Improved the OpenAI provider (#548)

app/MindWork AI Studio/Agents/AgentBase.cs

@@ -73,7 +73,6 @@ public abstract class AgentBase(ILogger<AgentBase> logger, SettingsManager setti
         WorkspaceId = Guid.Empty,
         ChatId = Guid.NewGuid(),
         Name = string.Empty,
-        Seed = this.RNG.Next(),
         SystemPrompt = systemPrompt,
         Blocks = [],
     };

app/MindWork AI Studio/Assistants/AssistantBase.razor.cs

@@ -21,9 +21,6 @@ public abstract partial class AssistantBase<TSettings> : AssistantLowerBase wher
     [Inject]
     protected IJSRuntime JsRuntime { get; init; } = null!;
     
-    [Inject]
-    protected ThreadSafeRandom RNG { get; init; } = null!;
-    
     [Inject]
     protected ISnackbar Snackbar { get; init; } = null!;
     
@@ -199,7 +196,6 @@ public abstract partial class AssistantBase<TSettings> : AssistantLowerBase wher
             WorkspaceId = Guid.Empty,
             ChatId = Guid.NewGuid(),
             Name = string.Format(this.TB("Assistant - {0}"), this.Title),
-            Seed = this.RNG.Next(),
             Blocks = [],
         };
     }
@@ -215,7 +211,6 @@ public abstract partial class AssistantBase<TSettings> : AssistantLowerBase wher
             WorkspaceId = workspaceId,
             ChatId = chatId,
             Name = name,
-            Seed = this.RNG.Next(),
             Blocks = [],
         };
         

app/MindWork AI Studio/Chat/ChatThread.cs

@@ -60,11 +60,6 @@ public sealed record ChatThread
     /// </summary>
     public string Name { get; set; } = string.Empty;
     
-    /// <summary>
-    /// The seed for the chat thread. Some providers use this to generate deterministic results.
-    /// </summary>
-    public int Seed { get; init; }
-    
     /// <summary>
     /// The current system prompt for the chat thread.
     /// </summary>

app/MindWork AI Studio/Components/ChatComponent.razor.cs

@@ -34,9 +34,6 @@ public partial class ChatComponent : MSGComponentBase, IAsyncDisposable
     [Inject]
     private ILogger<ChatComponent> Logger { get; set; } = null!;
     
-    [Inject]
-    private ThreadSafeRandom RNG { get; init; } = null!;
-    
     [Inject]
     private IDialogService DialogService { get; init; } = null!;
     
@@ -436,7 +433,6 @@ public partial class ChatComponent : MSGComponentBase, IAsyncDisposable
                 ChatId = Guid.NewGuid(),
                 DataSourceOptions = this.earlyDataSourceOptions,
                 Name = this.ExtractThreadName(this.userInput),
-                Seed = this.RNG.Next(),
                 Blocks = this.currentChatTemplate == ChatTemplate.NO_CHAT_TEMPLATE ? [] : this.currentChatTemplate.ExampleConversation.Select(x => x.DeepClone()).ToList(),
             };
             
@@ -674,7 +670,6 @@ public partial class ChatComponent : MSGComponentBase, IAsyncDisposable
                 WorkspaceId = this.currentWorkspaceId,
                 ChatId = Guid.NewGuid(),
                 Name = string.Empty,
-                Seed = this.RNG.Next(),
                 Blocks = this.currentChatTemplate == ChatTemplate.NO_CHAT_TEMPLATE ? [] : this.currentChatTemplate.ExampleConversation.Select(x => x.DeepClone()).ToList(),
             };
         }

app/MindWork AI Studio/Components/Workspaces.razor.cs

@@ -16,9 +16,6 @@ public partial class Workspaces : MSGComponentBase
     [Inject]
     private IDialogService DialogService { get; init; } = null!;
     
-    [Inject]
-    private ThreadSafeRandom RNG { get; init; } = null!;
-    
     [Inject]
     private ILogger<Workspaces> Logger { get; init; } = null!;
     
@@ -576,7 +573,6 @@ public partial class Workspaces : MSGComponentBase
             WorkspaceId = workspaceId,
             ChatId = Guid.NewGuid(),
             Name = string.Empty,
-            Seed = this.RNG.Next(),
             SystemPrompt = SystemPrompts.DEFAULT,
             Blocks = [],
         };

app/MindWork AI Studio/Pages/Writer.razor.cs

@@ -77,7 +77,6 @@ public partial class Writer : MSGComponentBase
             WorkspaceId = Guid.Empty,
             ChatId = Guid.NewGuid(),
             Name = string.Empty,
-            Seed = 798798,
             SystemPrompt =  """
                             You are an assistant who helps with writing documents. You receive a sample
                             from a document as input. As output, you provide how the begun sentence could

app/MindWork AI Studio/Provider/AlibabaCloud/ProviderAlibabaCloud.cs

@@ -36,7 +36,7 @@ public sealed class ProviderAlibabaCloud(ILogger logger) : BaseProvider("https:/
         };
         
         // Prepare the AlibabaCloud HTTP chat request:
-        var alibabaCloudChatRequest = JsonSerializer.Serialize(new ChatRequest
+        var alibabaCloudChatRequest = JsonSerializer.Serialize(new ChatCompletionAPIRequest
         {
             Model = chatModel.Id,
             
@@ -77,7 +77,7 @@ public sealed class ProviderAlibabaCloud(ILogger logger) : BaseProvider("https:/
             return request;
         }
         
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("AlibabaCloud", RequestBuilder, token))
+        await foreach (var content in this.StreamChatCompletionInternal<ChatCompletionDeltaStreamLine, NoChatCompletionAnnotationStreamLine>("AlibabaCloud", RequestBuilder, token))
             yield return content;
     }
 
@@ -156,7 +156,9 @@ public sealed class ProviderAlibabaCloud(ILogger logger) : BaseProvider("https:/
                     Capability.AUDIO_INPUT, Capability.SPEECH_INPUT,
                     Capability.VIDEO_INPUT,
 
-                    Capability.TEXT_OUTPUT, Capability.SPEECH_OUTPUT
+                    Capability.TEXT_OUTPUT, Capability.SPEECH_OUTPUT,
+                    
+                    Capability.CHAT_COMPLETION_API,
                 ];
             
             // Check for Qwen 3:
@@ -166,7 +168,8 @@ public sealed class ProviderAlibabaCloud(ILogger logger) : BaseProvider("https:/
                     Capability.TEXT_INPUT,
                     Capability.TEXT_OUTPUT,
                     
-                    Capability.OPTIONAL_REASONING, Capability.FUNCTION_CALLING
+                    Capability.OPTIONAL_REASONING, Capability.FUNCTION_CALLING,
+                    Capability.CHAT_COMPLETION_API,
                 ];
             
             if(modelName.IndexOf("-vl-") is not -1)
@@ -174,6 +177,8 @@ public sealed class ProviderAlibabaCloud(ILogger logger) : BaseProvider("https:/
                 [
                     Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
                     Capability.TEXT_OUTPUT,
+                    
+                    Capability.CHAT_COMPLETION_API,
                 ];
         }
         
@@ -185,7 +190,8 @@ public sealed class ProviderAlibabaCloud(ILogger logger) : BaseProvider("https:/
                 Capability.TEXT_INPUT, 
                 Capability.TEXT_OUTPUT,
                 
-                Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING
+                Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING,
+                Capability.CHAT_COMPLETION_API,
             ];
         }
         
@@ -197,7 +203,8 @@ public sealed class ProviderAlibabaCloud(ILogger logger) : BaseProvider("https:/
                 Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
                 Capability.TEXT_OUTPUT,
                 
-                Capability.ALWAYS_REASONING
+                Capability.ALWAYS_REASONING,
+                Capability.CHAT_COMPLETION_API,
             ];
         }
 
@@ -207,7 +214,8 @@ public sealed class ProviderAlibabaCloud(ILogger logger) : BaseProvider("https:/
             Capability.TEXT_INPUT,
             Capability.TEXT_OUTPUT,
             
-            Capability.FUNCTION_CALLING
+            Capability.FUNCTION_CALLING,
+            Capability.CHAT_COMPLETION_API,
         ];
     }
     

app/MindWork AI Studio/Provider/Anthropic/ProviderAnthropic.cs

@@ -72,7 +72,7 @@ public sealed class ProviderAnthropic(ILogger logger) : BaseProvider("https://ap
             return request;
         }
         
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("Anthropic", RequestBuilder, token))
+        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine, NoChatCompletionAnnotationStreamLine>("Anthropic", RequestBuilder, token))
             yield return content;
     }
 
@@ -122,7 +122,9 @@ public sealed class ProviderAnthropic(ILogger logger) : BaseProvider("https://ap
                 Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
                 Capability.TEXT_OUTPUT,
                 
-                Capability.OPTIONAL_REASONING, Capability.FUNCTION_CALLING];
+                Capability.OPTIONAL_REASONING, Capability.FUNCTION_CALLING,
+                Capability.CHAT_COMPLETION_API,
+            ];
         
         // Claude 3.7 is able to do reasoning:
         if(modelName.StartsWith("claude-3-7"))
@@ -130,7 +132,9 @@ public sealed class ProviderAnthropic(ILogger logger) : BaseProvider("https://ap
                 Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
                 Capability.TEXT_OUTPUT,
                 
-                Capability.OPTIONAL_REASONING, Capability.FUNCTION_CALLING];
+                Capability.OPTIONAL_REASONING, Capability.FUNCTION_CALLING,
+                Capability.CHAT_COMPLETION_API,
+            ];
         
         // All other 3.x models are able to process text and images as input:
         if(modelName.StartsWith("claude-3-"))
@@ -138,13 +142,17 @@ public sealed class ProviderAnthropic(ILogger logger) : BaseProvider("https://ap
                 Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
                 Capability.TEXT_OUTPUT,
                 
-                Capability.FUNCTION_CALLING];
+                Capability.FUNCTION_CALLING,
+                Capability.CHAT_COMPLETION_API,
+            ];
         
         // Any other model is able to process text only:
         return [
             Capability.TEXT_INPUT,
             Capability.TEXT_OUTPUT,
-            Capability.FUNCTION_CALLING];
+            Capability.FUNCTION_CALLING,
+            Capability.CHAT_COMPLETION_API,
+        ];
     }
     
     #endregion

app/MindWork AI Studio/Provider/Anthropic/ResponseStreamLine.cs

@@ -14,6 +14,21 @@ public readonly record struct ResponseStreamLine(string Type, int Index, Delta D
 
