Summarizer

Overview

The Summarizer node sends an input text to a language model with instructions to produce a condensed version. The output format can be a coherent paragraph, a bullet-point list, or a single TL;DR sentence, depending on the configured format. An optional preserve_entities flag instructs the model to keep named entities (people, organisations, products, dates) intact, which is important when the summary feeds a downstream extraction or fact-checking step.

Configuration

Field	Type	Required	Default	Description
model_key	string	Yes	—	Language model to use for summarisation (e.g. "gpt-4o-mini", "claude-3-haiku"). Must be registered in the platform's model registry.
max_tokens	integer	No	256	Target length of the summary in tokens. The model treats this as a soft upper bound; very long inputs may produce slightly longer summaries. Range: 32–2048.
format	string	No	"paragraph"	Summary style. One of "paragraph" (flowing prose), "bullets" (markdown bullet list), or "tldr" (single sentence prefixed with "TL;DR:").
preserve_entities	boolean	No	false	When true, the model is instructed not to paraphrase or drop named entities (people, organisations, dates, product names).
language	string	No	"en"	BCP-47 language code for the output summary. The model will produce the summary in this language regardless of the input language.
system_prompt	string	No	—	Custom system prompt to prepend to the summarisation instruction. Use to add domain context or tone guidance. If omitted, a default summarisation prompt is used.

Inputs

Port	Type	Description
text	string	The long-form text to be summarised. For structured inputs, extract the relevant field upstream with extract_field or json_transform before connecting to this port.

Outputs

Port	Type	Description
summary	string	The condensed summary in the configured format.
usage	object	Token usage for the LLM call: { prompt_tokens: number, completion_tokens: number, total_tokens: number }.

Example

{
  "nodeType": "summarizer",
  "config": {
    "model_key": "gpt-4o-mini",
    "max_tokens": 150,
    "format": "bullets",
    "preserve_entities": true,
    "language": "en"
  }
}

{
  "nodeType": "summarizer",
  "config": {
    "model_key": "claude-3-haiku",
    "max_tokens": 60,
    "format": "tldr",
    "preserve_entities": false,
    "system_prompt": "You are a legal document assistant. Summarise with precision."
  }
}

Notes

• Very short inputs (under ~100 tokens) may produce a summary nearly as long as the original; consider gating this node with an if_condition that checks input length first.
• The max_tokens limit applies to the completion only; the input text contributes to the total context window of the chosen model — stay within the model's context limit.
• When format is "bullets", the output is a raw markdown string starting with - ; render it in a markdown-aware UI component or pass it through output_parser if you need a structured array.
• For very long documents (articles, transcripts, PDFs), split the input first with the text_splitter node and summarise each chunk, then summarise the summaries in a second pass.