A plain chat with a model

The first thing you want to know about a platform is whether it answers at all. This is that answer: four nodes, no connections, works the moment you import it. Everything else on this page is this scenario plus a step or two.

What is inside

1. Customer message — the entry. Whatever the person typed into the chat on your site, the widget, or a messenger bot. The scenario does not know which door it came through, and does not need to.
2. Model — a node of its own rather than a field inside the agent. It is wired with a link_llm connection and says one thing: which model does the thinking. One model can serve several steps, and changing it is a single edit.
3. Model’s answer — the agent. It takes the message, calls the model, returns text.
4. Send the answer — the exit. What the customer ends up reading.

Three ready-made prompts

The only difference between the bots below is the text in the agent’s System prompt. The graph stays exactly the same.

Frequently asked questions. Takes the routine delivery, payment and returns questions off a human.

You answer frequently asked questions. Be brief, concrete and friendly, and rely only on
what the conversation gives you. When there is not enough to answer precisely, say so and
ask for the detail you need rather than inventing facts.

Summarising. Takes a long text and returns the gist.

You summarise text. Keep the facts, figures and names; drop the padding and repetition.
Three to five sentences by default. If the user asks for a different length or format,
follow that instead.

Translation. Translates incoming text into the requested language.

You are a translator. Translate the user's text into the language named in their message;
default to English when none is named. Preserve meaning, tone and formatting, and add no
commentary of your own.

What to set up for yourself

• The model. The template leaves it unset on purpose, so the import works on any plan. Open the Model node and pick one that suits your budget and language.
• The agent’s system prompt. An empty prompt means “answer like a general assistant”. This is where you write who your bot is and what it talks about.
• The greeting is not in the graph: it lives in the channel settings on the Publish tab.

What to connect

Nothing. The model is paid for in quarks from the workspace balance; no separate key is needed. If you would rather run on your own key, add an “own model key” connection in settings and pick its model in the Model node.