Skip to main content
querying.ai implements the common async task contract for answer-engine APIs. For the endpoints below, request and response schemas match, so a working client usually needs only a new base URL and key.

​
The change

Point your API-key constant at a querying.ai key from API keys. Webhook URLs, idempotency keys, task types and response parsing stay as they are.

​
What is implemented

Every engine runs asynchronously; there is no endpoint that blocks until an answer arrives.

​
Where behavior differs

There is no per-key requests-per-second limiter, but three limits apply:
  • The edge allows 1,200 requests per minute per Authorization credential and returns 429 above that.
  • Your plan caps how many tasks your account can have queued or processing at once. Over the cap, submit returns 429 PLAN_CONCURRENCY_LIMIT.
  • A key can carry a daily credit limit. Over it, submit returns 429 KEY_DAILY_CREDIT_LIMIT until the next UTC midnight.
For volume, send batches instead of many single requests. See Credits and limits and Batches and grids.
Each engine has its own price: for example 1 credit for ChatGPT and Perplexity, 1.25 for Gemini and Google AI Overviews. The engine catalog lists every price.Submit responses report { "creditsToCharge": 0, "creditsCharged": 0 }. Once a task is COMPLETED, polling and the webhook report the captured charge. Failed tasks are not charged.
An engine and country with no route returns 422 REGION_UNSUPPORTED immediately, instead of accepting a task that will fail. Retrying the same pair does not help.A client that assumes every well-formed submission is accepted needs a branch for 422. See Regions and location.
payload.include flags are not honored by every engine. An unsupported markdown, rawResponse or searchQueries flag is ignored. An unsupported html or analysis flag returns 422, and an unsupported citationAnalysis flag returns 400. Check the support matrix before you send them.
Every webhook carries querying-webhook-id, querying-webhook-timestamp and querying-webhook-signature headers. Your existing receiver keeps working if it ignores them, but you should verify them. See Webhooks.
GET /v1/async/task/{id} on a FAILED task returns a top-level error field. It is a plain string, not the { code, message, timestamp } object used by request errors.
A client that treats a truthy body.error as a failed request will misread a failed task. Branch on task.status.

​
Map engines to task types

GOOGLE returns a results-page envelope with the AI Overview in a nullable aioverview member. See Google AI Overviews.

​
Verify the cutover

Run your current integration and querying.ai in parallel for a while and diff the fields you consume, usually response.text, response.sources[] and response.markdown. A shadow comparison catches differences a schema check misses, such as a source list in a different order.