OpenAI API connection options
OpenAI accounts can use Responses API or Chat Completions and pass incremental replies to AI features that support streaming.
What it is
Provider API chooses which OpenAI API an account uses. INSIDE supports Responses API and Chat Completions. Both can return a complete reply or incremental output for AI features that support streaming.
Who uses it and why
Developers agree the API and model compatibility with Powerfront. Operators use the resulting AI replies in their enabled features; they do not choose an API for each conversation.
Where to find it
Powerfront manages Provider API in Settings › AI › Accounts. Ask Powerfront to change the API choice, endpoint or credentials for you.
How it works
- The feature sends its instructions and conversation through its account's selected OpenAI API.
- Where the model supports them, INSIDE includes tool requests, tool results, images, files and a structured reply format.
- With streaming, INSIDE assembles arriving text and tool arguments as the model produces them. It also receives progress, usage and completion information when supplied.
- The AI feature decides how to display or use that output. Receiving a partial reply does not establish that generation has finished.
For an explicitly selected OpenAI provider, Default selects Responses API. Older configurations without an explicit provider retain Chat Completions selection behaviour. Ask Powerfront to check the effective configuration when comparing older and newer accounts.
Configure it
Ask Powerfront to set Provider API to Default, Responses API or Chat Completions and confirm the selected Model supports your feature's inputs and tools.
Reasoning Effort offers Default, None, Low, Medium, High and XHigh. Verbosity offers Default, Low, Medium and High. The default choice leaves these options to the service. Model compatibility still applies.
A custom API Endpoint must accept the selected API format. Ask Powerfront to verify compatibility before using it with your AI features.
Settings
| Setting | Where | Applies to | Default | Who can change it | What it does |
|---|---|---|---|---|---|
| Provider API | Settings › AI › Accounts | website | Responses API (Default) | Powerfront | For an explicit OpenAI provider, Default selects Responses API; Chat Completions can be selected separately. |
| Reasoning Effort | Settings › AI › Accounts | website | Not set: the model service decides | Powerfront | Requests the selected reasoning effort from a model that supports it; Default leaves the service's choice unchanged. |
| Verbosity | Settings › AI › Accounts | website | Not set: the model service decides | Powerfront | Requests Low, Medium or High response detail from a model that supports the option. |
| API Endpoint | Settings › AI › Accounts | website | Not set | Powerfront | Overrides the OpenAI service address; an unset value uses the built-in OpenAI address for the selected API. |
Fields
| Field | Where | Required | What it is for |
|---|---|---|---|
| Model | Settings › AI › Accounts | Yes | Selects the configured OpenAI model for an AiFlow account. |
| API Key | Settings › AI › Accounts | No | Supplies the credential used for the OpenAI connection. |
Limits and good to know
- Streaming delivery to an operator or visitor depends on the AI feature, not just the account's API choice.
- A reply can stop because the provider's output limit was reached. INSIDE records supported token-limit stops as incomplete.
- Content filtering or a stream error can also interrupt output. A partial reply should not be treated as a completed action.
- Older account generation controls are not all applied by these newer OpenAI connections. Ask Powerfront to verify any required output limit for the actual execution path.
- Token information can include cached and reasoning usage when the service provides it. Those details are not available for every request.
- The presence of other OpenAI services does not make audio, image generation or realtime voice part of this account option.
Depends on and works with
View all connections as a list
This feature uses
- AI functions and required inputsEnhances
Supported models can request functions exposed by the feature.
- Rich content in AI repliesEnhances
Enabled rich replies let compatible OpenAI models return validated structured content.
- AI reply streaming settingsConfigured by
The AI feature's streaming options control whether supported incremental output is delivered progressively.
- Images and attachments for AI agentsEnhances
Supported models can receive image and file context.
- AI accounts and shared defaultsRequires
The account supplies model selection, API choice and credentials.
- AI model catalogueConfigured by
The account selects a website model with an OpenAI provider.
- Cloudflare AI Gateway connectionEnhances
OpenAI requests can be routed through a configured gateway.
Used by
- AI model provider selectionEnhances
OpenAI is an available provider with two API choices.
- Cloudflare AI Gateway connectionEnhances
OpenAI Chat Completions and Responses requests can use the gateway.
FAQ and troubleshooting
Does Default mean the same API on every account?
For the explicit OpenAI provider it means Responses API. Older configurations can retain Chat Completions behaviour. Powerfront can confirm your account's selection.
Why does my reply arrive all at once?
The AI feature may use complete replies, or its delivery path may not display incremental output. Ask Powerfront to check both the account and feature.
Why does the reply stop partway through?
An output limit, content filter or connection error may have stopped it. Give Powerfront the affected feature and conversation so the completion can be checked.
Will another OpenAI-compatible endpoint work automatically?
It must support the selected API format and the feature's required capabilities. Powerfront needs to verify that connection.