/Teams).
• Phone: event-driven via Twilio.

If a listener fails (e.g. credentials expired), the channel shows a red badge in the editor. You don’t lose inbound messages once credentials are fixed and the agent is redeployed — the listener resumes from where it left off.

​

Outbound (Initiator)

No listeners run. Deploying publishes the version that the conversations you start will run. You can start conversations before you deploy, too. The Start Conversation dialog says what they’ll run: Runs live v3, Runs v3, the last deployed version, or Runs your latest edits when the agent has never been deployed. A workflow’s Conversation Flow node can also start the agent whether or not it’s live. It runs the same version: the last deployed one, or your latest edits if the agent was never deployed.

​

Troubleshooting

​

Next