     /// <inheritdoc />
     public ContentStreamChunk GetContent() => new(this.Delta.Text, []);
+
+    #region Implementation of IAnnotationStreamLine
+
+    //
+    // Please note: Anthropic's API does not currently support sources in their
+    // OpenAI-compatible response stream.
+    //
+
+    /// <inheritdoc />
+    public bool ContainsSources() => false;
+
+    /// <inheritdoc />
+    public IList<ISource> GetSources() => [];
+
+    #endregion
 }
 
 /// <summary>

app/MindWork AI Studio/Provider/BaseProvider.cs

@@ -3,6 +3,7 @@ using System.Runtime.CompilerServices;
 using System.Text.Json;
 
 using AIStudio.Chat;
+using AIStudio.Provider.OpenAI;
 using AIStudio.Settings;
 using AIStudio.Tools.PluginSystem;
 using AIStudio.Tools.Services;
@@ -39,6 +40,7 @@ public abstract class BaseProvider : IProvider, ISecretId
     protected static readonly JsonSerializerOptions JSON_SERIALIZER_OPTIONS = new()
     {
         PropertyNamingPolicy = JsonNamingPolicy.SnakeCaseLower,
+        Converters = { new AnnotationConverter() }
     };
 
     /// <summary>
@@ -123,10 +125,12 @@ public abstract class BaseProvider : IProvider, ISecretId
                 break;
             }
 
+            var errorBody = await nextResponse.Content.ReadAsStringAsync(token);
             if (nextResponse.StatusCode is HttpStatusCode.Forbidden)
             {
                 await MessageBus.INSTANCE.SendError(new(Icons.Material.Filled.Block, string.Format(TB("Tried to communicate with the LLM provider '{0}'. You might not be able to use this provider from your location. The provider message is: '{1}'"), this.InstanceName, nextResponse.ReasonPhrase)));
                 this.logger.LogError($"Failed request with status code {nextResponse.StatusCode} (message = '{nextResponse.ReasonPhrase}').");
+                this.logger.LogDebug($"Error body: {errorBody}");
                 errorMessage = nextResponse.ReasonPhrase;
                 break;
             }
@@ -135,6 +139,7 @@ public abstract class BaseProvider : IProvider, ISecretId
             {
                 await MessageBus.INSTANCE.SendError(new(Icons.Material.Filled.CloudOff, string.Format(TB("Tried to communicate with the LLM provider '{0}'. The required message format might be changed. The provider message is: '{1}'"), this.InstanceName, nextResponse.ReasonPhrase)));
                 this.logger.LogError($"Failed request with status code {nextResponse.StatusCode} (message = '{nextResponse.ReasonPhrase}').");
+                this.logger.LogDebug($"Error body: {errorBody}");
                 errorMessage = nextResponse.ReasonPhrase;
                 break;
             }
@@ -143,6 +148,7 @@ public abstract class BaseProvider : IProvider, ISecretId
             {
                 await MessageBus.INSTANCE.SendError(new(Icons.Material.Filled.CloudOff, string.Format(TB("Tried to communicate with the LLM provider '{0}'. Something was not found. The provider message is: '{1}'"), this.InstanceName, nextResponse.ReasonPhrase)));
                 this.logger.LogError($"Failed request with status code {nextResponse.StatusCode} (message = '{nextResponse.ReasonPhrase}').");
+                this.logger.LogDebug($"Error body: {errorBody}");
                 errorMessage = nextResponse.ReasonPhrase;
                 break;
             }
@@ -151,6 +157,7 @@ public abstract class BaseProvider : IProvider, ISecretId
             {
                 await MessageBus.INSTANCE.SendError(new(Icons.Material.Filled.Key, string.Format(TB("Tried to communicate with the LLM provider '{0}'. The API key might be invalid. The provider message is: '{1}'"), this.InstanceName, nextResponse.ReasonPhrase)));
                 this.logger.LogError($"Failed request with status code {nextResponse.StatusCode} (message = '{nextResponse.ReasonPhrase}').");
+                this.logger.LogDebug($"Error body: {errorBody}");
                 errorMessage = nextResponse.ReasonPhrase;
                 break;
             }
@@ -159,6 +166,7 @@ public abstract class BaseProvider : IProvider, ISecretId
             {
                 await MessageBus.INSTANCE.SendError(new(Icons.Material.Filled.CloudOff, string.Format(TB("Tried to communicate with the LLM provider '{0}'. The server might be down or having issues. The provider message is: '{1}'"), this.InstanceName, nextResponse.ReasonPhrase)));
                 this.logger.LogError($"Failed request with status code {nextResponse.StatusCode} (message = '{nextResponse.ReasonPhrase}').");
+                this.logger.LogDebug($"Error body: {errorBody}");
                 errorMessage = nextResponse.ReasonPhrase;
                 break;
             }
@@ -167,6 +175,7 @@ public abstract class BaseProvider : IProvider, ISecretId
             {
                 await MessageBus.INSTANCE.SendError(new(Icons.Material.Filled.CloudOff, string.Format(TB("Tried to communicate with the LLM provider '{0}'. The provider is overloaded. The message is: '{1}'"), this.InstanceName, nextResponse.ReasonPhrase)));
                 this.logger.LogError($"Failed request with status code {nextResponse.StatusCode} (message = '{nextResponse.ReasonPhrase}').");
+                this.logger.LogDebug($"Error body: {errorBody}");
                 errorMessage = nextResponse.ReasonPhrase;
                 break;
             }
@@ -189,8 +198,20 @@ public abstract class BaseProvider : IProvider, ISecretId
         return new HttpRateLimitedStreamResult(true, false, string.Empty, response);
     }
 
-    protected async IAsyncEnumerable<ContentStreamChunk> StreamChatCompletionInternal<T>(string providerName, Func<Task<HttpRequestMessage>> requestBuilder, [EnumeratorCancellation] CancellationToken token = default) where T : struct, IResponseStreamLine
+    /// <summary>
+    /// Streams the chat completion from the provider using the Chat Completion API.
+    /// </summary>
+    /// <param name="providerName">The name of the provider.</param>
+    /// <param name="requestBuilder">A function that builds the request.</param>
+    /// <param name="token">The cancellation token to use.</param>
+    /// <typeparam name="TDelta">The type of the delta lines inside the stream.</typeparam>
+    /// <typeparam name="TAnnotation">The type of the annotation lines inside the stream.</typeparam>
+    /// <returns>The stream of content chunks.</returns>
+    protected async IAsyncEnumerable<ContentStreamChunk> StreamChatCompletionInternal<TDelta, TAnnotation>(string providerName, Func<Task<HttpRequestMessage>> requestBuilder, [EnumeratorCancellation] CancellationToken token = default) where TDelta : IResponseStreamLine where TAnnotation : IAnnotationStreamLine
     {
+        // Check if annotations are supported:
+        var annotationSupported = typeof(TAnnotation) != typeof(NoResponsesAnnotationStreamLine) && typeof(TAnnotation) != typeof(NoChatCompletionAnnotationStreamLine);
+        
         StreamReader? streamReader = null;
         try
         {
@@ -217,7 +238,9 @@ public abstract class BaseProvider : IProvider, ISecretId
         if (streamReader is null)
             yield break;
         
+        //
         // Read the stream, line by line:
+        //
         while (true)
         {
             try
@@ -240,7 +263,9 @@ public abstract class BaseProvider : IProvider, ISecretId
                 yield break;
             }
 
+            //
             // Read the next line:
+            //
             string? line;
             try
             {
@@ -266,28 +291,233 @@ public abstract class BaseProvider : IProvider, ISecretId
             if (line.StartsWith("data: [DONE]", StringComparison.InvariantCulture))
                 yield break;
 
-            T providerResponse;
+            //
+            // Process annotation lines:
+            //
+            if (annotationSupported && line.Contains("""
+                                                     "annotations":[
+                                                     """, StringComparison.InvariantCulture))
+            {
+                TAnnotation? providerResponse;
+                
+                try
+                {
+                    // We know that the line starts with "data: ". Hence, we can
+                    // skip the first 6 characters to get the JSON data after that.
+                    var jsonData = line[6..];
+
+                    // Deserialize the JSON data:
+                    providerResponse = JsonSerializer.Deserialize<TAnnotation>(jsonData, JSON_SERIALIZER_OPTIONS);
+
+                    if (providerResponse is null)
+                        continue;
+                }
+                catch
+                {
+                    // Skip invalid JSON data:
+                    continue;
+                }
+
+                // Skip empty responses:
+                if (!providerResponse.ContainsSources())
+                    continue;
+
+                // Yield the response:
+                yield return new(string.Empty, providerResponse.GetSources());
+            }
+            
+            //
+            // Process delta lines:
+            //
+            else
+            {
+                TDelta? providerResponse;
+                try
+                {
+                    // We know that the line starts with "data: ". Hence, we can
+                    // skip the first 6 characters to get the JSON data after that.
+                    var jsonData = line[6..];
+
+                    // Deserialize the JSON data:
+                    providerResponse = JsonSerializer.Deserialize<TDelta>(jsonData, JSON_SERIALIZER_OPTIONS);
+
+                    if (providerResponse is null)
+                        continue;
+                }
+                catch
+                {
+                    // Skip invalid JSON data:
+                    continue;
+                }
+
+                // Skip empty responses:
+                if (!providerResponse.ContainsContent())
+                    continue;
+
+                // Yield the response:
+                yield return providerResponse.GetContent();
+            }
+        }
+        
+        streamReader.Dispose();
+    }
+
+    /// <summary>
+    /// Streams the chat completion from the provider using the Responses API.
+    /// </summary>
+    /// <param name="providerName">The name of the provider.</param>
+    /// <param name="requestBuilder">A function that builds the request.</param>
+    /// <param name="token">The cancellation token to use.</param>
+    /// <typeparam name="TDelta">The type of the delta lines inside the stream.</typeparam>
+    /// <typeparam name="TAnnotation">The type of the annotation lines inside the stream.</typeparam>
+    /// <returns>The stream of content chunks.</returns>
+    protected async IAsyncEnumerable<ContentStreamChunk> StreamResponsesInternal<TDelta, TAnnotation>(string providerName, Func<Task<HttpRequestMessage>> requestBuilder, [EnumeratorCancellation] CancellationToken token = default) where TDelta : IResponseStreamLine where TAnnotation : IAnnotationStreamLine
+    {
+        // Check if annotations are supported:
+        var annotationSupported = typeof(TAnnotation) != typeof(NoResponsesAnnotationStreamLine) && typeof(TAnnotation) != typeof(NoChatCompletionAnnotationStreamLine);
+        
+        StreamReader? streamReader = null;
+        try
+        {
+            // Send the request using exponential backoff:
+            var responseData = await this.SendRequest(requestBuilder, token);
+            if(responseData.IsFailedAfterAllRetries)
+            {
+                this.logger.LogError($"The {providerName} responses call failed: {responseData.ErrorMessage}");
+                yield break;
+            }
+            
+            // Open the response stream:
+            var providerStream = await responseData.Response!.Content.ReadAsStreamAsync(token);
+
+            // Add a stream reader to read the stream, line by line:
+            streamReader = new StreamReader(providerStream);
+        }
+        catch(Exception e)
+        {
+            await MessageBus.INSTANCE.SendError(new(Icons.Material.Filled.Stream, string.Format(TB("Tried to communicate with the LLM provider '{0}'. There were some problems with the request. The provider message is: '{1}'"), this.InstanceName, e.Message)));
+            this.logger.LogError($"Failed to stream responses from {providerName} '{this.InstanceName}': {e.Message}");
+        }
+
+        if (streamReader is null)
+            yield break;
+        
+        //
+        // Read the stream, line by line:
+        //
+        while (true)
+        {
             try
             {
-                // We know that the line starts with "data: ". Hence, we can
+                if(streamReader.EndOfStream)
-                // skip the first 6 characters to get the JSON data after that.
+                    break;
-                var jsonData = line[6..];
-
-                // Deserialize the JSON data:
-                providerResponse = JsonSerializer.Deserialize<T>(jsonData, JSON_SERIALIZER_OPTIONS);
             }
-            catch
+            catch (Exception e)
             {
-                // Skip invalid JSON data:
+                await MessageBus.INSTANCE.SendError(new(Icons.Material.Filled.Stream, string.Format(TB("Tried to stream the LLM provider '{0}' answer. There were some problems with the stream. The message is: '{1}'"), this.InstanceName, e.Message)));
-                continue;
+                this.logger.LogWarning($"Failed to read the end-of-stream state from {providerName} '{this.InstanceName}': {e.Message}");
+                break;
             }
                 
