embed-model for OpenAI can not have its own base-url / api key

[pilhuhn]

I am trying to access 2 different remote OpenAI-API-compatible services for the chat and embed-models.
This fails, as the embedding model does not have its own base-url and api-key

Same applies to imageModel and moderationModel.

I think having those overrides is important as the OpenAI-API is kind of becoming a default API for many LLM-servers (InstructLab , Ollama in OpenAI mode, ..)

The following patch illustrates a potential fix, that works on my laptop :)

---
Index: model-providers/openai/openai-vanilla/runtime/src/main/java/io/quarkiverse/langchain4j/openai/runtime/config/EmbeddingModelConfig.java
IDEA additional info:
Subsystem: com.intellij.openapi.diff.impl.patch.CharsetEP
<+>UTF-8
===================================================================
diff --git a/model-providers/openai/openai-vanilla/runtime/src/main/java/io/quarkiverse/langchain4j/openai/runtime/config/EmbeddingModelConfig.java b/model-providers/openai/openai-vanilla/runtime/src/main/java/io/quarkiverse/langchain4j/openai/runtime/config/EmbeddingModelConfig.java
--- a/model-providers/openai/openai-vanilla/runtime/src/main/java/io/quarkiverse/langchain4j/openai/runtime/config/EmbeddingModelConfig.java	(revision a8864444916d915b178eda65ee7665934e0ac7e8)
+++ b/model-providers/openai/openai-vanilla/runtime/src/main/java/io/quarkiverse/langchain4j/openai/runtime/config/EmbeddingModelConfig.java	(date 1737111305494)
@@ -31,4 +31,14 @@
      * A unique identifier representing your end-user, which can help OpenAI to monitor and detect abuse.
      */
     Optional<String> user();
+
+    /**
+     * Base URL override for the embedding model
+     */
+    Optional<String> baseUrl();
+
+    /**
+     * Api key override for the embedding model
+     */
+    Optional<String> apiKey();
 }
Index: model-providers/openai/openai-vanilla/runtime/src/main/java/io/quarkiverse/langchain4j/openai/runtime/OpenAiRecorder.java
IDEA additional info:
Subsystem: com.intellij.openapi.diff.impl.patch.CharsetEP
<+>UTF-8
===================================================================
diff --git a/model-providers/openai/openai-vanilla/runtime/src/main/java/io/quarkiverse/langchain4j/openai/runtime/OpenAiRecorder.java b/model-providers/openai/openai-vanilla/runtime/src/main/java/io/quarkiverse/langchain4j/openai/runtime/OpenAiRecorder.java
--- a/model-providers/openai/openai-vanilla/runtime/src/main/java/io/quarkiverse/langchain4j/openai/runtime/OpenAiRecorder.java	(revision a8864444916d915b178eda65ee7665934e0ac7e8)
+++ b/model-providers/openai/openai-vanilla/runtime/src/main/java/io/quarkiverse/langchain4j/openai/runtime/OpenAiRecorder.java	(date 1737110791953)
@@ -189,8 +189,8 @@
             builder
                     .tlsConfigurationName(openAiConfig.tlsConfigurationName().orElse(null))
                     .configName(configName)
-                    .baseUrl(openAiConfig.baseUrl())
-                    .apiKey(apiKey)
+                    .baseUrl(embeddingModelConfig.baseUrl().orElse(openAiConfig.baseUrl()))
+                    .apiKey(embeddingModelConfig.apiKey().orElse(apiKey))
                     .timeout(openAiConfig.timeout().orElse(Duration.ofSeconds(10)))
                     .maxRetries(openAiConfig.maxRetries())
                     .logRequests(firstOrDefault(false, embeddingModelConfig.logRequests(), openAiConfig.logRequests()))

[geoand]

Thanks for looking into this!

To be honest, I am not terribly excited about the each component of LangChain4jOpenAiConfig having it's own base URL and api key configuration.
Can you please attach a sample project that you are using and that currently fails? I want to fully understand your use case and how you are trying to implement it.

Thanks