Extract data

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

POST /api/v1/extract-data MCP extract_data aiAI

~30–60 neurons per call with the `fast` model.

Playground

Result

Fill in the form and hit Run.

Call it from code

How it works

Two ways to say what you want:

fields — a list of names, optionally annotated: ["title", "price:number", "inStock:boolean", "tags:array"].
schema — a full JSON Schema object, when you need nesting or descriptions.

The model is constrained to JSON output and the result is validated before it is returned, so you either get conforming data or an invalid_input error — never half-parsed prose.

For pages with a stable DOM, scrape is cheaper and more repeatable. Use this one when the layout varies or the value you want is buried in prose.

Parameters

Parameter	Type	Default	Description
text	string	—	Source content.
url	string	—	Fetch this page and extract from it.
fields	string[]	—	Shorthand field list. Annotate types with `name:type` — string (default), number, boolean, array.
schema	object	—	Full JSON Schema for the desired object. Overrides `fields`.
instructions	string	—	Extra guidance, e.g. "prices in USD, dates as ISO 8601".
model	`fast` \| `balanced` \| `quality`	`"fast"`	Quality/cost tier. `fast` uses an 8B model (~30 neurons/call), `quality` a 70B model (~10× the neurons).

Response

`data` is the extracted object.

Parameter	Type	Default	Description
data	object	—	Extracted values.
model	string	—	Model used.
schemaUsed	object	—	The JSON Schema that was enforced.

Pairs well with

Scrape

Pull named fields out of a page with CSS selectors — deterministic, no AI tokens.

Summarize

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