Implement support for self-hosted and local LLMs (#20)

app/MindWork AI Studio/Provider/Providers.cs

@@ -1,6 +1,7 @@
 using AIStudio.Provider.Anthropic;
 using AIStudio.Provider.Mistral;
 using AIStudio.Provider.OpenAI;
+using AIStudio.Provider.SelfHosted;
 
 namespace AIStudio.Provider;
 
@@ -10,9 +11,12 @@ namespace AIStudio.Provider;
 public enum Providers
 {
     NONE,
+    
     OPEN_AI,
     ANTHROPIC,
     MISTRAL,
+    
+    SELF_HOSTED,
 }
 
 /// <summary>
@@ -27,11 +31,14 @@ public static class ExtensionsProvider
     /// <returns>The human-readable name of the provider.</returns>
     public static string ToName(this Providers provider) => provider switch
     {
+        Providers.NONE => "No provider selected",
+        
         Providers.OPEN_AI => "OpenAI",
         Providers.ANTHROPIC => "Anthropic",
         Providers.MISTRAL => "Mistral",
         
-        Providers.NONE => "No provider selected",
+        Providers.SELF_HOSTED => "Self-hosted",
+        
         _ => "Unknown",
     };
 
@@ -40,13 +47,16 @@ public static class ExtensionsProvider
     /// </summary>
     /// <param name="provider">The provider value.</param>
     /// <param name="instanceName">The used instance name.</param>
+    /// <param name="hostname">The hostname of the provider.</param>
     /// <returns>The provider instance.</returns>
-    public static IProvider CreateProvider(this Providers provider, string instanceName) => provider switch
+    public static IProvider CreateProvider(this Providers provider, string instanceName, string hostname = "http://localhost:1234") => provider switch
     {
         Providers.OPEN_AI => new ProviderOpenAI { InstanceName = instanceName },
         Providers.ANTHROPIC => new ProviderAnthropic { InstanceName = instanceName },
         Providers.MISTRAL => new ProviderMistral { InstanceName = instanceName },
         
+        Providers.SELF_HOSTED => new ProviderSelfHosted(hostname) { InstanceName = instanceName },
+        
         _ => new NoProvider(),
     };
 }

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

@@ -0,0 +1,16 @@
+namespace AIStudio.Provider.SelfHosted;
+
+/// <summary>
+/// The 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="MaxTokens">The maximum number of tokens to generate.</param>
+public readonly record struct ChatRequest(
+    string Model,
+    IList<Message> Messages,
+    bool Stream,
+    
+    int MaxTokens
+);

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

@@ -0,0 +1,8 @@
+namespace AIStudio.Provider.SelfHosted;
+
+/// <summary>
+/// Chat message model.
+/// </summary>
+/// <param name="Content">The text content of the message.</param>
+/// <param name="Role">The role of the message.</param>
+public readonly record struct Message(string Content, string Role);

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

@@ -0,0 +1,5 @@
+namespace AIStudio.Provider.SelfHosted;
+
+public readonly record struct ModelsResponse(string Object, Model[] Data);
+
+public readonly record struct Model(string Id, string Object, string OwnedBy);

