METRICRM
FeaturesAIPricingResources
//
Talk to us
METRICRM
ProductPricingFeaturesAIIntegrations
ResourcesUser GuideContact
LegalTerms of ServicePrivacy NoticePDPA NoticeData Processing AgreementCookie Policy
© 2026 Webbalances Solution (003096168-D). All rights reserved.Version 4.48.0
All guidesMetriAgentBuild an AI agent in Agent StudioTools, autonomy levels and guardrailsTest an agent in the simulatorPublish, roll out, roll back and stop agentsPut an AI agent on website chat and messaging appsLet AI agents answer email and portal casesApprove, reject and undo AI actions
User Guide › MetriAgent

Publish, roll out, roll back and stop agents

Publish the draft as a version, try it on a share of conversations, go back instantly and stop agents with a kill switch.

For: Admins

Steps

  1. 1Open the agent and go to Versions. Publishing needs instructions, a language, a greeting, at least one topic and one tool, and a saved draft.
  2. 2Add a note about what changed and select Publish. The first version serves every conversation.
  3. 3For later versions, enter the share of conversations (for example 10%) to try the new version on some customers first. Each customer stays on the same version.
  4. 4Raise the share or select Promote to everyone when you are happy. To go back, select Roll back on an earlier version — it serves every conversation again at once.
  5. 5Copy to draft brings an earlier version's settings back into the draft so you can change and republish them.
  6. 6To stop one agent, turn on Stop this agent at the top of its page. To stop every agent, turn on Stop all AI agents on the Agents page. A stopped agent stops at its next step and conversations go to your team.

Good to know

  • Published versions never change; only the draft does.
  • Every change, publish, rollback and kill switch is in the audit log.

Related guides

  • Build an AI agent in Agent StudioCreate an AI agent from a template and set its role, languages, greeting, instructions, topics and knowledge.
  • Tools, autonomy levels and guardrailsChoose what an agent may do, how much it may do on its own, and the limits that keep it safe.
  • Test an agent in the simulatorChat with an agent as a customer and see each step it takes — nothing is sent or changed.
  • Put an AI agent on website chat and messaging appsLet a published agent answer customers on your website chat, WhatsApp and other messaging apps, and hand over to your team when it should.

Still need help? Contact support

On this pageStepsGood to knowRelated guides