Skip to main content
POST
The POST /llm endpoint sends a prompt to an LLM (ChatGPT or Gemini), or a search query to Google AI Overview, through the LLM Scraping API. You send source, prompt, and optional countries. For ChatGPT, an include object returns the structured result below instead of the short answer. Access token: You need a Bringits access token (see Token acquisition). Send it as a Bearer token on every call and in Try it.

Endpoint

https://unblocker.bringits.com/llm

Headers

In Try it, enter your JWT only in Authorize (token field). Do not paste the full Authorization header or duplicate the token in a Headers box.

Response headers

Rate-limit headers (X-RateLimit-*, Retry-After) are currently returned on POST /request 429 responses, not on /llm.

Request body parameters

How the request is handled

  1. Quota — Your monthly allowance is checked before the call.
  2. Location — When you send countries, the answer is fetched from that region. Omit countries and the call uses ["US"].
  3. Fresh request — Each call is independent; there is no session to reuse from a previous call.
  4. Answer — On success you receive the answer as JSON (response.text and response.markdown). With ChatGPT include, you receive result instead (see below). Gemini and Google AI Overview calls are slower than ChatGPT, so use a long client timeout for them. Use a long client timeout for ChatGPT include calls as well.

Response format

When the call succeeds, the HTTP status is 200. Without ChatGPT include, the body is:

Google AI Overview

With "source": "google_ai_overview", prompt is a Google search query. The response uses the same object, plus the parts of the overview Google shows:
No overview for the query. Google does not show an AI Overview for every search. When the results page has none, the call still succeeds: HTTP 200, empty response.text, empty arrays, and "metadata": { "country": "US", "reason": "not_served" }. It counts toward your monthly quota like any successful call. If Google blocks the search, the call fails instead (see below).

ChatGPT include

Send include with "source": "chatgpt" and one prompt. The call returns one result object. HTTP 200 always has "success": true. A failed call is not 200, and the body is the error object below, not "success": false. A successful call counts once toward your quota.
This example was produced with "include": { "markdown": true, "shopping": true }, so it has markdown, shoppingCards, and inlineProducts. Flags you do not send are left out of result. A flag set to true adds its keys even when the page had nothing to put in them. An empty array (or null) means this answer had no data of that kind. It is still a success, and a later call with the same prompt can fill it. An unknown flag or a non-boolean value (for example "ads": "yes") fails with HTTP 400 and "error": "REQUEST_FAILED".

Fields that are always on result

sources

A Google Maps directions link is not a source.

citationPills

entities

A product for sale is not copied onto the entity. It is returned on inlineProducts.

map

Each place has name and position (1-based). Other keys are present only when the page sent them: id, provider, address, city, state, zipcode, countryCode, countryName, latitude, longitude, location, distanceMeters, rating, reviewCount, ratingScale, reviews, reviewHighlights, categories, description, enrichedDescription, descriptionCite, price, priceStr, phone, websiteUrl, providerUrl, imageUrl, imageUrls, hours, isOpen, isClosedPermanently, isClosedTemporarily, nextOpenHour, specialHours, attributes, tags, providerLogoUrl, providerLogoDarkUrl, providerImages, yelpMenuUrl, rank, popularityScore, isClaimed, dateOpened, dateClosed, reservationProviders, fromCache. A null the page sent stays null.

ads

shoppingCards and inlineProducts

shoppingCards is an array of cards. Each card has products. A product without position is numbered from 1 across the cards. inlineProducts is an array of { "product": { ... } }. product includes the fields the page sent, such as title, url, description, price, rating, num_reviews, image_urls, provider, merchant, and offers. product.position is 1-based when the page did not set it. product.productDetails is added only when productDetails is true and that product has a rationale or reviews. A product description by itself does not add it. When the flag is false, productDetails is omitted.