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

@@ -0,0 +1,162 @@
+using System.Runtime.CompilerServices;
+using System.Text;
+using System.Text.Json;
+
+using AIStudio.Chat;
+using AIStudio.Provider.OpenAI;
+using AIStudio.Settings;
+
+namespace AIStudio.Provider.SelfHosted;
+
+public sealed class ProviderSelfHosted(string hostname) : BaseProvider($"{hostname}/v1/"), IProvider
+{
+    private static readonly JsonSerializerOptions JSON_SERIALIZER_OPTIONS = new()
+    {
+        PropertyNamingPolicy = JsonNamingPolicy.SnakeCaseLower,
+    };
+
+    #region Implementation of IProvider
+
+    public string Id => "Self-hosted";
+    
+    public string InstanceName { get; set; } = "Self-hosted";
+    
+    public async IAsyncEnumerable<string> StreamChatCompletion(IJSRuntime jsRuntime, SettingsManager settings, Provider.Model chatModel, ChatThread chatThread, [EnumeratorCancellation] CancellationToken token = default)
+    {
+        // Prepare the system prompt:
+        var systemPrompt = new Message
+        {
+            Role = "system",
+            Content = chatThread.SystemPrompt,
+        };
+        
+        // Prepare the OpenAI HTTP chat request:
+        var providerChatRequest = JsonSerializer.Serialize(new ChatRequest
+        {
+            Model = (await this.GetTextModels(jsRuntime, settings, token: token)).First().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
+            {
+                Role = n.Role switch
+                {
+                    ChatRole.USER => "user",
+                    ChatRole.AI => "assistant",
+                    ChatRole.SYSTEM => "system",
+
+                    _ => "user",
+                },
+
+                Content = n.Content switch
+                {
+                    ContentText text => text.Text,
+                    _ => string.Empty,
+                }
+            }).ToList()],
+            
+            // Right now, we only support streaming completions:
+            Stream = true,
+            MaxTokens = -1,
+        }, JSON_SERIALIZER_OPTIONS);
+
+        // Build the HTTP post request:
+        var request = new HttpRequestMessage(HttpMethod.Post, "chat/completions");
+        
+        // Set the content:
+        request.Content = new StringContent(providerChatRequest, Encoding.UTF8, "application/json");
+        
+        // Send the request with the ResponseHeadersRead option.
+        // This allows us to read the stream as soon as the headers are received.
+        // This is important because we want to stream the responses.
+        var response = await this.httpClient.SendAsync(request, HttpCompletionOption.ResponseHeadersRead, token);
+        
+        // Open the response stream:
+        var providerStream = await response.Content.ReadAsStreamAsync(token);
+        
+        // Add a stream reader to read the stream, line by line:
+        var streamReader = new StreamReader(providerStream);
+        
+        // Read the stream, line by line:
+        while(!streamReader.EndOfStream)
+        {
+            // Check if the token is cancelled:
+            if(token.IsCancellationRequested)
+                yield break;
+            
+            // Read the next line:
+            var line = await streamReader.ReadLineAsync(token);
+            
+            // Skip empty lines:
+            if(string.IsNullOrWhiteSpace(line))
+                continue;
+            
+            // Skip lines that do not start with "data: ". Regard
+            // to the specification, we only want to read the data lines:
+            if(!line.StartsWith("data: ", StringComparison.InvariantCulture))
+                continue;
+
+            // Check if the line is the end of the stream:
+            if (line.StartsWith("data: [DONE]", StringComparison.InvariantCulture))
+                yield break;
+
+            ResponseStreamLine 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<ResponseStreamLine>(jsonData, JSON_SERIALIZER_OPTIONS);
+            }
+            catch
+            {
+                // Skip invalid JSON data:
+                continue;
+            }    
+            
+            // Skip empty responses:
+            if(providerResponse == default || providerResponse.Choices.Count == 0)
+                continue;
+            
+            // Yield the response:
+            yield return providerResponse.Choices[0].Delta.Content;
+        }
+    }
+
+    #pragma warning disable CS1998 // Async method lacks 'await' operators and will run synchronously
+    /// <inheritdoc />
+    public async IAsyncEnumerable<ImageURL> StreamImageCompletion(IJSRuntime jsRuntime, SettingsManager settings, Provider.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
+
+
+    public async Task<IEnumerable<Provider.Model>> GetTextModels(IJSRuntime jsRuntime, SettingsManager settings, string? apiKeyProvisional = null, CancellationToken token = default)
+    {
+        var request = new HttpRequestMessage(HttpMethod.Get, "models");
+        var response = await this.httpClient.SendAsync(request, token);
+        if(!response.IsSuccessStatusCode)
+            return [];
+
+        var modelResponse = await response.Content.ReadFromJsonAsync<ModelsResponse>(token);
+        if (modelResponse.Data.Length > 1)
+            Console.WriteLine("Warning: multiple models found; using the first one.");
+        
+        var firstModel = modelResponse.Data.First();
+        return [ new Provider.Model(firstModel.Id) ];
+    }
+
+    #pragma warning disable CS1998 // Async method lacks 'await' operators and will run synchronously
+    /// <inheritdoc />
+    public Task<IEnumerable<Provider.Model>> GetImageModels(IJSRuntime jsRuntime, SettingsManager settings, string? apiKeyProvisional = null, CancellationToken token = default)
+    {
+        return Task.FromResult(Enumerable.Empty<Provider.Model>());
+    }
+    #pragma warning restore CS1998 // Async method lacks 'await' operators and will run synchronously
+
+    #endregion
+}