COH-32886 - Implement a provider for the remote Anthropic models

[git-p4: depot-paths = "//dev/coherence-ce/main/": change = 118720]

Author: aseovic

prj/coherence-dependencies/pom.xml

@@ -246,7 +246,7 @@
     <jettison.version>1.5.4</jettison.version>
     <jhdf.version>0.7.0-alpha</jhdf.version>
     <jna.version>5.14.0</jna.version>
-    <langchain4j.version>1.0.1</langchain4j.version>
+    <langchain4j.version>1.2.0</langchain4j.version>
     <log4j2.version>2.25.1</log4j2.version>
     <micrometer.version>1.13.4</micrometer.version>
     <!-- NOTE: this version should ideally be in sync' with that used by Helidon

prj/coherence-rag-parent/coherence-rag/src/main/java/com/oracle/coherence/rag/ModelProvider.java

@@ -66,7 +66,7 @@ public interface ModelProvider
      */
     default ScoringModel getScoringModel(String sName)
         {
-        return null;
+        throw new UnsupportedOperationException("Scoring model is not supported by " + getClass().getName());
         }
 
     /**

prj/coherence-rag-parent/coherence-rag/src/test/java/com/oracle/coherence/rag/ModelProviderTest.java

@@ -196,10 +196,7 @@ void shouldReturnNullForScoringModelByDefault()
             ModelProvider basicProvider = new BasicModelProvider();
 
             // Act
-            ScoringModel model = basicProvider.getScoringModel("test-scoring-model");
-
-            // Assert
-            assertThat(model, is(nullValue()));
+            assertThrows(UnsupportedOperationException.class, () -> basicProvider.getScoringModel("test-scoring-model"));
             }
 
         @Test

prj/coherence-rag-parent/models/coherence-rag-anthropic/README-Integration-Testing.md

@@ -0,0 +1,133 @@
+# OpenAI Model Provider Integration Testing
+
+This module includes comprehensive integration tests for the OpenAI model provider using WireMock in proxy/replay mode. This allows testing against both real OpenAI APIs and recorded responses.
+
+## Overview
+
+The integration tests can operate in two modes:
+
+1. **Real API Mode**: Tests run against the actual OpenAI API
+2. **WireMock Mode**: Tests run against recorded API responses
+
+The mode is automatically determined based on the `ANTHROPIC_API_KEY` environment variable:
+- Real API: When `ANTHROPIC_API_KEY` does NOT start with "test"
+- WireMock: When `ANTHROPIC_API_KEY` starts with "test" or is not set
+
+## Cost-Effective Testing
+
+The tests use the most cost-effective OpenAI models:
+- **Embedding Model**: `text-embedding-3-small` (cheaper than text-embedding-3-large)
+- **Chat Model**: `gpt-3.5-turbo` (significantly cheaper than GPT-4 models)
+
+## Setup and Usage
+
+### 1. Running Tests Against Recorded Responses (Default)
+
+This is the recommended approach for CI/CD and regular development:
+
+```bash
+# Set test API key (or don't set it at all)
+export ANTHROPIC_API_KEY=test-key
+
+# Run the integration tests
+mvn test -Dtest=AnthropicModelProviderIT
+```
+
+### 2. Capturing New API Responses
+
+When you need to update the recorded responses or test new functionality:
+
+#### Step 1: Start WireMock Proxy
+```bash
+# Set your real OpenAI API key
+export ANTHROPIC_API_KEY=your-real-api-key
+
+# Start WireMock proxy
+mvn exec:java@wiremock-proxy
+```
+
+#### Step 2: Run Tests Through Proxy
+In another terminal:
+```bash
+# Keep the real API key set
+export ANTHROPIC_API_KEY=your-real-api-key
+
+# Point tests to WireMock proxy
+export ANTHROPIC_BASE_URL=http://localhost:8089/v1
+
+# Run tests to capture responses
+mvn test -Dtest=AnthropicModelProviderIT -nsu
+```
+
+#### Step 3: Stop Proxy and Verify
+- Stop the WireMock proxy (Ctrl+C)
+- Check that new mapping files were created in `target/wiremock/mappings/`
+- Check that response files were created in `target/wiremock/__files/`
+- Copy mapping and response files as needed to `test/resources/wiremock`
+
+### 3. Running Tests Against Real API
+
+For validation or testing new functionality:
+
+```bash
+# Set your real OpenAI API key (must NOT start with "test")
+export ANTHROPIC_API_KEY=sk-your-real-api-key
+
+# Run tests against real API
+mvn test -Dtest=AnthropicModelProviderIntegrationTest
+```
+
+## Directory Structure
+
+```
+src/test/resources/wiremock/
+├── mappings/                                    # Request/response mappings
+│   ├── chat-completion-request.json             # Chat completion API mapping
+│   └── streaming-chat-completion-request.json   # Streaming Chat completion API mapping
+└── __files/                                     # Response body files
+    ├── chat-completion-response.json            # Example chat completion response
+    └── streaming-chat-completion-response.txt   # Example streaming chat completion response
+```
+
+## Test Coverage
+
+The integration tests cover:
+
+- **Chat Model**: Conversation handling with response validation
+- **Streaming Chat Model**: Real-time conversation with streaming responses
+
+## Troubleshooting
+
+### WireMock Proxy Not Starting
+- Ensure port 8089 is available
+- Check that Java is installed and accessible
+
+### Tests Failing in WireMock Mode
+- Ensure recorded responses exist in `src/test/resources/wiremock/`
+- Check that mapping files match the expected request patterns
+- Verify response files contain valid JSON or SSE responses (for streaming chat)
+
+### Tests Failing in Real API Mode
+- Verify your OpenAI API key is valid and has sufficient credits
+- Check internet connectivity
+- Ensure you're not hitting rate limits
+
+### API Key Issues
+- Real API keys must start with `sk-`
+- Test API keys should start with `test` to trigger WireMock mode
+- Never commit real API keys to version control
+
+## Security Notes
+
+- **Never commit real API keys** to version control
+- Use test keys for recorded responses
+- The example responses contain only synthetic data
+- Real API responses may contain sensitive information - review before committing
+
+## Cost Management
+
+When testing against the real API:
+- Tests use the cheapest available models
+- Each test run costs approximately $0.01-0.02 USD
+- Monitor your OpenAI usage dashboard
+- Consider using lower limits on your API key for testing