Describe It — Stage 1 of Snap Studio¶

Audience: Everyone · Tags: studio, ai, describe, beginner

What you'll see. Stage 1 is one box: a textarea labeled "Tell me what you want to happen," with help text right underneath it — "Plain words are perfect — no technical terms needed." Below the box are three example prompts you can click to prefill it, so you never have to start from a blank page. When you're ready, submit it and SnapLink drafts a plan for you.

If you were already chatting with the Snap AI assistant elsewhere in the app and it offered an "Open in Snap Studio" button, you'll land here with your prompt already filled in — Stage 1 picks up right where that conversation left off.

Writing a good description¶

The assistant works best when your sentence answers two questions: what starts the flow, and what should happen next. You don't need to name a "trigger" or an "action" — just describe the moment and the result.

Too vague	Specific enough to draft from
"Send me alerts."	"When something happens, email the details to ops@example.com."
"Notify the team."	"When a new order comes in, send it to https://hooks.example.com/orders."
"Do the integration thing."	"When a customer signs up, email welcome@mycompany.com to let them know."

Notice the specific versions all include a real destination — an email address or a web address starting with http:// or https://. That's the one piece of information the assistant can't guess for you.

What the assistant can and can't do yet¶

Studio is deliberately narrow today, and it's worth knowing the edges before you start typing:

One trigger. Every flow Studio drafts starts with a webhook — something elsewhere sends SnapLink a signal, and that's what kicks the flow off. There's no way to draft a scheduled or time-based start from Stage 1 yet.
Two kinds of action. The assistant can only draft actions that either send an email or make a web request (REST call) to a URL you provide. If your prompt describes something else, the assistant will draft the closest fit or ask a question — it won't invent a connector that isn't there.
Fixed message text. Whatever text you (or the assistant) put in the message field is exactly what gets sent every time. Studio does not — not yet — pull live data out of the thing that triggered the flow and insert it into the message. If you write "New order received," every run says exactly that.
Two destination types only. An action's destination is either an email address or an http(s):// URL. Anything else won't validate in Stage 2.

None of this is hidden from you — if the assistant can't fully draft what you asked for, it will leave a field blank and ask a plain-language question in Stage 2 rather than guessing.

Two drafting engines — reported honestly¶

Every plan Studio hands back tells you, in its engine field, which of two drafting engines wrote it:

"llm" — a real large language model (Claude) produced the plan, through whichever AI provider your SnapLink operator has configured (for example, Azure AI Foundry or the first-party Anthropic API).
"rules" — a deterministic, keyword-based planner produced the plan instead. This isn't a failure state; it's an automatic, invisible fallback that kicks in whenever no AI provider key is configured, or if the AI call doesn't succeed for any reason. You'll never see an error for this — you'll just get a plan from the simpler engine.

The rules engine works by scanning your sentence for an email address or an http(s):// web address and building the matching action around it. If it can't find either, it defaults to an email action with an empty, required "send to" field and asks you to fill it in.

Both engines produce the exact same plan shape, and Stage 2 looks and behaves identically either way — there's no separate "AI mode" screen. If you want to check which engine drafted a particular plan, look at the engine value in the plan response:

{
  "engine": "rules",
  "name": "Email on new order",
  "summary": "When a webhook is received, send an email.",
  "steps": [
    {
      "id": "trigger-1",
      "kind": "trigger",
      "title": "When a webhook is received",
      "description": "Starts the flow when something sends SnapLink a signal.",
      "connectorKey": "webhook",
      "capability": "trigger",
      "fields": []
    },
    {
      "id": "action-1",
      "kind": "action",
      "title": "Then: send an email",
      "description": "Sends a fixed message to the address you choose.",
      "connectorKey": "email",
      "capability": "send_email",
      "fields": [
        { "key": "to", "label": "Send it to", "value": "", "required": true, "type": "email", "help": "Where should this email go?" },
        { "key": "subject", "label": "Subject", "value": "New order received", "required": false, "type": "text" },
        { "key": "text", "label": "Message", "value": "A new order just came in.", "required": true, "type": "textarea" }
      ]
    }
  ],
  "questions": ["What email address should the message go to?"]
}

In practice, which engine drafted your plan rarely matters to you: you review and fix the same kind of step cards either way (see Review & edit). It matters more to whoever administers your workspace, since it tells them whether an AI provider is actually configured.

If something goes wrong¶

See Snap Studio problems — for example, what to do when the assistant keeps asking the same question, or when your prompt doesn't produce the plan you expected.