Quick Prompt node
The Quick Prompt node sends a prompt to a language model (LLM) and returns the response. It is the primary node for AI-powered text generation, classification, extraction, and reasoning within a Task Agent workflow.
The node is called “Quick Prompt” in the palette (category: Actions). Internally it was formerly “LLM Task” — you may see this in older graphs. Both render the same node.
When to use
- Generate text responses (summaries, explanations, emails)
- Extract structured data from unstructured text
- Classify inputs into categories
- Transform or rewrite content
- Reason about data and make decisions
Configuration
Click the Quick Prompt node on the canvas to open its full-screen editor.
Fields
| Setting | Type | Description |
|---|---|---|
| Execution Model | Combobox | LLM model to use (default: “Default Project Config”). Select from project or system models. |
| Prompt | Code editor | The instruction template sent to the LLM. Supports {{variable.path}} syntax for variable interpolation. |
| Output Variable | Text field | Name of the variable that downstream nodes reference (e.g., prompt_result) |
Advanced Options
Expand Advanced Options to configure generation parameters:
| Setting | Type | Description |
|---|---|---|
| Temperature | Slider (0–1) | Controls randomness. 0 = deterministic, 1 = creative |
| Max Tokens | Number input (spinbutton) | Maximum response length in tokens |
| Response Format | Toggle buttons (Text / JSON) | Text for free-form responses; JSON for structured, schema-validated output |
Prompt template
The prompt template is the instruction sent to the LLM. Use variables from upstream nodes:
You are a helpful customer support agent.
The customer says: {{start.message}}
Their account status is: {{queryTable.status}}
Provide a concise, helpful response.Variables are referenced as {{nodeName.outputVariable}}.
Model selection
Choose which LLM model to use for this node. The combobox shows all LLM configurations available in your project. “Default Project Config” uses whatever model is set at the project level.
Use JSON output format when you need structured data for downstream Condition or Switch nodes. Define the expected JSON schema in the prompt (e.g., “Return JSON with keys: sentiment, confidence, category”).
Input/Output
| Direction | Variable | Type |
|---|---|---|
| Input | Any upstream variables referenced in the prompt | varies |
| Output | Configured output variable name | string or object (JSON mode) |
Common patterns
- Chain of thought — Set temperature to 0, ask the model to “think step by step”, use JSON output.
- Classification — Provide categories in the prompt, use JSON output with an enum field.
- Summarization — Feed long text from a RAG Search or Query Table node, ask for a summary.
- Rewriting — Take input text and rephrase it for a different audience or tone.
Common issues
- Empty response — Check that your input variables actually contain data. A missing variable renders as empty string.
- JSON parse error — If using JSON output mode, ensure your prompt clearly instructs the model to return valid JSON. Add “Return only JSON, no other text.”
- Slow response — Complex prompts with large context may take longer. Consider using a smaller model for simple tasks.