When a request fails

If the call does not succeed, the HTTP status is not 200 and the body is:
Treat any non-200 response as a failed call with "error": "REQUEST_FAILED". Use the HTTP status to decide next steps:
  • 400, 401, 403, 404, 422 — fix the prompt, JSON, token, or route. Do not retry the same request.
  • 429, 502, 503, and other non-200 server-side statuses — retry later with exponential backoff (for example 1s, 2s, 4s) and stop after a few attempts. /llm does not return Retry-After on 429.
Include x-request-id from the response headers when you contact support.

Usage example

On success, read response.text (and response.markdown) from the JSON body. To ask Gemini instead, send "source":"gemini". For Google AI Overview, send "source":"google_ai_overview" with a search query as prompt, and check metadata.reason before reading the answer. For a structured ChatGPT answer, send include and read result.text, result.sources, and the other keys you turned on:

Best practices

  • Store the token in a secret instead of hardcoding.
  • Success vs failure: HTTP 200 is a completed call. Read response.text (for Google AI Overview, unless metadata.reason is not_served). With ChatGPT include, success is true and the answer is result.text, which can be "" when the page sent no text. Any other status is a failed call with "error": "REQUEST_FAILED", not "success": false.
  • Retries: Retry only transient failures (429, 502, 503, and similar). Do not retry 400, 401, 403, 404, or 422 with the same payload. For repeated failures, contact support with x-request-id.
  • Overview – Token acquisition
  • Request – HTTP and browser requests (method + url)

Try it

Use the interactive Try it panel to send a real request.

Before you start

  1. Get a token – See Token acquisition.
  2. Authenticate – Enter the token in Authorize (Bearer is applied automatically).
  3. Edit the body – Default: {"source":"chatgpt","prompt":"...","countries":["US"]}. Set source to gemini to query Gemini, or to google_ai_overview to get Google’s AI Overview for a search query.
  4. Send – Success: response.text on the LLM object.
Security: Do not share your token or commit it to code. Use the Authorize dialog only in your browser; the token is not stored in the documentation.

Authorizations

Authorization
string
header
required

Use Authorize and enter your tenant JWT in the token field only. The gateway validates the token and injects x-tenant-id from the tenantId claim; do not set x-tenant-id manually in production.

Body

application/json
prompt
string
required

Question to send to the LLM, or the Google search query for google_ai_overview. One prompt per call. Must be non-empty after trimming.

Minimum string length: 1
Pattern: .*\S.*
Example:

"What is the best proxy provider?"

source
enum<string>
required

Where to send the prompt. Required: a request without source fails with HTTP 400 (REQUEST_FAILED).

Available options:
chatgpt,
gemini,
google_ai_overview
Example:

"chatgpt"

countries
string[]

Optional ISO country codes. When set, the request is routed through that location. Defaults to ["US"] if omitted. google_ai_overview uses the first code as the Google search country.

Example:
country
string

ChatGPT include: one ISO country code. When both country and countries are sent, country is used.

Example:

"US"

include
object

ChatGPT only. When present, the 200 body is one structured result object (success is true) instead of the short LLM object. Each flag is a boolean and defaults to off. Unknown keys or non-boolean flags fail with HTTP 400.

Response

LLM answer. Outer HTTP 200 with the Bringits LLM object (response.text), or the ChatGPT include object (result) when include is set. For google_ai_overview, a results page without an AI Overview is also 200, with empty response.text and metadata.reason set to not_served.

id
string
required

Request id (llm_ plus 12 uppercase hex characters from a UUID).

Pattern: ^llm_[0-9A-F]{12}$
Example:

"llm_A1B2C3D4E5F6"

source
enum<string>
required
Available options:
chatgpt,
gemini,
google_ai_overview
model
string | null
required

Reserved. The sources do not expose a model id.

Example:

null

response
object
required
metadata
object
required
citations
object[]

Citation chips in the answer. Filled for google_ai_overview; always empty for chatgpt and gemini.

sources
object[]

google_ai_overview only: source cards shown with the overview.

followUps
string[]

google_ai_overview only: follow-up questions Google suggests.