Basic Prompting

Every workflow you build will reuse this shape. Basic Prompting takes a message, frames it with a reusable prompt, sends it to a language model, and returns the answer — nothing hidden, nothing implicit. Start here to learn how the canvas connects components before you add memory, documents, or tools.

How it works

The flow is four components on a single path:

1. Chat Input captures the message — from the playground while you build, or from whatever calls the workflow once it's deployed.
2. Prompt holds your instructions — the framing the model should follow, written once and reused. It feeds the model as its own input, separate from the message, so you shape behavior here without touching the user's text.
3. Language Model takes the message and the instructions and runs the completion. The component is provider-agnostic: the same flow runs on any model you've added a credential for, so swapping the model is a credential change, not a rewire.
4. Chat Output returns the response to the caller.

That separation between instructions (the Prompt) and input (the Chat Input) is the whole lesson. Keep them apart and you can rewrite the system's behavior by editing one component, never by concatenating strings.

When to reach for it

Reach for Basic Prompting when the job is a single, self-contained turn — classification, rewriting, extraction, a one-shot question. It's also the right skeleton to fork when you're starting something larger and want a clean spine before adding components.

When it's not enough

This flow has no memory and no grounding. If the assistant needs to recall earlier turns, start from Memory Chatbot. If it needs to answer from a document or knowledge base instead of the model's own training, use Document Q&A or Vector Store RAG.

Try it

Set the Prompt to a clear instruction:

You are a precise technical editor. Rewrite whatever the user sends for clarity,
keeping every fact unchanged. Reply with only the rewrite.

Run the workflow with a clumsy paragraph as the chat input. Then edit only the Prompt — ask for bullet points, a different tone, a word limit — and run again. The behavior changes; the wiring doesn't. That's the model you'll carry into every other template.