Add support for client providers

Author: kunitoki

examples/graphics/source/examples/AI.h

@@ -91,7 +91,10 @@ LLMResponse LLMClient::runToolLoop (const Request& request, LLMToolRegistry& too
         for (const auto& toolCall : response.getToolCalls())
         {
             auto result = tools.dispatchToolCall (toolCall.name, toolCall.arguments);
-            current.messages.push_back (LLMMessage::toolResult (toolCall.id, JSON::toString (result, true)));
+
+            auto toolResultMsg = LLMMessage::toolResult (toolCall.id, JSON::toString (result, true));
+            toolResultMsg.name = toolCall.name; // preserved for providers that need name + id separately (e.g. Gemini)
+            current.messages.push_back (std::move (toolResultMsg));
         }
 
         response = complete (current);
@@ -124,14 +127,19 @@ String LLMClient::buildChatCompletionBody (const Request& request, bool stream)
     if (request.toolChoice.has_value())
         setLLMClientProperty (object, "tool_choice", toolChoiceToVar (*request.toolChoice));
 
-    if (request.temperature.has_value())
-        setLLMClientProperty (object, "temperature", static_cast<double> (*request.temperature));
+    if (! options.noTemperature)
+    {
+        if (request.temperature.has_value())
+            setLLMClientProperty (object, "temperature", static_cast<double> (*request.temperature));
+    }
 
     if (request.topP.has_value())
         setLLMClientProperty (object, "top_p", static_cast<double> (*request.topP));
 
-    if (request.maxTokens.has_value())
-        setLLMClientProperty (object, "max_tokens", *request.maxTokens);
+    // Per-request maxTokens overrides options.maxTokens; use max_completion_tokens for OpenAI-compatible APIs.
+    const int effectiveMaxTokens = request.maxTokens.value_or (options.maxTokens);
+    if (effectiveMaxTokens > 0)
+        setLLMClientProperty (object, "max_completion_tokens", effectiveMaxTokens);
 
     if (request.stopSequences.has_value())
     {
@@ -143,6 +151,40 @@ String LLMClient::buildChatCompletionBody (const Request& request, bool stream)
         setLLMClientProperty (object, "stop", stop);
     }
 
+    // Reasoning effort for o-series / GPT-5 models.
+    if (options.reasoningEffort.isNotEmpty())
+        setLLMClientProperty (object, "reasoning_effort", options.reasoningEffort);
+
+    // GBNF grammar for llama-server constrained decoding (per-request overrides config).
+    const auto& effectiveGrammar = request.grammar.isNotEmpty() ? request.grammar : options.grammar;
+    if (effectiveGrammar.isNotEmpty())
+        setLLMClientProperty (object, "grammar", effectiveGrammar);
+
+    // Prompt caching — bucket by application identity, retain for 24h.
+    if (options.userAgent.isNotEmpty())
+    {
+        setLLMClientProperty (object, "prompt_cache_key", options.userAgent);
+        setLLMClientProperty (object, "prompt_cache_retention", String ("24h"));
+    }
+
+    // Structured output via JSON Schema (built with LLMSchema helpers).
+    if (! request.schema.isVoid())
+    {
+        auto schemaWrapper = makeLLMClientObject();
+        setLLMClientProperty (schemaWrapper, "name", String ("response"));
+        setLLMClientProperty (schemaWrapper, "strict", true);
+        setLLMClientProperty (schemaWrapper, "schema", request.schema);
+
+        auto responseFormat = makeLLMClientObject();
+        setLLMClientProperty (responseFormat, "type", String ("json_schema"));
+        setLLMClientProperty (responseFormat, "json_schema", schemaWrapper);
+
+        setLLMClientProperty (object, "response_format", responseFormat);
+    }
+
+    // OpenRouter — application identification headers are injected at HTTP level,
+    // but some frontends read X-Title from the body; we skip that here.
+
     return JSON::toString (object, true);
 }
 

modules/yup_ai/llm/yup_LLMClient.cpp

@@ -30,26 +30,49 @@ namespace yup
 class YUP_API LLMClient
 {
 public:
+    /** The LLM provider type, used by LLMClientFactory to create the correct client. */
+    enum class Provider
+    {
+        OpenAIChat,      ///< OpenAI Chat Completions — also works with DeepSeek, OpenRouter, Ollama, llama-server.
+        OpenAIResponses, ///< OpenAI Responses API (GPT-5+).
+        Anthropic,       ///< Anthropic Messages API — Claude models.
+        Gemini           ///< Google Gemini generateContent API.
+    };
+
     struct Request
     {
         std::vector<LLMMessage> messages;
         std::optional<String> systemPrompt;
         std::vector<LLMTool> tools;
-        std::optional<String> toolChoice;
+        std::optional<String> toolChoice; ///< "auto", "none", "required", or a specific function name.
 