-            // Skip empty responses:
+            // Check if the token is canceled:
-            if (!providerResponse.ContainsContent())
+            if (token.IsCancellationRequested)
+            {
+                this.logger.LogWarning($"The user canceled the responses for {providerName} '{this.InstanceName}'.");
+                streamReader.Close();
+                yield break;
+            }
+
+            //
+            // Read the next line:
+            //
+            string? line;
+            try
+            {
+                line = await streamReader.ReadLineAsync(token);
+            }
+            catch (Exception e)
+            {
+                await MessageBus.INSTANCE.SendError(new(Icons.Material.Filled.Stream, string.Format(TB("Tried to stream the LLM provider '{0}' answer. Was not able to read the stream. The message is: '{1}'"), this.InstanceName, e.Message)));
+                this.logger.LogError($"Failed to read the stream from {providerName} '{this.InstanceName}': {e.Message}");
+                break;
+            }
+
+            // Skip empty lines:
+            if (string.IsNullOrWhiteSpace(line))
                 continue;
             
-            // Yield the response:
+            // Check if the line is the end of the stream:
-            yield return providerResponse.GetContent();
+            if (line.StartsWith("event: response.completed", StringComparison.InvariantCulture))
+                yield break;
+            
+            //
+            // Find delta lines:
+            //
+            if (line.StartsWith("""
+                                data: {"type":"response.output_text.delta"
+                                """, StringComparison.InvariantCulture))
+            {
+                TDelta? providerResponse;
+                try
+                {
+                    // We know that the line starts with "data: ". Hence, we can
+                    // skip the first 6 characters to get the JSON data after that.
+                    var jsonData = line[6..];
+
+                    // Deserialize the JSON data:
+                    providerResponse = JsonSerializer.Deserialize<TDelta>(jsonData, JSON_SERIALIZER_OPTIONS);
+
+                    if (providerResponse is null)
+                        continue;
+                }
+                catch
+                {
+                    // Skip invalid JSON data:
+                    continue;
+                }
+
+                // Skip empty responses:
+                if (!providerResponse.ContainsContent())
+                    continue;
+
+                // Yield the response:
+                yield return providerResponse.GetContent();
+            }
+            
+            //
+            // Find annotation added lines:
+            //
+            else if (annotationSupported && line.StartsWith(
+                         """
+                         data: {"type":"response.output_text.annotation.added"
+                         """, StringComparison.InvariantCulture))
+            {
+                TAnnotation? providerResponse;
+                try
+                {
+                    // We know that the line starts with "data: ". Hence, we can
+                    // skip the first 6 characters to get the JSON data after that.
+                    var jsonData = line[6..];
+
+                    // Deserialize the JSON data:
+                    providerResponse = JsonSerializer.Deserialize<TAnnotation>(jsonData, JSON_SERIALIZER_OPTIONS);
+
+                    if (providerResponse is null)
+                        continue;
+                }
+                catch
+                {
+                    // Skip invalid JSON data:
+                    continue;
+                }
+
+                // Skip empty responses:
+                if (!providerResponse.ContainsSources())
+                    continue;
+
+                // Yield the response:
+                yield return new(string.Empty, providerResponse.GetSources());
+            }
         }
         
         streamReader.Dispose();

app/MindWork AI Studio/Provider/CapabilitiesOpenSource.cs

@@ -34,11 +34,17 @@ public static class CapabilitiesOpenSource
                         Capability.TEXT_OUTPUT,
                         
                         Capability.FUNCTION_CALLING,
+                        Capability.CHAT_COMPLETION_API,
                     ];
 
             // The old vision models cannot do function calling:
             if (modelName.IndexOf("vision") is not -1)
-                return [Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT, Capability.TEXT_OUTPUT];
+                return [
+                    Capability.TEXT_INPUT,
+                    Capability.MULTIPLE_IMAGE_INPUT,
+                    Capability.TEXT_OUTPUT,
+                    Capability.CHAT_COMPLETION_API,
+                ];
             
             //
             // All models >= 3.1 are able to do function calling:
@@ -53,10 +59,14 @@ public static class CapabilitiesOpenSource
                         Capability.TEXT_OUTPUT,
                         
                         Capability.FUNCTION_CALLING,
+                        Capability.CHAT_COMPLETION_API,
                     ];
 
             // All other llama models can only do text input and output:
-            return [Capability.TEXT_INPUT, Capability.TEXT_OUTPUT];
+            return [
+                Capability.TEXT_INPUT, Capability.TEXT_OUTPUT,
+                Capability.CHAT_COMPLETION_API,
+            ];
         }
 
         //
@@ -66,9 +76,16 @@ public static class CapabilitiesOpenSource
         {
             if(modelName.IndexOf("deepseek-r1") is not -1 ||
                modelName.IndexOf("deepseek r1") is not -1)
-                return [Capability.TEXT_INPUT, Capability.TEXT_OUTPUT, Capability.ALWAYS_REASONING];
+                return [
+                    Capability.TEXT_INPUT, Capability.TEXT_OUTPUT,
+                    Capability.ALWAYS_REASONING,
+                    Capability.CHAT_COMPLETION_API,
+                ];
             
-            return [Capability.TEXT_INPUT, Capability.TEXT_OUTPUT]; 
+            return [
+                Capability.TEXT_INPUT, Capability.TEXT_OUTPUT,
+                Capability.CHAT_COMPLETION_API,
+            ]; 
         }
         
         //
@@ -77,9 +94,16 @@ public static class CapabilitiesOpenSource
         if (modelName.IndexOf("qwen") is not -1 || modelName.IndexOf("qwq") is not -1)
         {
             if (modelName.IndexOf("qwq") is not -1)
-                return [Capability.TEXT_INPUT, Capability.TEXT_OUTPUT, Capability.ALWAYS_REASONING];
+                return [
+                    Capability.TEXT_INPUT, Capability.TEXT_OUTPUT,
+                    Capability.ALWAYS_REASONING,
+                    Capability.CHAT_COMPLETION_API,
+                ];
             
-            return [Capability.TEXT_INPUT, Capability.TEXT_OUTPUT];
+            return [
+                Capability.TEXT_INPUT, Capability.TEXT_OUTPUT,
+                Capability.CHAT_COMPLETION_API,
+            ];
         }
         
         //
@@ -93,7 +117,8 @@ public static class CapabilitiesOpenSource
                     [
                         Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
                         Capability.TEXT_OUTPUT,
-                        Capability.FUNCTION_CALLING
+                        Capability.FUNCTION_CALLING,
+                        Capability.CHAT_COMPLETION_API,
                     ];
             
             if (modelName.IndexOf("3.1") is not -1)
@@ -101,7 +126,8 @@ public static class CapabilitiesOpenSource
                 [
                     Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
                     Capability.TEXT_OUTPUT,
-                    Capability.FUNCTION_CALLING
+                    Capability.FUNCTION_CALLING,
+                    Capability.CHAT_COMPLETION_API,
                 ];
             
             // Default:
@@ -109,7 +135,8 @@ public static class CapabilitiesOpenSource
                 [
                     Capability.TEXT_INPUT,
                     Capability.TEXT_OUTPUT,
-                    Capability.FUNCTION_CALLING
+                    Capability.FUNCTION_CALLING,
+                    Capability.CHAT_COMPLETION_API,
                 ];
         }
         
@@ -123,6 +150,7 @@ public static class CapabilitiesOpenSource
                 [
                     Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
                     Capability.TEXT_OUTPUT,
+                    Capability.CHAT_COMPLETION_API,
                 ];
         
             if(modelName.StartsWith("grok-3-mini"))
@@ -132,6 +160,7 @@ public static class CapabilitiesOpenSource
                     Capability.TEXT_OUTPUT,
                     
                     Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING,
+                    Capability.CHAT_COMPLETION_API,
                 ];
         
             if(modelName.StartsWith("grok-3"))
@@ -141,10 +170,41 @@ public static class CapabilitiesOpenSource
                     Capability.TEXT_OUTPUT,
                     
                     Capability.FUNCTION_CALLING,
+                    Capability.CHAT_COMPLETION_API,
+                ];
+        }
+        
+        //
+        // OpenAI models:
+        //
+        if (modelName.IndexOf("gpt-oss") is not -1 ||
+            modelName.IndexOf("gpt-3.5") is not -1)
+        {
+            if(modelName.IndexOf("gpt-oss") is not -1)
+                return 
+                [
+                    Capability.TEXT_INPUT,
+                    Capability.TEXT_OUTPUT,
+                    
+                    Capability.FUNCTION_CALLING,
+                    Capability.WEB_SEARCH,
+                    Capability.CHAT_COMPLETION_API,
+                ];
+            
+            if(modelName.IndexOf("gpt-3.5") is not -1)
+                return 
+                [
+                    Capability.TEXT_INPUT,
+                    Capability.TEXT_OUTPUT,
+                    
+                    Capability.CHAT_COMPLETION_API,
                 ];
         }
         
