AI Function Nodes
Home / ai / Summarize

Summarize

Compress text or a whole web page into Markdown at a length you choose.

POST /api/v1/summarize MCP summarize aiAI

~30 neurons with `fast`, ~300 with `quality`, of the daily free 10,000.

Playground

Content to summarise.
Fetch and summarise this page instead.
Steer the summary, e.g. "pricing and limits only".
Output language: auto mirrors the source, otherwise a name or code (zh, English, 日本語).
Quality/cost tier. fast uses an 8B model (~30 neurons/call), quality a 70B model (~10× the neurons).
Output shape.
Approximate length of the summary.
Runs against this deployment's live API.

Result

Fill in the form and hit Run.

Call it from code

How it works

Point it at raw text or at a url — when given a URL it fetches the page and converts it to Markdown first, so the model sees the article and not the navigation menu.

style changes the shape of the output, not just its length:

  • bullets — scannable bullet list
  • abstract — one dense paragraph, paper-abstract style
  • tldr — two sentences, maximum
  • outline — nested headings that mirror the document structure
  • keypoints — numbered claims, each with the supporting detail

The response reports compressionRatio so you can see exactly how much context you saved.

Parameters

ParameterTypeDefaultDescription
text string — Content to summarise.
url string — Fetch and summarise this page instead.
style `bullets` | `abstract` | `tldr` | `outline` | `keypoints` "bullets" Output shape.
targetWords integer 150 Approximate length of the summary.
language string "auto" Output language: auto mirrors the source, otherwise a name or code (zh, English, 日本語).
focus string — Steer the summary, e.g. "pricing and limits only".
model `fast` | `balanced` | `quality` "fast" Quality/cost tier. fast uses an 8B model (~30 neurons/call), quality a 70B model (~10× the neurons).

Response

The summary plus a compression report.

ParameterTypeDefaultDescription
summary string — Markdown summary.
title string — Source title, when a URL was given.
model string — Model actually used.
stats object — {inputChars, outputChars, compressionRatio}.

Pairs well with

HTML → Markdown

Strip the chrome, keep the article. Typically 80–90% fewer tokens than raw HTML.

Extract data

Turn messy text or a web page into JSON that matches your schema.

Translate

Translate text between 100+ languages while keeping Markdown structure intact.