Extract

Turn a URL into clean text or matching JSON.

Pass a page and the shape your application needs. Tabstack returns usable output without a parser, browser fleet, or second model call in your stack.

Extract a page Read the extract guide

Clean Markdown · Schema-matched JSON · Cache, effort and geo controls

tabstack.ai/pricing

TEXT

MARKDOWN

LINKS

extract.tsTypeScript

const res = await client.extract.json({
  url: 'https://example.com/pricing',
  json_schema: {
    type: 'object',
    properties: {
      plan: { type: 'string' },
      price: { type: 'number' },
    },
  },
})

res = client.extract.json(
    url='https://example.com/pricing',
    json_schema={
        'type': 'object',
        'properties': {
            'plan': {'type': 'string'},
            'price': {'type': 'number'},
        },
    },
)

tabstack extract json https://example.com/pricing \
  --schema '{"type":"object","properties":{"plan":{"type":"string"},"price":{"type":"number"}}}'

The Extract API

What does the Tabstack Extract API return?

Tabstack Extract takes a URL and returns clean Markdown or JSON matching the schema you provide. Fetching, rendering, content extraction, and schema enforcement happen before the response comes back.

the page changes

tabstack

fetch, render, extract
enforce the schema

the shape does not

Choose the output

Read the page or return the object.

/extract/markdown

Use it when your model, vector store, or application needs readable page content with headings and links intact.

/extract/json

Use it when the next step expects specific fields and types. Provide the schema with the request and receive matching output.

Inside the call

Four things happen before the response comes back.

Fetch the page

Retrieve the URL you passed, with the cache and geo controls you set.

Render what needs it

Run JavaScript when the content needs it.

Extract the content

Separate the page from its chrome and keep the structure.

Enforce the schema

Return the fields and types your request asked for.

extract.tsTypeScript

const res = await client.extract.json({
  url: 'https://example.com/pricing',
  json_schema: {
    type: 'object',
    properties: {
      plan:  { type: 'string' },
      price: { type: 'number' },
      seats: { type: 'number' },
    },
  },
})

// every key you ask for is present; validate the values

res = client.extract.json(
    url='https://example.com/pricing',
    json_schema={
        'type': 'object',
        'properties': {
            'plan':  {'type': 'string'},
            'price': {'type': 'number'},
            'seats': {'type': 'number'},
        },
    },
)

# every key you ask for is present; validate the values

tabstack extract json 'https://example.com/pricing' \
  --schema '{"type":"object","properties":{"plan":{"type":"string"},"price":{"type":"number"},"seats":{"type":"number"}}}'

Schemas

You define the schema. Tabstack returns the data.

Missing fields: Every key you ask for is present. A field the page does not state can come back null, empty, a placeholder number, or a guessed value. Validate values, not just keys.

Browse the schema library

In your pipeline

Keep extraction out of your application code.

Page layouts change. Rendering requirements vary. A field that looks simple can require navigation, retries, or model reasoning. Tabstack keeps those concerns behind the API while your application keeps a stable output contract. Use Extract for known URLs and repeated jobs.

one output contract

your code

one call
one shape

behind the api

fetch, render
retries, reasoning

Choosing the call

Extract returns what the page says. Generate works out what it means.

Need the fields that are on the page? This endpoint returns them in your schema.
Need interpretation, classification, or comparison? Generate adds instructions and returns that result in your schema.
Extract the plan names and prices, then ask Generate which audience each plan appears to target, and why.

Controls

Match the call to the page and the job.

Three parameters change how the page is retrieved.

Control	Use it when
`nocache`	The workflow cannot reuse a cached result
`effort`	You want to choose the retrieval depth
`geo_target`	The page varies by country

Use cases

Use Extract when a known page feeds another step.

Read a pasted link

Turn a URL someone dropped in as clean, readable Markdown.

Current docs for an assistant

Give a coding assistant documentation newer than its training data.

Keep a source set current

Refresh a defined set of pages that a retrieval index depends on.

Enrich an incoming record

Fill out a company or contact record from its public pages.

Monitor a public page

Track product, price, job, policy, or filing pages over time.

Typed records for a data product

Turn pages into rows your own product can query and ship.

Trust

How Extract handles your data.

Never trained on. Private by default. The Trust page explains what is processed and the account controls.

How data is handled

your url