         // Default:
-        return [Capability.TEXT_INPUT, Capability.TEXT_OUTPUT];
+        return [
+            Capability.TEXT_INPUT, Capability.TEXT_OUTPUT,
+            Capability.CHAT_COMPLETION_API,
+        ];
     }
 }

app/MindWork AI Studio/Provider/Capability.cs

@@ -94,4 +94,19 @@ public enum Capability
     /// The AI model can perform function calling, such as invoking APIs or executing functions.
     /// </summary>
     FUNCTION_CALLING,
+    
+    /// <summary>
+    /// The AI model can perform web search to retrieve information from the internet.
+    /// </summary>
+    WEB_SEARCH,
+    
+    /// <summary>
+    /// The AI model is used via the Chat Completion API.
+    /// </summary>
+    CHAT_COMPLETION_API,
+    
+    /// <summary>
+    /// The AI model is used via the Responses API.
+    /// </summary>
+    RESPONSES_API,
 }

app/MindWork AI Studio/Provider/DeepSeek/ProviderDeepSeek.cs

@@ -35,7 +35,7 @@ public sealed class ProviderDeepSeek(ILogger logger) : BaseProvider("https://api
         };
         
         // Prepare the DeepSeek HTTP chat request:
-        var deepSeekChatRequest = JsonSerializer.Serialize(new ChatRequest
+        var deepSeekChatRequest = JsonSerializer.Serialize(new ChatCompletionAPIRequest
         {
             Model = chatModel.Id,
             
@@ -76,7 +76,7 @@ public sealed class ProviderDeepSeek(ILogger logger) : BaseProvider("https://api
             return request;
         }
         
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("DeepSeek", RequestBuilder, token))
+        await foreach (var content in this.StreamChatCompletionInternal<ChatCompletionDeltaStreamLine, NoChatCompletionAnnotationStreamLine>("DeepSeek", RequestBuilder, token))
             yield return content;
     }
 
@@ -117,12 +117,14 @@ public sealed class ProviderDeepSeek(ILogger logger) : BaseProvider("https://api
                 Capability.TEXT_OUTPUT,
                 
                 Capability.ALWAYS_REASONING,
+                Capability.CHAT_COMPLETION_API,
             ];
         
         return
         [
             Capability.TEXT_INPUT,
             Capability.TEXT_OUTPUT,
+            Capability.CHAT_COMPLETION_API,
         ];
     }
 

app/MindWork AI Studio/Provider/Fireworks/ProviderFireworks.cs

@@ -4,6 +4,7 @@ using System.Text;
 using System.Text.Json;
 
 using AIStudio.Chat;
+using AIStudio.Provider.OpenAI;
 using AIStudio.Settings;
 
 namespace AIStudio.Provider.Fireworks;
@@ -77,7 +78,7 @@ public class ProviderFireworks(ILogger logger) : BaseProvider("https://api.firew
             return request;
         }
         
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("Fireworks", RequestBuilder, token))
+        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine, ChatCompletionAnnotationStreamLine>("Fireworks", RequestBuilder, token))
             yield return content;
     }
 

app/MindWork AI Studio/Provider/Fireworks/ResponseStreamLine.cs

@@ -15,6 +15,20 @@ public readonly record struct ResponseStreamLine(string Id, string Object, uint
 
     /// <inheritdoc />
     public ContentStreamChunk GetContent() => new(this.Choices[0].Delta.Content, []);
+
+    #region Implementation of IAnnotationStreamLine
+
+    //
+    // Currently, Fireworks does not provide source citations in their response stream.
+    //
+    
+    /// <inheritdoc />
+    public bool ContainsSources() => false;
+
+    /// <inheritdoc />
+    public IList<ISource> GetSources() => [];
+
+    #endregion
 }
 
 /// <summary>

app/MindWork AI Studio/Provider/GWDG/ProviderGWDG.cs

@@ -35,7 +35,7 @@ public sealed class ProviderGWDG(ILogger logger) : BaseProvider("https://chat-ai
         };
         
         // Prepare the GWDG HTTP chat request:
-        var gwdgChatRequest = JsonSerializer.Serialize(new ChatRequest
+        var gwdgChatRequest = JsonSerializer.Serialize(new ChatCompletionAPIRequest
         {
             Model = chatModel.Id,
             
@@ -76,7 +76,7 @@ public sealed class ProviderGWDG(ILogger logger) : BaseProvider("https://chat-ai
             return request;
         }
         
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("GWDG", RequestBuilder, token))
+        await foreach (var content in this.StreamChatCompletionInternal<ChatCompletionDeltaStreamLine, ChatCompletionAnnotationStreamLine>("GWDG", RequestBuilder, token))
             yield return content;
     }
 

app/MindWork AI Studio/Provider/Google/ProviderGoogle.cs

@@ -78,7 +78,7 @@ public class ProviderGoogle(ILogger logger) : BaseProvider("https://generativela
             return request;
         }
         
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("Google", RequestBuilder, token))
+        await foreach (var content in this.StreamChatCompletionInternal<ChatCompletionDeltaStreamLine, NoChatCompletionAnnotationStreamLine>("Google", RequestBuilder, token))
             yield return content;
     }
 
@@ -136,6 +136,7 @@ public class ProviderGoogle(ILogger logger) : BaseProvider("https://generativela
                     Capability.TEXT_OUTPUT,
                     
                     Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING,
+                    Capability.CHAT_COMPLETION_API,
                 ];
 
             // Image generation:
@@ -146,6 +147,7 @@ public class ProviderGoogle(ILogger logger) : BaseProvider("https://generativela
                     Capability.SPEECH_INPUT, Capability.VIDEO_INPUT,
                     
                     Capability.TEXT_OUTPUT, Capability.IMAGE_OUTPUT,
+                    Capability.CHAT_COMPLETION_API,
                 ];
             
             // Realtime model:
@@ -158,6 +160,7 @@ public class ProviderGoogle(ILogger logger) : BaseProvider("https://generativela
                     Capability.TEXT_OUTPUT, Capability.SPEECH_OUTPUT,
                     
                     Capability.FUNCTION_CALLING,
+                    Capability.CHAT_COMPLETION_API,
                 ];
             
             // The 2.0 flash models cannot call functions:
@@ -168,6 +171,7 @@ public class ProviderGoogle(ILogger logger) : BaseProvider("https://generativela
                     Capability.SPEECH_INPUT, Capability.VIDEO_INPUT,
                     
                     Capability.TEXT_OUTPUT,
+                    Capability.CHAT_COMPLETION_API,
                 ];
             
             // The old 1.0 pro vision model:
@@ -177,6 +181,7 @@ public class ProviderGoogle(ILogger logger) : BaseProvider("https://generativela
                     Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
                     
                     Capability.TEXT_OUTPUT,
+                    Capability.CHAT_COMPLETION_API,
                 ];
             
             // Default to all other Gemini models:
@@ -188,6 +193,7 @@ public class ProviderGoogle(ILogger logger) : BaseProvider("https://generativela
                 Capability.TEXT_OUTPUT,
                 
                 Capability.FUNCTION_CALLING,
+                Capability.CHAT_COMPLETION_API,
             ];
         }
         
@@ -199,6 +205,7 @@ public class ProviderGoogle(ILogger logger) : BaseProvider("https://generativela
             Capability.TEXT_OUTPUT,
             
             Capability.FUNCTION_CALLING,
+            Capability.CHAT_COMPLETION_API,
         ];
     }
     

app/MindWork AI Studio/Provider/Groq/ProviderGroq.cs

@@ -61,8 +61,6 @@ public class ProviderGroq(ILogger logger) : BaseProvider("https://api.groq.com/o
                 }
             }).ToList()],
             
-            Seed = chatThread.Seed,
-            
             // Right now, we only support streaming completions:
             Stream = true,
         }, JSON_SERIALIZER_OPTIONS);
@@ -80,7 +78,7 @@ public class ProviderGroq(ILogger logger) : BaseProvider("https://api.groq.com/o
             return request;
         }
         
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("Groq", RequestBuilder, token))
+        await foreach (var content in this.StreamChatCompletionInternal<ChatCompletionDeltaStreamLine, ChatCompletionAnnotationStreamLine>("Groq", RequestBuilder, token))
             yield return content;
     }
 

app/MindWork AI Studio/Provider/Helmholtz/ProviderHelmholtz.cs

@@ -35,7 +35,7 @@ public sealed class ProviderHelmholtz(ILogger logger) : BaseProvider("https://ap
         };
         
         // Prepare the Helmholtz HTTP chat request:
-        var helmholtzChatRequest = JsonSerializer.Serialize(new ChatRequest
+        var helmholtzChatRequest = JsonSerializer.Serialize(new ChatCompletionAPIRequest
         {
             Model = chatModel.Id,
             
@@ -76,7 +76,7 @@ public sealed class ProviderHelmholtz(ILogger logger) : BaseProvider("https://ap
             return request;
         }
         
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("Helmholtz", RequestBuilder, token))
+        await foreach (var content in this.StreamChatCompletionInternal<ChatCompletionDeltaStreamLine, ChatCompletionAnnotationStreamLine>("Helmholtz", RequestBuilder, token))
             yield return content;
     }
 

app/MindWork AI Studio/Provider/HuggingFace/ProviderHuggingFace.cs

@@ -40,7 +40,7 @@ public sealed class ProviderHuggingFace : BaseProvider
         };
         
         // Prepare the HuggingFace HTTP chat request:
-        var huggingfaceChatRequest = JsonSerializer.Serialize(new ChatRequest
+        var huggingfaceChatRequest = JsonSerializer.Serialize(new ChatCompletionAPIRequest
         {
             Model = chatModel.Id,
             
@@ -81,7 +81,7 @@ public sealed class ProviderHuggingFace : BaseProvider
             return request;
         }
         
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("HuggingFace", RequestBuilder, token))
+        await foreach (var content in this.StreamChatCompletionInternal<ChatCompletionDeltaStreamLine, ChatCompletionAnnotationStreamLine>("HuggingFace", RequestBuilder, token))
             yield return content;
     }
 

app/MindWork AI Studio/Provider/IAnnotationStreamLine.cs

@@ -0,0 +1,19 @@
+namespace AIStudio.Provider;
+
+/// <summary>
+/// A contract for a line in a response stream that can provide annotations such as sources.
+/// </summary>
+public interface IAnnotationStreamLine
+{
+    /// <summary>
+    /// Checks if the response line contains any sources.
+    /// </summary>
+    /// <returns>True when the response line contains sources, false otherwise.</returns>
+    public bool ContainsSources();
+
+    /// <summary>
+    /// Gets the sources of the response line.
+    /// </summary>
+    /// <returns>The sources of the response line.</returns>
+    public IList<ISource> GetSources();
+}

app/MindWork AI Studio/Provider/IResponseStreamLine.cs

@@ -1,6 +1,9 @@
 namespace AIStudio.Provider;
 
