Open Responses API doesn't accept "output_text" as an input Item

[Zoarial94]

LocalAI version:
v4.10.0 (7ad0cbf)

Environment, CPU architecture, OS, and Version:
Docker, x86, Linux

Describe the bug
Open Response API doesn't accept output_text as an Item per the spec.
https://www.openresponses.org/reference#request-body

LocalAI/core/http/endpoints/openresponses/responses.go

Line 727 in b77937f

func convertORMessageItem(itemMap map[string]any, cfg *config.ModelConfig) (schema.Message, error) {

You should be able to exactly copy the "output" JSON from one call into the "input" JSON of the next call.

To Reproduce
Start a new OR API request (/v1/responses) with output messages from a previous response.

Expected behavior
The model sees the previous responses in the context for the new response

Logs

Example input:

Input JSON

{
  "model": "gemma-4-26b-a4b-it",
  "instructions": "These are scanned hand-written notes. Combine the notes, but do not reorder or change the wording of the notes except for the footnotes and endnotes. ",
  "input": [
    {
      "id": "msg_8c030eaa-8eca-402d-9a37-3b81097dedf8",
      "role": "assistant",
      "type": "message",
      "status": "completed",
      "content": [
        {
          "text": "text 1",
          "type": "output_text",
          "logprobs": [],
          "annotations": []
        }
      ],
      "summary": [],
      "arguments": ""
    },
    {
      "id": "msg_936e093d-e8b7-46b1-a3ed-c5058f5092af",
      "role": "assistant",
      "type": "message",
      "status": "completed",
      "content": [
        {
          "text": "text 2",
          "type": "output_text",
          "logprobs": [],
          "annotations": []
        }
      ],
      "summary": [],
      "arguments": ""
    },
    {
      "id": "msg_259556ea-700d-42e6-905a-9897fb6674d9",
      "role": "assistant",
      "type": "message",
      "status": "completed",
      "content": [
        {
          "text": "text 3",
          "type": "output_text",
          "logprobs": [],
          "annotations": []
        }
      ],
      "summary": [],
      "arguments": ""
    },
    {
      "id": "msg_427a8eff-6185-402e-99a6-1a93c3b3f705",
      "role": "assistant",
      "type": "message",
      "status": "completed",
      "content": [
        {
          "text": "text 4",
          "type": "output_text",
          "logprobs": [],
          "annotations": []
        }
      ],
      "summary": [],
      "arguments": ""
    }
  ],
  "temperature": 0.8
}

Example Backend Input:

Backend Input

[
  {
    "role": "system",
    "content": "These are scanned hand-written notes. Combine the notes, but do not reorder or change the wording of the notes except for the footnotes and endnotes. "
  }
]

Backend Output

{
  "id": "resp_ac7c7791-7680-4685-a4cc-c206264b073f",
  "object": "response",
  "created_at": 1790610865,
  "completed_at": 1790610910,
  "status": "completed",
  "model": "gemma-4-26b-a4b-it",
  "output": [
    {
      "type": "message",
      "id": "msg_8f7ba482-ca80-4cc9-8633-9d71dd0e8fa3",
      "status": "completed",
      "role": "assistant",
      "content": [
        {
          "type": "output_text",
          "text": "Please provide the scanned hand-written notes (by uploading the images or providing the text). \n\nOnce you provide them, I will combine them according to your exact constraints:\n1. **No reordering:** I will keep the notes in their original sequence.\n2. **No wording changes:** I will not change a single word in the body of the notes.\n3. **Footnotes/Endnotes exception:** I will only adjust the footnotes and endnotes to ensure they are correctly placed or formatted according to your requirements.",
          "annotations": [],
          "logprobs": []
        }
      ],
      "arguments": "",
      "summary": []
    }
  ],
  "error": null,
  "incomplete_details": null,
  "previous_response_id": null,
  "instructions": "These are scanned hand-written notes. Combine the notes, but do not reorder or change the wording of the notes except for the footnotes and endnotes. ",
  "tools": [],
  "tool_choice": "none",
  "parallel_tool_calls": true,
  "max_tool_calls": null,
  "temperature": 0.8,
  "top_p": 1,
  "presence_penalty": 0,
  "frequency_penalty": 0,
  "top_logprobs": 0,
  "max_output_tokens": null,
  "text": {
    "format": {
      "type": "text"
    }
  },
  "truncation": "auto",
  "reasoning": null,
  "usage": {
    "input_tokens": 43,
    "output_tokens": 1546,
    "total_tokens": 1589,
    "input_tokens_details": {
      "cached_tokens": 0
    },
    "output_tokens_details": {
      "reasoning_tokens": 0
    }
  },
  "metadata": {},
  "store": true,
  "background": false,
  "service_tier": "default",
  "safety_identifier": null,
  "prompt_cache_key": null
}

Additional context
I believe | "output_text" (Pseudo code since I am unfamiliar with go) needs to be added here as a quick solution, but I'm unsure if it will cause downstream problems.

LocalAI/core/http/endpoints/openresponses/responses.go

Lines 750 to 754 in b77937f

	switch partType {
	case "input_text":
	if text, ok := partMap["text"].(string); ok {
	textContent += text
	}

[localai-org-maint-bot]

I have picked this up. Confirmed on latest master (84a2b20): request message conversion handles input_text but skips output_text, so replayed assistant messages lose their text. I will add a focused fix, regression tests for replaying serialized response output, and a documentation example.