> ## Documentation Index
> Fetch the complete documentation index at: https://docs.m9tz.de/llms.txt
> Use this file to discover all available pages before exploring further.

# Deploy Phase: Moving Your AI Agent Into Live Operations

> Learn how go-live works, what the pre-deploy checklist covers, how the first week of monitoring runs, and what your 14-day hypercare period includes.

Staging approval means your agent is ready for the real world. The Deploy phase moves it from the sandboxed environment into your live business operations — carefully, with a clear checklist, a shared go-live call, and a full week of close monitoring before we hand you the keys.

## Pre-Deploy Checklist

Nothing goes live until every item on this checklist is confirmed. We review this together before scheduling the go-live call.

<Steps>
  <Step title="All integrations authorised in production">
    Every tool connection — CRM, inbox, helpdesk, database — must be authorised against your live production credentials, not the staging equivalents. Verify that each connection is active and has the correct permission scope.
  </Step>

  <Step title="All test cases passed">
    The full staging test suite must have passed with no outstanding failures. Any open issues flagged during your staging review must be resolved and re-tested before this step is signed off.
  </Step>

  <Step title="Your team has been notified">
    Anyone who interacts with the workflows the agent will touch — support staff, ops team, managers — should know the agent is going live, what it does, and how to recognise its activity.
  </Step>

  <Step title="Rollback plan is in place">
    We document the exact steps to pause or disable the agent if something unexpected happens after go-live. You should know where the pause control is before the agent is live.
  </Step>
</Steps>

<Warning>
  Do not skip the team notification step. An agent acting on live data without your colleagues knowing it exists leads to confusion, duplicated effort, and eroded trust in the system. A brief internal message before go-live prevents all of that.
</Warning>

## The Go-Live Process

Go-live is a structured, supervised event — not a silent deployment.

<CardGroup cols={3}>
  <Card title="Production Deployment" icon="rocket">
    We deploy the agent to your production environment at the agreed time. The deployment itself takes minutes; the verification that follows is what matters.
  </Card>

  <Card title="Live Smoke Test" icon="vial">
    With you on a call, we trigger the agent against a real input and walk through the output together — verifying that every step behaves exactly as designed in the live environment.
  </Card>

  <Card title="Go/No-Go Confirmation" icon="circle-check">
    Only after the smoke test confirms correct behaviour do we formally mark the agent as live. If the smoke test surfaces an issue, we pause and resolve it before proceeding.
  </Card>
</CardGroup>

<Note>
  The go-live call typically takes 30–45 minutes. Plan for it to run slightly longer on your first agent deployment — it's worth taking the time to feel confident before the call ends.
</Note>

## First-Week Monitoring

The first seven days after launch are the highest-risk window for any agent. We treat them accordingly.

During this period, our team actively monitors:

* **Error rates** — any unexpected failures in the agent's workflow steps
* **Escalation frequency** — whether the agent is handing off to humans more or less often than the design anticipated
* **Tool connection health** — confirming that every integration stays stable under real traffic
* **Output quality** — spot-checking agent responses or actions against the expected behaviour defined in the design

<Note>
  If we identify anything that looks off during the first seven days — an unexpected error pattern, a guardrail triggering too frequently, a tool behaving differently in production — we investigate and fix it at **no extra charge**.
</Note>

## How to Pause or Stop the Agent

You are always in control. If something unexpected comes up — a business process changes, a tool has an outage, or you simply want to review behaviour before letting the agent continue — you can pause it immediately.

<Tip>
  Log in to your **control panel** and navigate to your agent's settings to find the pause control. Pausing is instant and non-destructive — nothing in progress is lost, and the agent can be resumed just as quickly. Your account manager can walk you through this during the go-live call if you'd like a live demo.
</Tip>

Pausing does not affect any data already processed. It simply stops the agent from acting on new triggers until you resume it.

## 14-Day Hypercare Period

After go-live, you enter a **14-day hypercare period**. During this window, our team is on standby for fast fixes — no new scoping, no project queue, no waiting.

<CardGroup cols={2}>
  <Card title="What Hypercare Covers" icon="shield-check">
    Bug fixes, unexpected behaviour, integration issues, and adjustments to guardrail thresholds that arise from real-world usage. If it's a result of the build, it's covered.
  </Card>

  <Card title="What Hypercare Doesn't Cover" icon="circle-xmark">
    New features, workflow additions, new integrations, or scope changes. Those are handled as change requests after hypercare ends.
  </Card>
</CardGroup>

Once hypercare closes, you move into the [Handoff phase](/blueprint/handoff), where we make sure your team can own and operate the agent independently.