Add interleaved thinking and cache prompting

Author: ejstuart

model-providers/anthropic/runtime/src/main/java/io/quarkiverse/langchain4j/anthropic/QuarkusAnthropicClient.java

@@ -59,11 +59,13 @@ public class QuarkusAnthropicClient extends AnthropicClient {
     public static final String BETA = "tools-2024-04-04";
     private final String apiKey;
     private final String anthropicVersion;
+    private final String configuredBeta;
     private final AnthropicRestApi restApi;
 
     public QuarkusAnthropicClient(Builder builder) {
         this.apiKey = builder.apiKey;
         this.anthropicVersion = builder.version;
+        this.configuredBeta = builder.beta;
 
         try {
             var restApiBuilder = QuarkusRestClientBuilder.newBuilder().baseUri(new URI(builder.baseUrl))
@@ -104,12 +106,35 @@ private AnthropicRestApi.ApiMetadata createMetadata(AnthropicCreateMessageReques
         var builder = AnthropicRestApi.ApiMetadata.builder()
                 .apiKey(apiKey)
                 .anthropicVersion(anthropicVersion);
-        if (hasTools(request)) {
-            builder.beta(BETA);
+
+        String betaHeader = buildBetaHeaderForRequest(request);
+        if (betaHeader != null) {
+            builder.beta(betaHeader);
         }
+
         return builder.build();
     }
 
+    private String buildBetaHeaderForRequest(AnthropicCreateMessageRequest request) {
+        boolean toolsPresent = hasTools(request);
+
+        // If beta configured, use it (may need to combine with tools beta)
+        if (configuredBeta != null) {
+            // If tools present and tools beta not already in configured header
+            if (toolsPresent && !configuredBeta.contains("tools-")) {
+                return BETA + "," + configuredBeta;
+            }
+            return configuredBeta;
+        }
+
+        // Backward compatibility: default tools beta when tools present
+        if (toolsPresent) {
+            return BETA;
+        }
+
+        return null;
+    }
+
     private boolean hasTools(AnthropicCreateMessageRequest request) {
         if (!isNullOrEmpty(request.getTools())) {
             return true;

model-providers/anthropic/runtime/src/main/java/io/quarkiverse/langchain4j/anthropic/runtime/AnthropicRecorder.java

@@ -98,6 +98,15 @@ public Function<SyntheticCreationalContext<ChatModel>, ChatModel> chatModel(Stri
                 builder.sendThinking(thinkingConfig.sendThinking().get());
             }
 
+            // Pass caching configuration to builder
+            builder.cacheSystemMessages(chatModelConfig.cacheSystemMessages());
+            builder.cacheTools(chatModelConfig.cacheTools());
+
+            // Add beta header for interleaved thinking if enabled
+            if (thinkingConfig.interleaved().orElse(false)) {
+                builder.beta("interleaved-thinking-2025-05-14");
+            }
+
             return new Function<>() {
                 @Override
                 public ChatModel apply(SyntheticCreationalContext<ChatModel> context) {
@@ -172,6 +181,15 @@ public Function<SyntheticCreationalContext<StreamingChatModel>, StreamingChatMod
                 builder.sendThinking(thinkingConfig.sendThinking().get());
             }
 
+            // Pass caching configuration to builder
+            builder.cacheSystemMessages(chatModelConfig.cacheSystemMessages());
+            builder.cacheTools(chatModelConfig.cacheTools());
+
+            // Add beta header for interleaved thinking if enabled
+            if (thinkingConfig.interleaved().orElse(false)) {
+                builder.beta("interleaved-thinking-2025-05-14");
+            }
+
             return new Function<>() {
                 @Override
                 public StreamingChatModel apply(SyntheticCreationalContext<StreamingChatModel> context) {

model-providers/anthropic/runtime/src/main/java/io/quarkiverse/langchain4j/anthropic/runtime/config/ChatModelConfig.java

@@ -77,6 +77,22 @@ public interface ChatModelConfig {
     @ConfigDocDefault("false")
     Optional<Boolean> logResponses();
 
+    /**
+     * Cache system messages to reduce costs for repeated prompts.
+     * Requires minimum 1024 tokens (Claude Opus/Sonnet) or 2048-4096 tokens (Haiku).
+     * Supported models: Claude Opus 4.1, Sonnet 4.5, Haiku 4.5, and later models.
+     */
+    @WithDefault("false")
+    Boolean cacheSystemMessages();
+
+    /**
+     * Cache tool definitions to reduce costs.
+     * Requires minimum 1024 tokens (Claude Opus/Sonnet) or 2048-4096 tokens (Haiku).
+     * Supported models: Claude Opus 4.1, Sonnet 4.5, Haiku 4.5, and later models.
+     */
+    @WithDefault("false")
+    Boolean cacheTools();
+
     /**
      * Thinking related configuration
      */
@@ -106,5 +122,12 @@ interface ThinkingConfig {
          */
         @ConfigDocDefault("true")
         Optional<Boolean> sendThinking();
+
+        /**
+         * Enable interleaved thinking for Claude 4 models, allowing reasoning between tool calls.
+         * Requires Claude 4 model (e.g., claude-opus-4-20250514) and thinking.type: enabled.
+         */
+        @WithDefault("false")
+        Optional<Boolean> interleaved();
     }
 }