-public interface IResponseStreamLine
+/// <summary>
+/// A contract for a streamed response line that may contain content and annotations.
+/// </summary>
+public interface IResponseStreamLine : IAnnotationStreamLine
 {
     /// <summary>
     /// Checks if the response line contains any content.
@@ -13,16 +16,4 @@ public interface IResponseStreamLine
     /// </summary>
     /// <returns>The content of the response line.</returns>
     public ContentStreamChunk GetContent();
-    
-    /// <summary>
-    /// Checks if the response line contains any sources.
-    /// </summary>
-    /// <returns>True when the response line contains sources, false otherwise.</returns>
-    public bool ContainsSources() => false;
-    
-    /// <summary>
-    /// Gets the sources of the response line.
-    /// </summary>
-    /// <returns>The sources of the response line.</returns>
-    public IList<ISource> GetSources() => [];
 }

app/MindWork AI Studio/Provider/Mistral/ProviderMistral.cs

@@ -59,8 +59,6 @@ public sealed class ProviderMistral(ILogger logger) : BaseProvider("https://api.
                 }
             }).ToList()],
             
-            RandomSeed = chatThread.Seed,
-            
             // Right now, we only support streaming completions:
             Stream = true,
             SafePrompt = false,
@@ -79,7 +77,7 @@ public sealed class ProviderMistral(ILogger logger) : BaseProvider("https://api.
             return request;
         }
         
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("Mistral", RequestBuilder, token))
+        await foreach (var content in this.StreamChatCompletionInternal<ChatCompletionDeltaStreamLine, NoChatCompletionAnnotationStreamLine>("Mistral", RequestBuilder, token))
             yield return content;
     }
 
@@ -134,6 +132,7 @@ public sealed class ProviderMistral(ILogger logger) : BaseProvider("https://api.
                 Capability.TEXT_OUTPUT,
                 
                 Capability.FUNCTION_CALLING,
+                Capability.CHAT_COMPLETION_API,
             ];
         
         // Mistral medium:
@@ -144,6 +143,7 @@ public sealed class ProviderMistral(ILogger logger) : BaseProvider("https://api.
                 Capability.TEXT_OUTPUT,
                 
                 Capability.FUNCTION_CALLING,
+                Capability.CHAT_COMPLETION_API,
             ];
         
         // Mistral small:
@@ -154,6 +154,7 @@ public sealed class ProviderMistral(ILogger logger) : BaseProvider("https://api.
                 Capability.TEXT_OUTPUT,
                 
                 Capability.FUNCTION_CALLING,
+                Capability.CHAT_COMPLETION_API,
             ];
         
         // Mistral saba:
@@ -162,6 +163,7 @@ public sealed class ProviderMistral(ILogger logger) : BaseProvider("https://api.
             [
                 Capability.TEXT_INPUT,
                 Capability.TEXT_OUTPUT,
+                Capability.CHAT_COMPLETION_API,
             ];
         
         // Default:

app/MindWork AI Studio/Provider/NoChatCompletionAnnotationStreamLine.cs

@@ -0,0 +1,15 @@
+namespace AIStudio.Provider;
+
+/// <summary>
+/// A marker record indicating that no chat completion annotation line is expected in that stream.
+/// </summary>
+public sealed record NoChatCompletionAnnotationStreamLine : IAnnotationStreamLine
+{
+    #region Implementation of IAnnotationStreamLine
+
+    public bool ContainsSources() => false;
+
+    public IList<ISource> GetSources() => [];
+
+    #endregion
+}

app/MindWork AI Studio/Provider/NoResponsesAnnotationStreamLine.cs

@@ -0,0 +1,15 @@
+namespace AIStudio.Provider;
+
+/// <summary>
+/// A marker record indicating that no annotation line is expected in that Responses API stream.
+/// </summary>
+public sealed record NoResponsesAnnotationStreamLine : IAnnotationStreamLine
+{
+    #region Implementation of IAnnotationStreamLine
+
+    public bool ContainsSources() => false;
+
+    public IList<ISource> GetSources() => [];
+
+    #endregion
+}

app/MindWork AI Studio/Provider/OpenAI/AnnotatingUnknown.cs

@@ -0,0 +1,7 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Represents an unknown annotation type.
+/// </summary>
+/// <param name="Type">The type of the unknown annotation.</param>
+public sealed record AnnotatingUnknown(string Type) : Annotation(Type);

app/MindWork AI Studio/Provider/OpenAI/Annotation.cs

@@ -0,0 +1,10 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Base class for different types of annotations.
+/// </summary>
+/// <remarks>
+/// We use this base class to represent various annotation types for all types of LLM providers.
+/// </remarks>
+/// <param name="Type">The type of the annotation.</param>
+public abstract record Annotation(string Type);

app/MindWork AI Studio/Provider/OpenAI/AnnotationConverter.cs

@@ -0,0 +1,62 @@
+using System.Text.Json;
+using System.Text.Json.Serialization;
+
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Custom JSON converter for the annotation class to handle polymorphic deserialization.
+/// </summary>
+/// <remarks>
+/// We use this converter for chat completion API and responses API annotation deserialization.
+/// </remarks>
+public sealed class AnnotationConverter : JsonConverter<Annotation>
+{
+    public override Annotation? Read(ref Utf8JsonReader reader, Type typeToConvert, JsonSerializerOptions options)
+    {
+        using var doc = JsonDocument.ParseValue(ref reader);
+        var root = doc.RootElement;
+        
+        if (!root.TryGetProperty("type", out var typeElement))
+            return null;
+        
+        var type = typeElement.GetString();
+        var rawText = root.GetRawText();
+        
+        Annotation? annotation;
+        switch (type)
+        {
+            case "url_citation":
+                
+                // Let's check the responses API data type first:
+                var responsesAnnotation = JsonSerializer.Deserialize<ResponsesAnnotatingUrlCitationData>(rawText, options);
+                
+                // If it fails, let's try the chat completion API data type:
+                if(responsesAnnotation is null || string.IsNullOrWhiteSpace(responsesAnnotation.Title) || string.IsNullOrWhiteSpace(responsesAnnotation.URL))
+                {
+                    // Try chat completion API data type:
+                    var chatCompletionAnnotation = JsonSerializer.Deserialize<ChatCompletionAnnotatingURL>(rawText, options);
+                    
+                    // If both fail, we return the unknown type:
+                    if(chatCompletionAnnotation is null)
+                        annotation = new AnnotatingUnknown(type);
+                    else
+                        annotation = chatCompletionAnnotation;
+                }
+                else
+                    annotation = responsesAnnotation;
+                
+                break;
+            
+            default:
+                annotation = new AnnotatingUnknown(type ?? "unknown");
+                break;
+        }
+
+        return annotation;
+    }
+
+    public override void Write(Utf8JsonWriter writer, Annotation value, JsonSerializerOptions options)
+    {
+        JsonSerializer.Serialize(writer, value, value.GetType(), options);
+    }
+}

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionAPIRequest.cs

@@ -0,0 +1,18 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// The OpenAI's legacy chat completion request model.
+/// </summary>
+/// <param name="Model">Which model to use for chat completion.</param>
+/// <param name="Messages">The chat messages.</param>
+/// <param name="Stream">Whether to stream the chat completion.</param>
+public record ChatCompletionAPIRequest(
+    string Model,
+    IList<Message> Messages,
+    bool Stream
+)
+{
+    public ChatCompletionAPIRequest() : this(string.Empty, [], true)
+    {
+    }
+}

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionAnnotatingURL.cs

@@ -0,0 +1,16 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Data structure for URL annotation in chat completions.
+/// </summary>
+/// <remarks>
+/// Although this class is not directly intended for the Responses API, it is
+/// used there as a fallback solution. One day, one of the open source LLM
+/// drivers may use this data structure for their responses API.
+/// </remarks>
+/// <param name="Type">The type of annotation, typically "url_citation".</param>
+/// <param name="UrlCitation">The URL citation details.</param>
+public sealed record ChatCompletionAnnotatingURL(
+    string Type,
+    ChatCompletionUrlCitationData UrlCitation
+) : Annotation(Type);

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionAnnotationChoice.cs

@@ -0,0 +1,8 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Data structure representing a choice in a chat completion annotation response.
+/// </summary>
+/// <param name="Index">The index of the choice.</param>
+/// <param name="Delta">The delta information for the choice.</param>
+public record ChatCompletionAnnotationChoice(int Index, ChatCompletionAnnotationDelta Delta);

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionAnnotationDelta.cs

@@ -0,0 +1,7 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Data structure representing annotation deltas in chat completions.
+/// </summary>
+/// <param name="Annotations">The list of annotations, which can be null.</param>
+public record ChatCompletionAnnotationDelta(IList<Annotation>? Annotations);

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionAnnotationStreamLine.cs

@@ -0,0 +1,57 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Represents a line of a chat completion annotation stream.
+/// </summary>
+/// <param name="Id">The unique identifier of the chat completion.</param>
+/// <param name="Object">The type of object returned, typically "chat.completion".</param>
+/// <param name="Created">The creation timestamp of the chat completion in Unix epoch format.</param>
+/// <param name="Model">The model used for the chat completion.</param>
+/// <param name="SystemFingerprint">The system fingerprint associated with the chat completion.</param>
+/// <param name="Choices">The list of choices returned in the chat completion.</param>
+public record ChatCompletionAnnotationStreamLine(string Id, string Object, uint Created, string Model, string SystemFingerprint, IList<ChatCompletionAnnotationChoice> Choices) : IAnnotationStreamLine
+{
+    #region Implementation of IAnnotationStreamLine
+
+    /// <inheritdoc />
+    public bool ContainsSources() => this.Choices.Any(choice => choice.Delta.Annotations is not null && choice.Delta.Annotations.Any(annotation => annotation is not AnnotatingUnknown));
+
+    /// <inheritdoc />
+    public IList<ISource> GetSources()
+    {
+        var sources = new List<ISource>();
+        foreach (var choice in this.Choices)
+        {
+            if (choice.Delta.Annotations is null)
+                continue;
+
+            // Iterate through all annotations:
+            foreach (var annotation in choice.Delta.Annotations)
+            {
+                // Check if the annotation is of the expected type and extract the source information:
+                if (annotation is ChatCompletionAnnotatingURL urlAnnotation)
+                    sources.Add(new Source(urlAnnotation.UrlCitation.Title, urlAnnotation.UrlCitation.URL));
+
+                //
+                // Check for the unexpected annotation type of the Responses API.
+                //
+                // This seems weird at first. But there are two possibilities why this could happen:
+                // - Anyone of the open source providers such as ollama, LM Studio, etc. could
+                //   implement & use the Responses API data structures for annotations in their
+                //   chat completion endpoint.
+                //
+                // - Our custom JSON converter checks for the Responses API data type first. If it
+                //   fails, it checks for the chat completion API data type. So, when the Responses
+                //   API data type is valid, it will be deserialized into that type, even though
+                //   we are calling the chat completion endpoint.
+                //
+                if (annotation is ResponsesAnnotatingUrlCitationData citationData)
+                    sources.Add(new Source(citationData.Title, citationData.URL));
+            }
+        }
+
+        return sources;
+    }
+
+    #endregion
+}

