Open AI response api and assistant api

[JP-droidic]

Hi

Are you able to share with me an example of response API with threadID and also assistant API? It would be interesting to see calling built-in e.g. web search, coding, passing vectoriId and referring and downloading files generated.

Also, do you have a simple example of a web-based conversation? Where the user asks a question, the LLM answers and then in a second call we need to rebuild the full chat history and pass it to the LLM.

Sorry for so many questions, but it would be great to see examples ,especially for us moving from SK.

[lofcz]

No worries, I love providing unpaid consultancy.

Are you able to share with me an example of response API with threadID and also assistant API? It would be interesting to see calling built-in e.g. web search, coding, passing vectoriId and referring and downloading files generated.

Tornado has a unique feature which enables using Chat with Responses exclusive models/parameters - we can automatically route your requests via /chat or /responses, so I am not sure what you'd like to see. There is nothing like threadID in the OpenAI's Responses API. Responses have PreviousResponseId - do you mean that? Please see whether some of these demos do what you are looking for:
https://github.com/lofcz/LlmTornado/blob/master/src/LlmTornado.Demo/ResponsesDemo.cs

Also, do you have a simple example of a web-based conversation? Where the user asks a question, the LLM answers and then in a second call we need to rebuild the full chat history and pass it to the LLM.

Use Conversation, do your web searching either via overpriced built-in tools, or use something like Tavily, add the messages to the conversation - we handle "rebuilding" for you. Here is an example of doing this with Grok (this uses their live-search):
https://github.com/lofcz/LlmTornado/blob/5e5350f15058686fa093d5e5f8fa35daeae7f6f9/src/LlmTornado.Demo/ChatDemo2.cs#L326-L381

Here is a minimal example*:

[TornadoTest]
public static async Task MockWebSearch()
{
    Conversation chat = Program.Connect().Chat.CreateConversation(new ChatRequest
    {
        Model = ChatModel.Google.Gemini.Gemini25Pro,
        Tools = [
            new Tool(async (
                [Description("The query to search for")] string query, 
                ToolArguments args) =>
            {
                // use Tavily, or whatever
                return new
                {
                    results = new List<string>
                    {
                        """
                        2025-08-14, Version 24.6.0 (Current), @RafaelGSS

                        Notable Changes
                        [471fe712b3] - (SEMVER-MINOR) cli: add NODE_USE_SYSTEM_CA=1 (Joyee Cheung) #59276
                        [38aedfbf73] - (SEMVER-MINOR) crypto: support ML-DSA KeyObject, sign, and verify (Filip Skokan) #59259
                        [201304537e] - (SEMVER-MINOR) zlib: add dictionary support to zstdCompress and zstdDecompress (lluisemper) #59240
                        [e79c93a5d0] - (SEMVER-MINOR) http: add server.keepAliveTimeoutBuffer option (Haram Jeong) #59243
                        [c144d69efc] - lib: docs deprecate _http_* (Sebastian Beltran) #59293
                        [aeb4de55a7] - (SEMVER-MINOR) fs: port SonicBoom module to fs module as Utf8Stream (James M Snell) #58897
                        """
                    }
                };
            }, "search_internet", "Searches the internet")
        ]
    });
    
    chat.AddSystemMessage("You are an assistant. Search internet when appropriate to answer queries. Reply curtly.");
    chat.AddUserMessage("What is the latest NodeJS version?");

    // run the first conversation round, here we naively suppose the model will use search_internet
    // ToolCallsHandler.ContinueConversation <-- passing this in results in the result of the tool call being added to the conversation
    Console.WriteLine(await chat.GetResponseRich(ToolCallsHandler.ContinueConversation));
    
    // run round two, the model will use the result of the internet search to inform the user about what it found
    Console.WriteLine(await chat.GetResponseRich(ToolCallsHandler.ContinueConversation));
}

*please use 3.7.29 for this example, there was some edge case bug with async tools which aren't really async I found when writing this.

If you like this way of writing tools, see this file: https://github.com/lofcz/LlmTornado/blob/master/src/LlmTornado.Demo/InfraDemo.cs

This prints something like this:

[block: Function]
name: search_internet
arguments:
{
  "query": "latest NodeJS version"
}
[block: Reasoning]
**Navigating the NodeJS Landscape**

Okay, so I need to find the most current NodeJS version.  That's straightforward enough – I'll fire off a quick internet search. The tool I need is `search_internet`, which I know handles this kind of query.  The only parameter it requires is, naturally, the `query`.  And that's simply "latest NodeJS version."  To get the answer, I'll execute `print(search_internet(query="latest NodeJS version"))`.  That should give me the information I'm after, directly and efficiently. No need to overcomplicate things!


[block: Reasoning]
**Determining the Current NodeJS Version**

Okay, so I need to find the current, most up-to-date version of NodeJS. Given that this information is constantly changing, a direct search is the way to go.  I'll leverage the "search_internet" tool and use the query "latest NodeJS version."  That should be precise enough to retrieve the necessary information quickly.  After the search runs, I'll analyze the results.  I'll look for a clear indication of the latest release, extract that version number, and present it as the definitive answer.

[block: Message]
24.6.0

Demo finished

and also assistant API

If by assistants you mean the obsolete technology OpenAI will remove soon, we have demos for that here:
https://github.com/lofcz/LlmTornado/blob/master/src/LlmTornado.Demo/AssistantsDemo.cs