OpenAI Responses API

The Responses API is a recent addition to the OpenAI platform. It allows developers to build conversational tools by referencing the previous_response_id rather than sending the full message history on every request. TeXRA supports this API as an alternative to the Chat Completions API.

Key differences

• Continuations: Provide previous_response_id from the prior response to continue a conversation. Only the new messages are sent.
• Input types: Message parts use input_text or input_image objects.
• Output format: Instead of choices, text is found in response.output[0].content[0].text (or output_text).
• Instructions: The instructions parameter applies only to the current request. When using previous_response_id, you must resend any system instructions you want applied.
• No stop sequences: The Responses API does not accept a stop parameter. If your agent requires an end tag, handle it in post-processing rather than sending it to the API.

See the OpenAI Responses documentation for full details.

Using with TeXRA

When "texra.model.useOpenAIResponsesAPI" is enabled, the extension automatically:

1. Converts chat message parts into input_text/input_image objects.
2. Tracks the last response.id and sends it as previous_response_id for subsequent rounds.
3. Reads the returned text from output_text or the output array.

This keeps requests small and simplifies conversation management.

The open-weight models gpt-oss-120b and gpt-oss-20b (available only via OpenRouter) also use the Responses API automatically, even if the setting isn't enabled.