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionChoice.cs

@@ -0,0 +1,13 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Data model for a choice made by the AI.
+/// </summary>
+/// <param name="Index">The index of the choice.</param>
+/// <param name="Delta">The delta text of the choice.</param>
+public record ChatCompletionChoice(int Index, ChatCompletionDelta Delta)
+{
+    public ChatCompletionChoice() : this(0, new (string.Empty))
+    {
+    }
+}

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionDelta.cs

@@ -0,0 +1,12 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// The delta text of a choice.
+/// </summary>
+/// <param name="Content">The content of the delta text.</param>
+public record ChatCompletionDelta(string Content)
+{
+    public ChatCompletionDelta() : this(string.Empty)
+    {
+    }
+}

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionDeltaStreamLine.cs

@@ -0,0 +1,45 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Data model for a delta line in the chat completion response stream.
+/// </summary>
+/// <param name="Id">The id of the response.</param>
+/// <param name="Object">The object describing the response.</param>
+/// <param name="Created">The timestamp of the response.</param>
+/// <param name="Model">The model used for the response.</param>
+/// <param name="SystemFingerprint">The system fingerprint; together with the seed, this allows you to reproduce the response.</param>
+/// <param name="Choices">The choices made by the AI.</param>
+public record ChatCompletionDeltaStreamLine(string Id, string Object, uint Created, string Model, string SystemFingerprint, IList<ChatCompletionChoice> Choices) : IResponseStreamLine
+{
+    public ChatCompletionDeltaStreamLine() : this(string.Empty, string.Empty, 0, string.Empty, string.Empty, [])
+    {
+    }
+    
+    /// <inheritdoc />
+    public bool ContainsContent() => this.Choices.Count > 0;
+
+    /// <inheritdoc />
+    public ContentStreamChunk GetContent() => new(this.Choices[0].Delta.Content, []);
+
+    #region Implementation of IAnnotationStreamLine
+
+    //
+    // Please note that there are multiple options where LLM providers might stream sources:
+    //
+    // - As part of the delta content while streaming. That would be part of this class.
+    // - By using a dedicated stream event and data structure. That would be another class implementing IResponseStreamLine.
+    //
+    // Right now, OpenAI uses the latter approach, so we don't have any sources here. And
+    // because no other provider does it yet, we don't have any implementation here either.
+    //
+    // One example where sources are part of the delta content is the Perplexity provider.
+    //
+    
+    /// <inheritdoc />
+    public bool ContainsSources() => false;
+
+    /// <inheritdoc />
+    public IList<ISource> GetSources() => [];
+
+    #endregion
+}

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionUrlCitationData.cs

@@ -0,0 +1,14 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Represents citation data for a URL in a chat completion response.
+/// </summary>
+/// <param name="EndIndex">The end index of the citation in the response text.</param>
+/// <param name="StartIndex">The start index of the citation in the response text.</param>
+/// <param name="Title">The title of the cited source.</param>
+/// <param name="URL">The URL of the cited source.</param>
+public sealed record ChatCompletionUrlCitationData(
+    int EndIndex,
+    int StartIndex,
+    string Title,
+    string URL);

app/MindWork AI Studio/Provider/OpenAI/ChatRequest.cs

@@ -1,21 +0,0 @@
-using System.ComponentModel.DataAnnotations;
-
-namespace AIStudio.Provider.OpenAI;
-
-/// <summary>
-/// The OpenAI chat request model.
-/// </summary>
-/// <param name="Model">Which model to use for chat completion.</param>
-/// <param name="Messages">The chat messages.</param>
-/// <param name="Stream">Whether to stream the chat completion.</param>
-/// <param name="Seed">The seed for the chat completion.</param>
-/// <param name="FrequencyPenalty">The frequency penalty for the chat completion.</param>
-public readonly record struct ChatRequest(
-    string Model,
-    IList<Message> Messages,
-    bool Stream,
-    int Seed,
-    
-    [Range(-2.0f, 2.0f)]
-    float FrequencyPenalty
-);

app/MindWork AI Studio/Provider/OpenAI/ProviderOpenAI.cs

@@ -30,16 +30,20 @@ public sealed class ProviderOpenAI(ILogger logger) : BaseProvider("https://api.o
             yield break;
         
         // Unfortunately, OpenAI changed the name of the system prompt based on the model.
-        // All models that start with "o" (the omni aka reasoning models) and all GPT4o models
+        // All models that start with "o" (the omni aka reasoning models), all GPT4o models,
-        // have the system prompt named "developer". All other models have the system prompt
+        // and all newer models have the system prompt named "developer". All other models
-        // named "system". We need to check this to get the correct system prompt.
+        // have the system prompt named "system". We need to check this to get the correct
+        // system prompt.
         //
         // To complicate it even more: The early versions of reasoning models, which are released
         // before the 17th of December 2024, have no system prompt at all. We need to check this
         // as well.
         
         // Apply the basic rule first:
-        var systemPromptRole = chatModel.Id.StartsWith('o') || chatModel.Id.Contains("4o") ? "developer" : "system";
+        var systemPromptRole =
+            chatModel.Id.StartsWith('o') ||
+            chatModel.Id.StartsWith("gpt-5", StringComparison.Ordinal) ||
+            chatModel.Id.Contains("4o") ? "developer" : "system";
         
         // Check if the model is an early version of the reasoning models:
         systemPromptRole = chatModel.Id switch
@@ -52,7 +56,16 @@ public sealed class ProviderOpenAI(ILogger logger) : BaseProvider("https://api.o
             _ => systemPromptRole,
         };
 
-        this.logger.LogInformation($"Using the system prompt role '{systemPromptRole}' for model '{chatModel.Id}'.");
+        // Read the model capabilities:
+        var modelCapabilities = this.GetModelCapabilities(chatModel);
+        
+        // Check if we are using the Responses API or the Chat Completion API:
+        var usingResponsesAPI = modelCapabilities.Contains(Capability.RESPONSES_API);
+        
+        // Prepare the request path based on the API we are using:
+        var requestPath = usingResponsesAPI ? "responses" : "chat/completions";
+        
+        this.logger.LogInformation("Using the system prompt role '{SystemPromptRole}' and the '{RequestPath}' API for model '{ChatModelId}'.", systemPromptRole, requestPath, chatModel.Id);
         
         // Prepare the system prompt:
         var systemPrompt = new Message
@@ -61,43 +74,94 @@ public sealed class ProviderOpenAI(ILogger logger) : BaseProvider("https://api.o
             Content = chatThread.PrepareSystemPrompt(settingsManager, chatThread, this.logger),
         };
 
-        // Prepare the OpenAI HTTP chat request:
+        //
-        var openAIChatRequest = JsonSerializer.Serialize(new ChatRequest
+        // Prepare the tools we want to use:
+        //
+        IList<Tool> tools = modelCapabilities.Contains(Capability.WEB_SEARCH) switch
         {
-            Model = chatModel.Id,
+            true => [ Tools.WEB_SEARCH ],
+            _ => []
+        };
         
-            // Build the messages:
+        //
-            // - First of all the system prompt
+        // Create the request: either for the Responses API or the Chat Completion API
-            // - Then none-empty user and AI messages
+        //
-            Messages = [systemPrompt, ..chatThread.Blocks.Where(n => n.ContentType is ContentType.TEXT && !string.IsNullOrWhiteSpace((n.Content as ContentText)?.Text)).Select(n => new Message
+        var openAIChatRequest = usingResponsesAPI switch
+        {
+            // Chat Completion API request:
+            false => JsonSerializer.Serialize(new ChatCompletionAPIRequest
             {
-                Role = n.Role switch
+                Model = chatModel.Id,
+            
+                // Build the messages:
+                // - First of all the system prompt
+                // - Then none-empty user and AI messages
+                Messages = [systemPrompt, ..chatThread.Blocks.Where(n => n.ContentType is ContentType.TEXT && !string.IsNullOrWhiteSpace((n.Content as ContentText)?.Text)).Select(n => new Message
                 {
-                    ChatRole.USER => "user",
+                    Role = n.Role switch
-                    ChatRole.AI => "assistant",
+                    {
-                    ChatRole.AGENT => "assistant",
+                        ChatRole.USER => "user",
-                    ChatRole.SYSTEM => systemPromptRole,
+                        ChatRole.AI => "assistant",
+                        ChatRole.AGENT => "assistant",
+                        ChatRole.SYSTEM => systemPromptRole,
 
-                    _ => "user",
+                        _ => "user",
-                },
+                    },
 
-                Content = n.Content switch
+                    Content = n.Content switch
+                    {
+                        ContentText text => text.Text,
+                        _ => string.Empty,
+                    }
+                }).ToList()],
+            
+                // Right now, we only support streaming completions:
+                Stream = true,
+            }, JSON_SERIALIZER_OPTIONS),
+            
+            // Responses API request:
+            true => JsonSerializer.Serialize(new ResponsesAPIRequest
+            {
+                Model = chatModel.Id,
+            
+                // Build the messages:
+                // - First of all the system prompt
+                // - Then none-empty user and AI messages
+                Input = [systemPrompt, ..chatThread.Blocks.Where(n => n.ContentType is ContentType.TEXT && !string.IsNullOrWhiteSpace((n.Content as ContentText)?.Text)).Select(n => new Message
                 {
-                    ContentText text => text.Text,
+                    Role = n.Role switch
-                    _ => string.Empty,
+                    {
-                }
+                        ChatRole.USER => "user",
-            }).ToList()],
+                        ChatRole.AI => "assistant",
+                        ChatRole.AGENT => "assistant",
+                        ChatRole.SYSTEM => systemPromptRole,
 
-            Seed = chatThread.Seed,
+                        _ => "user",
+                    },
 
-            // Right now, we only support streaming completions:
+                    Content = n.Content switch
-            Stream = true,
+                    {
-        }, JSON_SERIALIZER_OPTIONS);
+                        ContentText text => text.Text,
+                        _ => string.Empty,
+                    }
+                }).ToList()],
+            
+                // Right now, we only support streaming completions:
+                Stream = true,
+                
+                // We do not want to store any data on OpenAI's servers:
+                Store = false,
+                
+                // Tools we want to use:
+                Tools = tools,
+                
+            }, JSON_SERIALIZER_OPTIONS),
+        };
 
         async Task<HttpRequestMessage> RequestBuilder()
         {
             // Build the HTTP post request:
-            var request = new HttpRequestMessage(HttpMethod.Post, "chat/completions");
+            var request = new HttpRequestMessage(HttpMethod.Post, requestPath);
 
