AI Agents

LLM Nodes

LLM nodes let you send structured prompts to language models and capture the output for further processing.

Basic Prompt Node

The simplest way to use an LLM. Configure:

System Prompt — sets context and rules
User Message — the actual question or instruction
Output Field — where to store the response in $json

Example: Summarise Customer Feedback

System Prompt:

You are a customer success analyst. Summarise the feedback in 2 sentences.
Focus on the main issue and sentiment. Be neutral and factual.

User Message:

{{ $json.feedbackText }}

Structured Output (JSON Mode)

Force the model to return valid JSON by enabling JSON Mode:

System Prompt:

Extract information from the text. Return ONLY valid JSON with this schema:
{
  "name": string,
  "email": string | null,
  "intent": "purchase" | "support" | "inquiry" | "other",
  "urgency": "high" | "medium" | "low"
}

Then connect a Parse JSON node to extract fields safely.

Chat History (Multi-turn)

Connect a Memory node to maintain conversation history. The LLM node automatically injects prior messages into the prompt.

Model Parameters

Parameter	Effect
`temperature`	Randomness (0 = deterministic, 1 = creative)
`max_tokens`	Maximum response length
`top_p`	Nucleus sampling threshold
`frequency_penalty`	Reduce repetition
`presence_penalty`	Encourage new topics

Cost Management

The LLM node tracks token usage per execution. Use the Analytics dashboard to:

Monitor spend per model
Set budget alerts
Identify high-cost workflows