Examples for files upload into conversations in GPTs API

Starting a Conversation

In order to upload files using GPTs API, you need to a new conversation using gpt-4o, then to get the conversationId from the event-stream.

Link to doc https://www.goapi.ai/docs/gpts-api

Call the endpoint with the following sample cURL code

Then data from the event stream is as follows:

data: {"message": {"id": "b63eacbc-a225-4dee-9498-eb62a5950786", "author": {"role": "assistant", "name": null, "metadata": {}}, "create_time": 1727273868.262348, "update_time": null, "content": {"content_type": "text", "parts": ["Hello! How can I assist you today?"]}, "status": "in_progress", "end_turn": null, "weight": 1.0, "metadata": {"citations": [], "content_references": [], "gizmo_id": null, "message_type": "next", "model_slug": "gpt-4o", "default_model_slug": "gpt-4o", "pad": "AAAAAAAAAAAAAAAAAAAAAAAAAA", "parent_id": "8cc00bc8-c9d3-4362-841e-ffbed24aca60", "model_switcher_deny": []}, "recipient": "all", "channel": null}, "conversation_id": "66f41b8c-0d58-800b-8dfb-da993cc5fbfb", "error": null}

Then, the user should upload a file using this conversation_id (66f41b8c-0d58-800b-8dfb-da993cc5fbfb), Content-Type should be form-data.

The response should be below:

{
    "code": 200,
    "data": {
        "file_name": "DINSTXPN8D]HG280Y(BCX`2.png",
        "file_id": "file-gEHVo1KU9irGjZCbdARYUFy2",
        "file_size": 121851
    },
    "message": "success"
}

In order to use this file In the following conversations, we will need to start a conversation using this file. Depending the file format, the API call will differ.

Creating a conversation with file

We will use the previous endpoint to start the conversation with the file, with the same conversationID.

Conversation with image files

Image file is the most unique file format to be used in the conversation, since only for this file format would we use multimodal_text of content_type (perhaps it is due to the web version of ChatGPT would also need to displayed the uploaded picture as well?).

{
    "content": {
        "content_type": "multimodal_text",
        "parts": [
            {
                "content_type": "image_asset_pointer",
                "asset_pointer": "file-service://file-gEHVo1KU9irGjZCbdARYUFy2",
                "size_bytes": 121851,
                "width": 128,
                "height": 128
            },
            "describe this image"
        ]
    },
    "metadata": {
        "attachments": [
            {
                "name": "DINSTXPN8D]HG280Y(BCX`2.png",
                "id": "file-gEHVo1KU9irGjZCbdARYUFy2",
                "size": 121851,
                "mimeType": "image/png",
                "width": 128,
                "height": 128
            }
        ]
    }
}

To note:

“asset_pointer” and its values (which is the same with the value for “id” within “metadata”)

“mimeType”

“parts” and its associated content.

If you need to use multiple image files, you could upload multiple times, as shown below.


      "content": {
        "content_type": "multimodal_text",
        "parts": [
          {
            "content_type": "image_asset_pointer",
            "asset_pointer": "file-service://file-otNCSKxKA5LwzE839wbJZbTc",
            "size_bytes": 57766,
            "width": 271,
            "height": 325
          },
          {
            "content_type": "image_asset_pointer",
            "asset_pointer": "file-service://file-HO3i69pHaQZEiWdKzUAxAr9M",
            "size_bytes": 8205,
            "width": 128,
            "height": 128
          },
          "Test_test_test"
        ]
      },
      "metadata": {
        "attachments": [
          {
            "id": "file-otNCSKxKA5LwzE839wbJZbTc",
            "size": 57766,
            "name": ")JUM75}DS$IG`OESMUGD3Q8.png",
            "mime_type": "image/png",
            "width": 271,
            "height": 325
          },
          {
            "id": "file-HO3i69pHaQZEiWdKzUAxAr9M",
            "size": 8205,
            "name": "mai.png",
            "mime_type": "image/png",
            "width": 128,
            "height": 128
          }
        ]
      }

Conversation with txt files

This largely follows the same steps as the image file workflow.

With regarding to file_size_token, as per our test, the workflow should work without this parameter. Thus we are not going to check for this parameter specifically.

You need to find the corresponding mime_type by using ChatGPT Web Version, and using Developer’s Tool (pressing F12 for windows users) to check for the metadata within conversation.

{ 
    "content": {
        "content_type": "text",
        "parts": [
            "describe this file"
        ]
    },
    "metadata": {
        "attachments": [
            {
                "name": "hello.txt",
                "id": "file-WHgV7URVkoI7iJnpgg0dVU2D",
                "size": 12,
                //"file_size_tokens": 4,
                "mime_type": "text/plain"
            }
        ]
    }
}

Conversation with PDF files

{
    "content": {
        "content_type": "text",
        "parts": [
            "describe this file"
        ]
    },
    "metadata": {
        "attachments": [
            {
                "name": "hello.pdf",
                "id": "file-WHgV7URVkoI7iJnpgg0dVU2D",
                "size": 12,
                "mime_type": "application/pdf"
            }
        ]
    }
}

Conversation with word files

{
    "content": {
        "content_type": "text",
        "parts": [
            "describe this file"
        ]
    },
    "metadata": {
        "attachments": [
            {
                "name": "hello.docx",
                "id": "file-WHgV7URVkoI7iJnpgg0dVU2D",
                "size": 12,
                "mime_type": "application/pdf"
            }
        ]
    }
}

Conversation with excel files

{
    "content": {
        "content_type": "text",
        "parts": [
            "describe this file"
        ]
    },
    "metadata": {
        "attachments": [
            {
                "name": "work.xlsx",
                "id": "file-WHgV7URVkoI7iJnpgg0dVU2D",
                "size": 121,
                "mime_type": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet"
            }
        ]
    }
}

Subsequent Conversations

Because GPTs has "conversations", thus if you have already incorporated a file into a conversation, you no longer need to do it again for the same conversation.

You don’t need to use the metadata parameter either.

{
    "parent_message_id": "",
    "content": {
        "content_type": "text",
        "parts": [
          "What is the picture I just sent"
        ]
    }
}

Examples for files upload into conversations in GPTs API

Starting a Conversation#

Creating a conversation with file#

Conversation with image files#

Conversation with txt files#

Conversation with PDF files#

Conversation with word files#

Conversation with excel files#

Subsequent Conversations#