             // Set the authorization header:
             request.Headers.Authorization = new AuthenticationHeaderValue("Bearer", await requestedSecret.Secret.Decrypt(ENCRYPTION));
@@ -107,28 +171,34 @@ public sealed class ProviderOpenAI(ILogger logger) : BaseProvider("https://api.o
             return request;
         }
 
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("OpenAI", RequestBuilder, token))
+        if (usingResponsesAPI)
-            yield return content;
+            await foreach (var content in this.StreamResponsesInternal<ResponsesDeltaStreamLine, ResponsesAnnotationStreamLine>("OpenAI", RequestBuilder, token))
+                yield return content;
+        
+        else
+            await foreach (var content in this.StreamChatCompletionInternal<ChatCompletionDeltaStreamLine, ChatCompletionAnnotationStreamLine>("OpenAI", RequestBuilder, token))
+                yield return content;
     }
 
     #pragma warning disable CS1998 // Async method lacks 'await' operators and will run synchronously
+    
     /// <inheritdoc />
     public override async IAsyncEnumerable<ImageURL> StreamImageCompletion(Model imageModel, string promptPositive, string promptNegative = FilterOperator.String.Empty, ImageURL referenceImageURL = default, [EnumeratorCancellation] CancellationToken token = default)
     {
         yield break;
     }
+    
     #pragma warning restore CS1998 // Async method lacks 'await' operators and will run synchronously
 
     /// <inheritdoc />
     public override async Task<IEnumerable<Model>> GetTextModels(string? apiKeyProvisional = null, CancellationToken token = default)
     {
-        var models = await this.LoadModels(["gpt-", "o1-", "o3-", "o4-"], token, apiKeyProvisional);
+        var models = await this.LoadModels(["chatgpt-", "gpt-", "o1-", "o3-", "o4-"], token, apiKeyProvisional);
         return models.Where(model => !model.Id.Contains("image", StringComparison.OrdinalIgnoreCase) &&
                                      !model.Id.Contains("realtime", StringComparison.OrdinalIgnoreCase) &&
                                      !model.Id.Contains("audio", StringComparison.OrdinalIgnoreCase) &&
                                      !model.Id.Contains("tts", StringComparison.OrdinalIgnoreCase) &&
-                                     !model.Id.Contains("transcribe", StringComparison.OrdinalIgnoreCase) &&
+                                     !model.Id.Contains("transcribe", StringComparison.OrdinalIgnoreCase));
-                                     !model.Id.Contains("o1-pro", StringComparison.OrdinalIgnoreCase));
     }
 
     /// <inheritdoc />
@@ -147,6 +217,26 @@ public sealed class ProviderOpenAI(ILogger logger) : BaseProvider("https://api.o
     {
         var modelName = model.Id.ToLowerInvariant().AsSpan();
         
+        if (modelName is "gpt-4o-search-preview")
+            return
+                [
+                    Capability.TEXT_INPUT,
+                    Capability.TEXT_OUTPUT,
+                    
+                    Capability.WEB_SEARCH,
+                    Capability.CHAT_COMPLETION_API,
+                ];
+        
+        if (modelName is "gpt-4o-mini-search-preview")
+            return
+                [
+                    Capability.TEXT_INPUT,
+                    Capability.TEXT_OUTPUT,
+                    
+                    Capability.WEB_SEARCH,
+                    Capability.CHAT_COMPLETION_API,
+                ];
+        
         if (modelName.StartsWith("o1-mini"))
             return
                 [
@@ -154,32 +244,63 @@ public sealed class ProviderOpenAI(ILogger logger) : BaseProvider("https://api.o
                     Capability.TEXT_OUTPUT,
                     
                     Capability.ALWAYS_REASONING,
+                    Capability.CHAT_COMPLETION_API,
                 ];
         
+        if(modelName is "gpt-3.5-turbo")
+            return
+            [
+                Capability.TEXT_INPUT,
+                Capability.TEXT_OUTPUT,
+                Capability.RESPONSES_API,
+            ];
+        
+        if(modelName.StartsWith("gpt-3.5"))
+            return
+            [
+                Capability.TEXT_INPUT,
+                Capability.TEXT_OUTPUT,
+                Capability.CHAT_COMPLETION_API,
+            ];
+
+        if (modelName.StartsWith("chatgpt-4o-"))
+            return
+            [
+                Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
+                Capability.TEXT_OUTPUT,
+                Capability.RESPONSES_API,
+            ];
+        
         if (modelName.StartsWith("o3-mini"))
             return
                 [
                     Capability.TEXT_INPUT,
                     Capability.TEXT_OUTPUT,
                     
-                    Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING
+                    Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING,
+                    Capability.RESPONSES_API,
                 ];
         
-        if (modelName.StartsWith("o4-mini") || modelName.StartsWith("o1") || modelName.StartsWith("o3"))
+        if (modelName.StartsWith("o4-mini") || modelName.StartsWith("o3"))
             return
                 [
                     Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
                     Capability.TEXT_OUTPUT,
                     
-                    Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING
+                    Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING,
+                    Capability.WEB_SEARCH,
+                    Capability.RESPONSES_API,
                 ];
         
-        if(modelName.StartsWith("gpt-3.5"))
+        if (modelName.StartsWith("o1"))
             return
-                [
+            [
-                    Capability.TEXT_INPUT,
+                Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
-                    Capability.TEXT_OUTPUT,
+                Capability.TEXT_OUTPUT,
-                ];
+                    
+                Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING,
+                Capability.RESPONSES_API,
+            ];
         
         if(modelName.StartsWith("gpt-4-turbo"))
             return
@@ -187,7 +308,8 @@ public sealed class ProviderOpenAI(ILogger logger) : BaseProvider("https://api.o
                     Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
                     Capability.TEXT_OUTPUT,
                     
-                    Capability.FUNCTION_CALLING
+                    Capability.FUNCTION_CALLING,
+                    Capability.RESPONSES_API,
                 ];
         
         if(modelName is "gpt-4" || modelName.StartsWith("gpt-4-"))
@@ -195,14 +317,37 @@ public sealed class ProviderOpenAI(ILogger logger) : BaseProvider("https://api.o
                 [
                     Capability.TEXT_INPUT,
                     Capability.TEXT_OUTPUT,
+                    Capability.RESPONSES_API,
                 ];
         
+        if(modelName.StartsWith("gpt-5-nano"))
+            return
+            [
+                Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
+                Capability.TEXT_OUTPUT,
+                
+                Capability.FUNCTION_CALLING, Capability.ALWAYS_REASONING,
+                Capability.RESPONSES_API,
+            ];
+        
+        if(modelName is "gpt-5" || modelName.StartsWith("gpt-5-"))
+            return
+            [
+                Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
+                Capability.TEXT_OUTPUT,
+                
+                Capability.FUNCTION_CALLING, Capability.ALWAYS_REASONING,
+                Capability.WEB_SEARCH,
+                Capability.RESPONSES_API,
+            ];
+        
         return
             [
                 Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
                 Capability.TEXT_OUTPUT,
                 
                 Capability.FUNCTION_CALLING,
+                Capability.RESPONSES_API,
             ];
     }
     

app/MindWork AI Studio/Provider/OpenAI/ResponseStreamLine.cs

@@ -1,32 +0,0 @@
-namespace AIStudio.Provider.OpenAI;
-
-/// <summary>
-/// Data model for a line in the response stream, for streaming completions.
-/// </summary>
-/// <param name="Id">The id of the response.</param>
-/// <param name="Object">The object describing the response.</param>
-/// <param name="Created">The timestamp of the response.</param>
-/// <param name="Model">The model used for the response.</param>
-/// <param name="SystemFingerprint">The system fingerprint; together with the seed, this allows you to reproduce the response.</param>
-/// <param name="Choices">The choices made by the AI.</param>
-public readonly record struct ResponseStreamLine(string Id, string Object, uint Created, string Model, string SystemFingerprint, IList<Choice> Choices) : IResponseStreamLine
-{
-    /// <inheritdoc />
-    public bool ContainsContent() => this != default && this.Choices.Count > 0;
-
-    /// <inheritdoc />
-    public ContentStreamChunk GetContent() => new(this.Choices[0].Delta.Content, []);
-}
-
-/// <summary>
-/// Data model for a choice made by the AI.
-/// </summary>
-/// <param name="Index">The index of the choice.</param>
-/// <param name="Delta">The delta text of the choice.</param>
-public readonly record struct Choice(int Index, Delta Delta);
-
-/// <summary>
-/// The delta text of a choice.
-/// </summary>
-/// <param name="Content">The content of the delta text.</param>
-public readonly record struct Delta(string Content);

app/MindWork AI Studio/Provider/OpenAI/ResponsesAPIRequest.cs

@@ -0,0 +1,21 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// The request body for the Responses API.
+/// </summary>
+/// <param name="Model">Which model to use.</param>
+/// <param name="Input">The chat messages.</param>
+/// <param name="Stream">Whether to stream the response.</param>
+/// <param name="Store">Whether to store the response on the server (usually OpenAI's infrastructure).</param>
+/// <param name="Tools">The tools to use for the request.</param>
+public record ResponsesAPIRequest(
+    string Model,
+    IList<Message> Input,
+    bool Stream,
+    bool Store,
+    IList<Tool> Tools)
+{
+    public ResponsesAPIRequest() : this(string.Empty, [], true, false, [])
+    {
+    }
+}

app/MindWork AI Studio/Provider/OpenAI/ResponsesAnnotatingURL.cs

@@ -0,0 +1,16 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Data structure for URL citation annotations in the OpenAI Responses API.
+/// </summary>
+/// <param name="Type">The type of annotation, typically "url_citation".</param>
+/// <param name="EndIndex">The end index of the annotated text in the response.</param>
+/// <param name="StartIndex">The start index of the annotated text in the response.</param>
+/// <param name="Title">The title of the cited URL.</param>
+/// <param name="URL">The URL being cited.</param>
+public sealed record ResponsesAnnotatingUrlCitationData(
+    string Type,
+    int EndIndex,
+    int StartIndex,
+    string Title,
+    string URL) : Annotation(Type);

