AI Extract

AI Extract loads one page and turns it into the data you asked for. You give it a URL and a description of what you need — optionally a JSON Schema — and it returns the fields, not the page. It is synchronous: one call, one result, no session to poll. A run costs a flat 5 points.

When to use it

Use AI Extract when the source is known and the shape of the answer matters:

fields from a company website: pricing, offices, contacts, legal entity
a profile, listing or product page turned into a record
one-off scraping where writing a parser is not worth it
intake pipelines that must produce the same JSON every time

If you don’t know which page holds the answer, start with Search or let AI Research find and read pages for you. If the source is a provider we already cover — LinkedIn, a marketplace, a company registry — a Data Source returns a stable schema with no prompt to write.

Example request

curl -X POST "https://api.neuralverge.ai/functions/v1/run-extract" \
  -H "Authorization: Bearer YOUR_ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://openai.com/careers/",
    "instructions": "Extract the open job listings: title, team, location and the link to the posting.",
    "settings": { "country_code": "us" }
  }'

Without a schema, the fields come back as readable text in human and machine is null:

{
  "session_id": "11111111-1111-1111-1111-111111111111",
  "kind": "extract",
  "url": "https://openai.com/careers/",
  "human": "- Title: OpenAI Residency\n  Link: https://openai.com/residency/\n- Title: Emerging talent\n  Link: https://openai.com/careers/emerging-talent/",
  "machine": null,
  "total_points": 5
}

Asking for a schema

Pass a JSON Schema as a string in settings.extract_schema_json and the same page comes back as machine, shaped exactly like your schema. This is the mode to use in a pipeline — the response stays stable even when the page is redesigned.

curl -X POST "https://api.neuralverge.ai/functions/v1/run-extract" \
  -H "Authorization: Bearer YOUR_ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "url": "https://openai.com/careers/",
    "instructions": "Extract the open job listings.",
    "settings": {
      "country_code": "us",
      "extract_schema_json": "{\"type\":\"object\",\"properties\":{\"jobs\":{\"type\":\"array\",\"items\":{\"type\":\"object\",\"properties\":{\"title\":{\"type\":\"string\"},\"team\":{\"type\":\"string\"},\"location\":{\"type\":\"string\"},\"url\":{\"type\":\"string\"}}}}}}"
    }
  }'

{
  "session_id": "11111111-1111-1111-1111-111111111111",
  "machine": {
    "jobs": [
      { "title": "OpenAI Residency", "team": null, "location": null, "url": "https://openai.com/residency/" },
      { "title": "Emerging talent", "team": null, "location": null, "url": "https://openai.com/careers/emerging-talent/" }
    ]
  },
  "total_points": 5
}

A field the page does not state comes back null rather than guessed — team and location above are not on that page.

Request reference

Field	Type	Description
`url`	string	The page to load. Required unless you use `base_url` with a URL builder.
`instructions`	string	What to extract, in plain language.
`settings.extract_schema_json`	string	JSON Schema, as a string. Sets the shape of `machine`.
`settings.extract_schema_mode`	string	`provide` to use your schema, `generate` to let the run derive one from your instructions.
`settings.country_code`	string	Country the page is loaded from, for geo-gated content.
`settings.url_builder`	object	Build the target URL from parts instead of passing a finished `url` — see `base_url` and `url_builder_values`.

Response reference

Field	Description
`session_id`	The run, readable again through Get Session Status.
`url` / `used_url`	What you asked for, and the URL actually loaded after any URL building or redirects.
`human`	Readable rendering of what was found.
`machine`	Your schema’s shape when a schema was given, otherwise `null`.
`total_points`	Points charged: 5 per run.

Notes worth knowing

One page per run. AI Extract does not follow links or paginate. For many pages, call it per URL, or use AI Research with a page budget.
Pages behind hard anti-bot defences may fail. When a page cannot be loaded the run fails rather than returning an empty object, so a failed load is never silently an empty record.
The schema is a contract, not a filter. Asking for a field the page lacks yields null; it does not drop the key.

Next steps

Run Extract — full request and response reference
AI Research — when the pages still have to be found
Data Sources — fixed schemas for known providers

Introduction

Product

API Basics

Integrations

AI Extract

AI Extract

When to use it

Example request

Asking for a schema

Request reference

Response reference

Notes worth knowing

Next steps

​AI Extract

​When to use it

​Example request

​Asking for a schema

​Request reference

​Response reference

​Notes worth knowing

​Next steps

AI Extract

When to use it

Example request

Asking for a schema

Request reference

Response reference

Notes worth knowing

Next steps