# Human in the Loop

Human in the Loop (HITL) adds an approval step to any action your agent takes. When enabled, the agent prepares its response or action but waits for a human to approve it before executing.

## How it works

```
  Without HITL                          With HITL

  Customer ──▶ Agent ──▶ Reply sent     Customer ──▶ Agent ──▶ Draft ──▶ Human ──▶ Reply sent
              (auto)                                (waits)           approves
```

With HITL enabled:

1. Your agent processes the ticket and decides what to do
2. Instead of executing immediately, it creates a draft or holds the action
3. A human reviews the proposed action in your helpdesk
4. The human approves (sends as-is), edits and sends, or discards

## When to use HITL

| Scenario                    | Recommendation                                               |
| --------------------------- | ------------------------------------------------------------ |
| **First time going live**   | Enable for all reply actions — review before trusting        |
| **Sensitive topics**        | Enable for actions involving refunds, account changes, legal |
| **Routine replies**         | Disable once you're confident in the agent's answers         |
| **Ticket management**       | Usually safe to disable for tagging, categorization          |
| **Customer-facing replies** | Consider enabling until you've verified quality              |

## Setting it up

HITL is configured per action:

1. Go to **Settings > Actions** in your agent workspace
2. Select the action you want to configure
3. Toggle **Require human approval** on or off
4. Save

That's it. When enabled, the action will wait for approval before executing.

## HITL vs. Draft Reply

There are two ways to keep a human in the loop:

| Approach                 | How it works                                    | Best for                                     |
| ------------------------ | ----------------------------------------------- | -------------------------------------------- |
| **HITL on Public Reply** | Agent queues a public reply that needs approval | Reviewing AI replies before they send        |
| **Draft Reply action**   | Agent creates a draft — never sends on its own  | Always-on human review, no risk of auto-send |

**Draft Reply** is the safest option — it never sends automatically, even if HITL is misconfigured. Use it when you want guaranteed human review.

**HITL on Public Reply** is more flexible — you can turn it on and off as you build confidence. Start with HITL on, then disable it once you trust the agent.

## Tips

**Start with HITL on, then gradually disable.** The best way to build trust is to review agent responses for a few days, then relax the requirement as you see consistent quality.

**Use HITL selectively.** You don't need it on every action. It makes sense for customer-facing replies but is usually unnecessary for tagging or categorization.

**Combine with good instructions.** HITL catches mistakes, but well-written [instructions](https://docs.eesel.ai/configure-your-agent/instructions) prevent them. Both together give you the best results.