app/MindWork AI Studio/Provider/OpenAI/ResponsesAnnotationStreamLine.cs

@@ -0,0 +1,45 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Data structure for a line in the response stream of the Responses API, containing an annotation.
+/// </summary>
+/// <param name="Type">The type of the annotation.</param>
+/// <param name="AnnotationIndex">The continuous index of the annotation in the response.</param>
+/// <param name="Annotation">The annotation details.</param>
+public sealed record ResponsesAnnotationStreamLine(string Type, int AnnotationIndex, Annotation Annotation) : IAnnotationStreamLine
+{
+    #region Implementation of IAnnotationStreamLine
+
+    /// <inheritdoc />
+    public bool ContainsSources()
+    {
+        return this.Annotation is not AnnotatingUnknown;
+    }
+
+    /// <inheritdoc />
+    public IList<ISource> GetSources()
+    {
+        //
+        // Check for the unexpected annotation type of the chat completion API.
+        //
+        // This seems weird at first. But there are two possibilities why this could happen:
+        // - Anyone of the open source providers such as ollama, LM Studio, etc. could
+        //   implement and use the chat completion API data structures for annotations in their
+        //   Responses API endpoint.
+        //
+        // - Our custom JSON converter checks for all possible annotation data types. So,
+        //   when the streamed data is valid for any annotation type, it will be deserialized
+        //   into that type, even though we are calling the Responses API endpoint.
+        //
+        if (this.Annotation is ChatCompletionAnnotatingURL urlAnnotation)
+            return [new Source(urlAnnotation.UrlCitation.Title, urlAnnotation.UrlCitation.URL)];
+        
+        // Check for the expected annotation type of the Responses API:
+        if (this.Annotation is ResponsesAnnotatingUrlCitationData urlCitationData)
+            return [new Source(urlCitationData.Title, urlCitationData.URL)];
+
+        return [];
+    }
+
+    #endregion
+}

app/MindWork AI Studio/Provider/OpenAI/ResponsesDeltaStreamLine.cs

@@ -0,0 +1,39 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Data model for a delta line in the Response API chat completion stream.
+/// </summary>
+/// <param name="Type">The type of the response.</param>
+/// <param name="Delta">The delta content of the response.</param>
+public record ResponsesDeltaStreamLine(
+    string Type,
+    string Delta) : IResponseStreamLine
+{
+    #region Implementation of IResponseStreamLine
+
+    /// <inheritdoc />
+    public bool ContainsContent() => !string.IsNullOrWhiteSpace(this.Delta);
+
+    /// <inheritdoc />
+    public ContentStreamChunk GetContent() => new(this.Delta, this.GetSources());
+
+    //
+    // Please note that there are multiple options where LLM providers might stream sources:
+    //
+    // - As part of the delta content while streaming. That would be part of this class.
+    // - By using a dedicated stream event and data structure. That would be another class implementing IResponseStreamLine.
+    //
+    // Right now, OpenAI uses the latter approach, so we don't have any sources here. And
+    // because no other provider does it yet, we don't have any implementation here either.
+    //
+    // One example where sources are part of the delta content is the Perplexity provider.
+    //
+    
+    /// <inheritdoc />
+    public bool ContainsSources() => false;
+
+    /// <inheritdoc />
+    public IList<ISource> GetSources() => [];
+
+    #endregion
+}

app/MindWork AI Studio/Provider/OpenAI/Tool.cs

@@ -0,0 +1,12 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Represents a tool used by the AI model.
+/// </summary>
+/// <remarks>
+/// Right now, only our OpenAI provider is using tools. Thus, this class is located in the
+/// OpenAI namespace. In the future, when other providers also support tools, this class can
+/// be moved into the provider namespace.
+/// </remarks>
+/// <param name="Type">The type of the tool.</param>
+public record Tool(string Type);

app/MindWork AI Studio/Provider/OpenAI/Tools.cs

@@ -0,0 +1,14 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// Known tools for LLM providers.
+/// </summary>
+/// <remarks>
+/// Right now, only our OpenAI provider is using tools. Thus, this class is located in the
+/// OpenAI namespace. In the future, when other providers also support tools, this class can
+/// be moved into the provider namespace.
+/// </remarks>
+public static class Tools
+{
+    public static readonly Tool WEB_SEARCH = new("web_search");
+}

app/MindWork AI Studio/Provider/Perplexity/Choice.cs

@@ -0,0 +1,8 @@
+namespace AIStudio.Provider.Perplexity;
+
+/// <summary>
+/// Data model for a choice made by the AI.
+/// </summary>
+/// <param name="Index">The index of the choice.</param>
+/// <param name="Delta">The delta text of the choice.</param>
+public readonly record struct Choice(int Index, Delta Delta);

app/MindWork AI Studio/Provider/Perplexity/Delta.cs

@@ -0,0 +1,7 @@
+namespace AIStudio.Provider.Perplexity;
+
+/// <summary>
+/// The delta text of a choice.
+/// </summary>
+/// <param name="Content">The content of the delta text.</param>
+public readonly record struct Delta(string Content);

app/MindWork AI Studio/Provider/Perplexity/ProviderPerplexity.cs

@@ -44,7 +44,7 @@ public sealed class ProviderPerplexity(ILogger logger) : BaseProvider("https://a
         };
         
         // Prepare the Perplexity HTTP chat request:
-        var perplexityChatRequest = JsonSerializer.Serialize(new ChatRequest
+        var perplexityChatRequest = JsonSerializer.Serialize(new ChatCompletionAPIRequest
         {
             Model = chatModel.Id,
             
@@ -85,7 +85,7 @@ public sealed class ProviderPerplexity(ILogger logger) : BaseProvider("https://a
             return request;
         }
         
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("Perplexity", RequestBuilder, token))
+        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine, NoChatCompletionAnnotationStreamLine>("Perplexity", RequestBuilder, token))
             yield return content;
     }
 
@@ -130,6 +130,8 @@ public sealed class ProviderPerplexity(ILogger logger) : BaseProvider("https://a
                 Capability.IMAGE_OUTPUT,
                 
                 Capability.ALWAYS_REASONING,
+                Capability.WEB_SEARCH,
+                Capability.CHAT_COMPLETION_API,
             ];
         
         return
@@ -139,6 +141,9 @@ public sealed class ProviderPerplexity(ILogger logger) : BaseProvider("https://a
             
             Capability.TEXT_OUTPUT,
             Capability.IMAGE_OUTPUT,
+            
+            Capability.WEB_SEARCH,
+            Capability.CHAT_COMPLETION_API,
         ];
     }
     

app/MindWork AI Studio/Provider/Perplexity/ResponseStreamLine.cs

@@ -23,23 +23,3 @@ public readonly record struct ResponseStreamLine(string Id, string Object, uint
     /// <inheritdoc />
     public IList<ISource> GetSources() => this.SearchResults.Cast<ISource>().ToList();
 }
-
-/// <summary>
-/// Data model for a choice made by the AI.
-/// </summary>
-/// <param name="Index">The index of the choice.</param>
-/// <param name="Delta">The delta text of the choice.</param>
-public readonly record struct Choice(int Index, Delta Delta);
-
-/// <summary>
-/// The delta text of a choice.
-/// </summary>
-/// <param name="Content">The content of the delta text.</param>
-public readonly record struct Delta(string Content);
-
-/// <summary>
-/// Data model for a search result.
-/// </summary>
-/// <param name="Title">The title of the search result.</param>
-/// <param name="URL">The URL of the search result.</param>
-public sealed record SearchResult(string Title, string URL) : Source(Title, URL);

app/MindWork AI Studio/Provider/Perplexity/SearchResult.cs

@@ -0,0 +1,8 @@
+namespace AIStudio.Provider.Perplexity;
+
+/// <summary>
+/// Data model for a search result.
+/// </summary>
+/// <param name="Title">The title of the search result.</param>
+/// <param name="URL">The URL of the search result.</param>
+public sealed record SearchResult(string Title, string URL) : Source(Title, URL);

app/MindWork AI Studio/Provider/SelfHosted/ProviderSelfHosted.cs

@@ -75,7 +75,7 @@ public sealed class ProviderSelfHosted(ILogger logger, Host host, string hostnam
             return request;
         }
         
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("self-hosted provider", RequestBuilder, token))
+        await foreach (var content in this.StreamChatCompletionInternal<ChatCompletionDeltaStreamLine, ChatCompletionAnnotationStreamLine>("self-hosted provider", RequestBuilder, token))
             yield return content;
     }
 

app/MindWork AI Studio/Provider/X/ProviderX.cs

@@ -35,7 +35,7 @@ public sealed class ProviderX(ILogger logger) : BaseProvider("https://api.x.ai/v
         };
         
         // Prepare the xAI HTTP chat request:
-        var xChatRequest = JsonSerializer.Serialize(new ChatRequest
+        var xChatRequest = JsonSerializer.Serialize(new ChatCompletionAPIRequest
         {
             Model = chatModel.Id,
             
@@ -61,8 +61,6 @@ public sealed class ProviderX(ILogger logger) : BaseProvider("https://api.x.ai/v
                 }
             }).ToList()],
             
-            Seed = chatThread.Seed,
-            
             // Right now, we only support streaming completions:
             Stream = true,
         }, JSON_SERIALIZER_OPTIONS);
@@ -80,7 +78,7 @@ public sealed class ProviderX(ILogger logger) : BaseProvider("https://api.x.ai/v
             return request;
         }
         
-        await foreach (var content in this.StreamChatCompletionInternal<ResponseStreamLine>("xAI", RequestBuilder, token))
+        await foreach (var content in this.StreamChatCompletionInternal<ChatCompletionDeltaStreamLine, NoChatCompletionAnnotationStreamLine>("xAI", RequestBuilder, token))
             yield return content;
     }
 

app/MindWork AI Studio/wwwroot/changelog/v0.9.51.md

@@ -4,11 +4,14 @@
 - Added the ability to control the update installation behavior by configuration plugins.
 - Added the option for LLM providers to stream citations or sources.
 - Added support for citations to the chat interface. This feature is invisible unless an LLM model is streaming citations or sources.
+- Added the Responses API according to the OpenAI documentation. It is currently only used by OpenAI, but we could use the API for other providers as soon as someone offers it. This means that all text-based LLMs from OpenAI can now be used in MindWork AI Studio. For example, the Deep Research models for comprehensive research tasks.
+- Added support for web searches. Currently supported by some OpenAI models (e.g., GPT5, newer Omni models, Deep Research models) and Perplexity. Used sources are displayed visually in the chat interface.
 - Improved memory usage in several areas of the app.
 - Improved plugin management for configuration plugins so that hot reload detects when a provider or chat template has been removed.
 - Improved the dialog for naming chats and workspaces to ensure valid inputs are entered.
 - Improved the dialog invocation by making parameter provision more robust.
 - Improved the text summarizer assistant by allowing users to specify important aspects & optimized the generated prompt.
+- Improved the OpenAI provider by supporting more models and capabilities.
 - Changed the configuration plugin setting name for how often to check for updates from `UpdateBehavior` to `UpdateInterval`.
 - Fixed a bug in various assistants where some text fields were not reset when resetting.
 - Fixed the input field header in the dialog for naming chats and workspaces.