# Live Mode

Source: https://www.hyperagent.com/docs/concepts/agents/invocations/live-mode

Keep an agent on a checklist so it notices changes and only pings you when something needs attention.

**Live Mode** is how an agent proactively manages a job without intervention. It wakes on an interval, works your checklist, and decides whether something needs to be resolved. Live Mode is how the agent proactively **notices** risk, drift, or opportunity and brings you the decision instead of waiting to be asked.



## When Live Mode is the right tool [#when-live-mode-is-the-right-tool]

Use it when the situation changes over time and silence is a valid outcome.

Reach for Live Mode when:

* Something important moves without a human remembering to check
* You want continuity: yesterday's flags should inform today's check
* "Nothing changed" should not spam you

**Why continuity matters:** imagine yesterday's check flagged an enterprise deal with no next step. Today's check should ask whether that specific risk was fixed, not rediscover the account from scratch. Live Mode keeps the watch in one thread so each check can compare against what it already reported.

### Example Live Mode agents [#example-live-mode-agents]







Use a **schedule** when you want a predictable deliverable on a calendar and a fresh thread is fine.

## The difference between Live Mode and scheduled invocations [#the-difference-between-live-mode-and-scheduled-invocations]

Same product family, different jobs.

Built for watching. Checklist-shaped work. Can suppress "nothing to report."
Continues context across checks. Floor of **30 minutes** between wakes. New
setups often start around once a day until you tighten the interval.

Built for delivery. Standing task prompt on a calendar. Often a **new
thread** each run so Monday's brief does not inherit last week's draft.
Always produces the run you asked for on that cadence.

## What you configure [#what-you-configure]

The settings that shape every wake-up.

What to review each wake-up. Observable items beat "make sure everything is
fine." Leave empty only if the agent's default instructions already define
the watch.

How often to check. Floor is **30 minutes**. Default for many new setups is
about **once a day**.

The UI calls this the **heartbeat model**. It runs the recurring check. Many
teams use a leaner model here and keep a stronger model on the main agent
for interactive work.

Optional Slack channel, DM, Telegram, or email. The thread still holds the
run; the channel is how you hear about alerts.



## Checklist design and quiet runs [#checklist-design-and-quiet-runs]

Write the watch so "nothing happened" is a valid, quiet outcome.

The agent should review a checklist, decide whether a human is needed, and report status so Hyperagent knows whether to notify you. When nothing needs attention, Live Mode can **suppress** the alert.

### Strong checklist items [#strong-checklist-items]

Strong items name **what to look at**, **what changed**, and **when to alert**:

```text
Check Salesforce opportunities that moved stage since the last run.
Alert only if a deal over $50k lost a champion or has no next step.
Include account name, owner, and the evidence you used.
```

```text
Scan #customer-escalations for threads newer than the last run.
Summarize severity. Stay quiet if nothing is P1/P2.
```

### Weak checklist items [#weak-checklist-items]

* "Make sure everything is fine."
* "Be helpful."
* "Check the business."

Vague watches produce noisy alerts or useless silence.

## Live Mode intervals [#live-mode-intervals]

The shortest Live mode interval is **every 30 minutes**. Every interval wakes the agent. For near-real-time reactions to external events, use a [webhook](/docs/concepts/agents/invocations#webhooks) trigger.

## FAQs [#faqs]

No. It is periodic. Hyperagent wakes the agent on your interval, runs the checklist, then stops until the next tick.

No. The minimum gap is 30 minutes.

No. Quiet runs can be suppressed. You should hear about alerts, not every
all-clear.

Yes. They solve different jobs.

Prefer a capable-enough, cost-aware model for routine checks. See [Supported models](/docs/concepts/agents/models). Use a stronger model when a human is in the loop or the alert needs deep analysis.