         std::optional<float> temperature;
         std::optional<float> topP;
-        std::optional<int> maxTokens;
+        std::optional<int> maxTokens; ///< Per-request override; falls back to Options::maxTokens.
         std::optional<std::vector<String>> stopSequences;
+
+        var schema;                    ///< Optional JSON Schema for structured output (built with LLMSchema).
+        String grammar;                ///< Optional per-request GBNF (llama-server) or Lark (OpenAI Responses) grammar.
+        String grammarToolName;        ///< Tool name for grammar-constrained output (OpenAI Responses API only).
+        String grammarToolDescription; ///< Tool description for grammar output; defaults to system prompt if empty.
     };
 
     struct Options
     {
+        Provider provider = Provider::OpenAIChat; ///< LLM backend provider — used by LLMClientFactory.
+
         String model;
         String baseUrl = "http://localhost:11434/v1";
         String apiKey;
         int timeoutMs = 120000;
         int maxRetries = 2;
+        int maxTokens = 0; ///< Default max output tokens (0 = provider default); per-request value overrides.
+
+        String reasoningEffort;     ///< "none", "low", "medium", "high" — for OpenAI o-series and Gemini 2.5 models.
+        String grammar;             ///< Default GBNF grammar for llama-server constrained decoding (per-request overrides).
+        bool noTemperature = false; ///< Set true for models that reject the temperature parameter (e.g. GPT-5 series).
+        String userAgent;           ///< Application identifier used for User-Agent header and prompt cache key.
+        String appUrl;              ///< Application URL sent as HTTP-Referer on OpenRouter requests.
     };
 
     explicit LLMClient (Options options);

modules/yup_ai/llm/yup_LLMClient.h

@@ -0,0 +1,96 @@
+/*
+  ==============================================================================
+
+   This file is part of the YUP library.
+   Copyright (c) 2026 - kunitoki@gmail.com
+
+   YUP is an open source library subject to open-source licensing.
+
+   The code included in this file is provided under the terms of the ISC license
+   http://www.isc.org/downloads/software-support-policy/isc-license. Permission
+   to use, copy, modify, and/or distribute this software for any purpose with or
+   without fee is hereby granted provided that the above copyright notice and
+   this permission notice appear in all copies.
+
+   YUP IS PROVIDED "AS IS" WITHOUT ANY WARRANTY, AND ALL WARRANTIES, WHETHER
+   EXPRESSED OR IMPLIED, INCLUDING MERCHANTABILITY AND FITNESS FOR PURPOSE, ARE
+   DISCLAIMED.
+
+  ==============================================================================
+*/
+
+namespace yup
+{
+
+std::unique_ptr<LLMClient> LLMClientFactory::create (LLMClient::Options options)
+{
+    switch (options.provider)
+    {
+        case LLMClient::Provider::OpenAIChat:
+            return std::make_unique<LLMOpenAIChatClient> (std::move (options));
+
+        case LLMClient::Provider::OpenAIResponses:
+            return std::make_unique<LLMOpenAIResponsesClient> (std::move (options));
+
+        case LLMClient::Provider::Anthropic:
+            return std::make_unique<LLMAnthropicClient> (std::move (options));
+
+        case LLMClient::Provider::Gemini:
+            return std::make_unique<LLMGeminiClient> (std::move (options));
+
+        default:
+            jassertfalse; // Unknown provider
+            return nullptr;
+    }
+}
+
+//==============================================================================
+std::unique_ptr<LLMClient> LLMClientFactory::openAIChat (String model,
+                                                         String baseUrl,
+                                                         String apiKey)
+{
+    LLMClient::Options opts;
+    opts.provider = LLMClient::Provider::OpenAIChat;
+    opts.model = std::move (model);
+    opts.baseUrl = std::move (baseUrl);
+    opts.apiKey = std::move (apiKey);
+    return create (std::move (opts));
+}
+
+std::unique_ptr<LLMClient> LLMClientFactory::openAIResponses (String model,
+                                                              String apiKey,
+                                                              String baseUrl)
+{
+    LLMClient::Options opts;
+    opts.provider = LLMClient::Provider::OpenAIResponses;
+    opts.model = std::move (model);
+    opts.apiKey = std::move (apiKey);
+    opts.baseUrl = std::move (baseUrl);
+    return create (std::move (opts));
+}
+
+std::unique_ptr<LLMClient> LLMClientFactory::anthropic (String model,
+                                                        String apiKey,
+                                                        String baseUrl)
+{
+    LLMClient::Options opts;
+    opts.provider = LLMClient::Provider::Anthropic;
+    opts.model = std::move (model);
+    opts.apiKey = std::move (apiKey);
+    opts.baseUrl = std::move (baseUrl);
+    return create (std::move (opts));
+}
+
+std::unique_ptr<LLMClient> LLMClientFactory::gemini (String model,
+                                                     String apiKey,
+                                                     String baseUrl)
+{
+    LLMClient::Options opts;
+    opts.provider = LLMClient::Provider::Gemini;
+    opts.model = std::move (model);
+    opts.apiKey = std::move (apiKey);
+    opts.baseUrl = std::move (baseUrl);
+    return create (std::move (opts));
+}
+
+} // namespace yup

