[QUESTION] - tool_search support

[mjmustafaev]

How can we help?

Is tool_search, as noted below, supported through the library? If not, what are the future-proof suggestions to hand-build it for now?

https://developers.openai.com/api/docs/guides/tools-tool-search

[jsquire]

Hi @mjmustafaev. Thanks for reaching out and for your inquiry. The tool_search tool has not yet been added to the Responses convenience surface. Until it is added, you can either use the protocol method overload, the JSON patch API, or deserialize JSON as the base tool. The "base tool" approach is demonstrated below.

Guide example without namespace tool

var client = new ResponsesClient("<< YOUR API KEY >>");

// Create the CRM tool
var crmTool = ResponseTool.CreateFunctionTool(
  functionName: "list_open_orders",
  functionDescription: "List open orders for a customer ID.",
  functionParameters: BinaryData.FromObjectAsJson(new
  {
      type = "object",
      properties = new
      {
          customer_id = new { type = "string" }
      },
      required = new[] { "customer_id" },
      additionalProperties = false
  }),
  strictModeEnabled: false);

// Defer loading is a new property which needs to be patched in        
crmTool.Patch.Set("$.defer_loading"u8, true);

// The tool search tool can most easily be represented as a base response tool
var toolSearchTool = ModelReaderWriter.Read<ResponseTool>(BinaryData.FromObjectAsJson(new
{
    type = "tool_search"
}));

// Both tools are added to the options when creating the response
var promptItem = ResponseItem.CreateUserMessageItem("Tell me the story of the three little pigs - with enthusiasm!");

var options = new CreateResponseOptions("gpt-5.4", [ promptItem ])
{
    Instructions = "You are Terry Crews and have had 12 cans of Red Bull today.  You are PUMPED to help!"        
};

options.Tools.Add(crmTool);
options.Tools.Add(toolSearchTool);

var response = await client.CreateResponseAsync(options);
Console.WriteLine(response.Value.GetOutputText());

Guide example exactly

There's not much purpose to the namespace tool in the example, but if you wanted to reproduce the guide exactly as written, the easiest approach would be to create the namespace tool directly with the function tool embedded. As this is also a newer tool that does not yet have representation in the convenience API, the easiest approach is also to represent it as a base response tool.

var client = new ResponsesClient("<< YOUR API KEY >>");

// Create the namespace tool with embedded function tool
var namespaceTool = ModelReaderWriter.Read<ResponseTool>(BinaryData.FromObjectAsJson(new
{
    type = "namespace",
    name = "crm",
    description = "CRM tools for customer lookup and order management.",
    tools = new object[]
    {
        new
        {
            type = "function",
            name = "list_open_orders",
            description =  "List open orders for a customer ID.",
            defer_loading = true,
            parameters = new
            {
                type = "object",
                properties = new
                {
                    customer_id = new { type = "string" }
                },
                required = new[] { "customer_id" },
                additionalProperties = false
          }
        }
    }
}));

var toolSearchTool = ModelReaderWriter.Read<ResponseTool>(BinaryData.FromObjectAsJson(new
{
    type = "tool_search"
}));

// Both tools are added to the options when creating the response
var promptItem = ResponseItem.CreateUserMessageItem("Tell me the story of the three little pigs - with enthusiasm!");

var options = new CreateResponseOptions("gpt-5.4", [ promptItem ])
{
    Instructions = "You are Terry Crews and have had 12 cans of Red Bull today.  You are PUMPED to help!"        
};

options.Tools.Add(namespaceTool );
options.Tools.Add(toolSearchTool);

var response = await client.CreateResponseAsync(options);
Console.WriteLine(response.Value.GetOutputText());

[mjmustafaev]

Got it, thanks very much, @jsquire !! 👍🏼