Responses Interaction Example

Responses Interaction Example

Back to OpenAI Responses.

This compact example mirrors the detailed Chinese walkthrough. It shows why Responses is item-oriented: hosted tools, function calls, and assistant messages all appear as output items.

1. 1
   client request

   Client sends `input`, a `function` tool, a hosted `web_search` tool, and `stream: true`.

2. 2
   response.created

   The stream starts a response resource with `status: in_progress`.

3. 3
   hosted tool events

   `response.web_search_call.*` events report hosted tool lifecycle by `item_id` and `output_index`.

4. 4
   function_call item

   `response.output_item.added` creates a function call item; argument deltas join by `item_id`.

5. 5
   function_call_output

   The next request sends the local tool result using the function call `call_id`.

6. 6
   final message item

   The final turn emits a message item, text deltas, `output_text.done`, and `response.completed`.

Join keys

Entity	Stable key	Notes
response	response.id	Can connect follow-up calls via previous_response_id.
output item	item_id plus output_index	item_id is the primary join key for deltas.
function result	call_id	The client returns tool output with function_call_output.call_id.
content part	item_id + content_index	Text deltas and done events target one content part.

ProxAI behavior

For openai_responses -> openai_responses, ProxAI preserves SSE bytes while the Responses observer watches terminal events and streamed tool-call arguments for compact diagnostics.