modules/yup_ai/llm/yup_LLMClientFactory.cpp

@@ -0,0 +1,79 @@
+/*
+  ==============================================================================
+
+   This file is part of the YUP library.
+   Copyright (c) 2026 - kunitoki@gmail.com
+
+   YUP is an open source library subject to open-source licensing.
+
+   The code included in this file is provided under the terms of the ISC license
+   http://www.isc.org/downloads/software-support-policy/isc-license. Permission
+   to use, copy, modify, and/or distribute this software for any purpose with or
+   without fee is hereby granted provided that the above copyright notice and
+   this permission notice appear in all copies.
+
+   YUP IS PROVIDED "AS IS" WITHOUT ANY WARRANTY, AND ALL WARRANTIES, WHETHER
+   EXPRESSED OR IMPLIED, INCLUDING MERCHANTABILITY AND FITNESS FOR PURPOSE, ARE
+   DISCLAIMED.
+
+  ==============================================================================
+*/
+
+namespace yup
+{
+
+//==============================================================================
+/** Factory that creates the correct LLMHttpClient subclass from an Options struct.
+
+    Use LLMClientFactory::create() to instantiate an LLM client for any supported
+    provider.  The Provider enum in LLMClient::Options selects the concrete class.
+
+    @code
+    yup::LLMClient::Options opts;
+    opts.provider  = yup::LLMClient::Provider::Anthropic;
+    opts.model     = "claude-opus-4-5";
+    opts.apiKey    = "sk-ant-...";
+    opts.baseUrl   = "https://api.anthropic.com/v1";
+
+    auto client = yup::LLMClientFactory::create (opts);
+    auto response = client->chat ("Hello, Claude!");
+    @endcode
+
+    Convenience static methods are provided for the most common provider setups.
+
+    @tags{AI}
+*/
+class YUP_API LLMClientFactory
+{
+public:
+    /** Creates an LLM client for the provider specified in @p options.
+
+        @param options  Full options struct.  options.provider selects the concrete class.
+        @returns        A heap-allocated concrete LLMHttpClient subclass, or nullptr if
+                        the provider enum value is unrecognised.
+    */
+    static std::unique_ptr<LLMClient> create (LLMClient::Options options);
+
+    //==============================================================================
+    /** Convenience factory — OpenAI Chat Completions (also Ollama, DeepSeek, OpenRouter, llama-server). */
+    static std::unique_ptr<LLMClient> openAIChat (String model,
+                                                  String baseUrl = "http://localhost:11434/v1",
+                                                  String apiKey = {});
+
+    /** Convenience factory — OpenAI Responses API (GPT-5+, reasoning models). */
+    static std::unique_ptr<LLMClient> openAIResponses (String model,
+                                                       String apiKey,
+                                                       String baseUrl = "https://api.openai.com/v1");
+
+    /** Convenience factory — Anthropic Messages API (Claude models). */
+    static std::unique_ptr<LLMClient> anthropic (String model,
+                                                 String apiKey,
+                                                 String baseUrl = "https://api.anthropic.com/v1");
+
+    /** Convenience factory — Google Gemini generateContent API. */
+    static std::unique_ptr<LLMClient> gemini (String model,
+                                              String apiKey,
+                                              String baseUrl = "https://generativelanguage.googleapis.com");
+};
+
+} // namespace yup