This article lists every OpenAI endpoint the integration calls. Its main purpose is deprecation tracking: if OpenAI announces a change to one of these endpoints, this list shows immediately whether the integration is affected. It is also useful for firewall allow-listing, for scoping the API key's permissions, and for troubleshooting.

Subscribers manage their OpenAI account, API keys and webhooks in the OpenAI console at https://platform.openai.com.

OpenAI exposes a JSON REST API. The integration uses a single area of it — the Responses API — plus one lightweight call used to confirm the connection.

The base URL is set per subscription in the API Url setting and ships pre-filled with OpenAI's production address:

Throughout this article [Base URL] stands for whatever that setting holds, so the paths below apply regardless of which address a subscription is pointed at. OpenAI does not publish separate sandbox and production hosts; there is one address, and separating test from live traffic is done with separate API keys or projects.

Areas covered: Response Endpoints · Model Endpoints · Authentication · Rate Limits and Concurrency · Authorization Endpoint · Pagination

The Responses API is where all enrichment happens. A request is submitted when a Customer or Product is transferred to OpenAI, and the completed result is fetched back when OpenAI reports it is ready.

Used during FROM iPaaS.com transfers to submit a record for enrichment. The request carries the record as JSON together with the instruction configured in the collection's PromptText mapping.

Parameters sent on this call:

Subscribers can supply additional parameters through a collection's ModelInstructions mapping, which accepts a JSON object. Values supplied there are applied first and the parameters above are then set by the integration, so model, input, background, store and metadata cannot be overridden.

Used during TO iPaaS.com transfers to fetch a completed enrichment. {response_id} is the identifier OpenAI assigned when the response was created — the same value stored on the enriched record in its OpenAI Last Response ID custom field, and the value entered on the Manual Sync page for a TO collection.

Because the integration submits requests with store set to false, OpenAI retains the response only temporarily to support background execution. A response fetched long after it completed may no longer be available.

Used to validate the connection. When subscription settings are saved, the integration makes this call to confirm the API key is accepted before reporting the connection as established. It returns the models available to the account.

This is the only call the integration makes outside the Responses API, and it reads nothing into iPaaS.com — the response is used solely as a success or failure signal.

OpenAI uses a secret API key presented as a bearer token on every request. There is no OAuth flow, no consent screen, and therefore no authorization endpoint — the key is created once in the OpenAI console and entered on the subscription.

The key must carry at least Write permission on the Responses capability. A key restricted below that is accepted when the connection is validated, because listing models needs fewer permissions than creating a response, but enrichment then fails at transfer time.

Requests travelling in the other direction — OpenAI calling iPaaS.com to report a completed enrichment — are authenticated separately, using the Webhook API Key embedded in the webhook address the subscriber configures in the OpenAI console. See the OpenAI Installation Instructions.

OpenAI applies rate limits at the organization and project level rather than per user, and the limits vary by model. They are measured across several dimensions at once — requests per minute, requests per day, tokens per minute and tokens per day — and whichever is reached first applies. Every response carries headers reporting the applicable limits, what remains, and when each resets. Current limits for an account are shown in the OpenAI console.

When OpenAI reports that a limit has been reached, the transfer is rescheduled automatically and resumes once the limit window has passed. No subscriber action is required and no data is lost.

On the iPaaS.com side, subscribers or their MiSP control how hard the integration pushes using API Throttle Limit, API Throttle Seconds, Concurrent Connections and Concurrent Batch Executions. These are described in the OpenAI Connections and Settings article. Setting them conservatively during initial configuration reduces the chance of reaching OpenAI's limits while mappings are still being validated.

Source: OpenAI Rate limits guide, retrieved 29 July 2026 — https://developers.openai.com/api/docs/guides/rate-limits

Not applicable. OpenAI authenticates with a secret API key presented on each request rather than an OAuth flow, so there is no token endpoint for the integration to call and no authorization URL that varies by region. The key is created once in the OpenAI console and entered on the subscription — see Authentication above.

Not applicable. Every call the integration makes returns a single record or a short fixed list, so no pagination parameters are used.