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

# Talking to Hamla

> What to say, what Hamla does on its own, and what it will always stop and ask you about.

Most marketing tools are something you operate. Hamla is something you talk to. That's the whole interface — you describe what you want, it does the work and shows you the result.

Which raises three fair questions: what do I say, what does it do without me, and what can it do behind my back? This page answers all three.

## What to say

Describe the outcome, not the steps.

| Instead of                         | Say                                 |
| ---------------------------------- | ----------------------------------- |
| Navigating to the campaign builder | "Set up cart recovery"              |
| Finding the subject-line field     | "Make the subject shorter"          |
| Opening the analytics tab          | "How did last week go?"             |
| Building an audience filter        | "Who hasn't ordered in two months?" |

You don't need the right words. "أرسل تذكير للي تركوا السلة" and "remind people who left stuff in their cart" get you the same campaign.

**Hamla replies in whatever language you write in** — including transliterated Arabic. Write "3amel campaign" and you get Arabic back. It won't switch you into a language you're not using, and product names, brand names, and coupon codes stay exactly as you typed them.

### Be specific about what you care about

Hamla builds immediately when it has enough to go on, and only asks when something essential is genuinely missing — the goal, the audience, or the offer. So the more you put in the first message, the less back-and-forth:

> Set up a win-back for customers quiet for 60 days, lead with new arrivals, no discount

If it does need to ask, it asks once, with a few focused questions — not a wizard.

### Correcting it is just talking

There's no field to find. Say what's wrong:

> Too long. Cut it to two sentences.

> خلّي النبرة أبسط

> Drop the coupon from the first message but keep it in the follow-up.

## Two modes

There's a toggle for how much rope Hamla has.

**Plan** reads, analyses, and proposes. It cannot change anything — the tools that would modify a campaign aren't even loaded. Use it when you want to think out loud without anything moving.

**Do** is the default and can build, edit, and draft. It still can't put anything live on its own; see below.

## What Hamla does without being asked

Some things happen on their own, on a schedule.

|                              | What it does                                                                                         |
| ---------------------------- | ---------------------------------------------------------------------------------------------------- |
| **After you connect**        | Reads your products, orders, and customers, writes you a briefing, and drafts two or three campaigns |
| **Health checks**            | Watches live campaigns for breakage and for metrics falling off                                      |
| **Daily and weekly reports** | Summarises what happened and what changed                                                            |
| **Anomaly detection**        | Flags spikes and drops you didn't ask about                                                          |

These arrive as alerts, which come in five kinds: something looks wrong, something looks like an opportunity, something needs you, something was done automatically, or here's your report.

## What it can do on its own — and how to change that

There is one setting that governs autonomous action:

| Setting     | Behaviour                                                         |
| ----------- | ----------------------------------------------------------------- |
| **Notify**  | Tells you. Does nothing.                                          |
| **Suggest** | Tells you, recommends a fix, one click applies it.                |
| **Auto**    | Carries out **low-risk** actions itself and tells you afterwards. |

Even on Auto, the only action Hamla takes by itself is **pausing a campaign that's misbehaving** — the safe direction. Anything classified as high-risk always waits for you, on every setting.

<Note>
  **Going live is always yours.** Launching a campaign, resuming a paused one, and scheduling one for later all require your click on a confirmation card. This isn't a policy — those three actions are refused on Hamla's side. There's no autonomy setting that unlocks them.
</Note>

Scheduling counts as launching, just deferred. It needs the same approval.

## What it remembers

Hamla carries context between conversations, so you're not re-explaining yourself:

* **Your brand voice**, picked up from your site and your edits. It only saves a voice profile after seeing enough consistent signals, and after you confirm.
* **Preferences you state.** Say "never use emojis in emails" and it will offer once to save that as a standing rule. It only saves with an explicit yes, and only your own rules — it can't alter Hamla's built-in playbooks.
* **What you've asked before**, so "like last time" means something.

If you'd rather it forgot something, say so.

## When it's unsure

Hamla is built to say "I'm not certain" rather than guess. If you ask whether it can do something and it can't find the answer in its own documentation, it will tell you it's unsure and offer to check — not tell you no.

If you get a confident answer that turns out to be wrong, that's a bug worth reporting to [support@hamla.io](mailto:support@hamla.io).

## Related

<CardGroup cols={2}>
  <Card title="Your first campaign" href="/start/first-campaign">
    The ten-minute version, end to end.
  </Card>

  <Card title="Plans, limits, and AI credits" href="/guides/plans-and-credits">
    What talking to Hamla actually costs.
  </Card>
</CardGroup>