fix: create and use Gemini context caches when ContextCacheConfig is set

core/src/main/java/com/google/adk/agents/ContextCacheConfig.java

@@ -15,43 +15,62 @@
  */
 package com.google.adk.agents;
 
+import static com.google.common.base.Preconditions.checkArgument;
+import static com.google.common.base.Preconditions.checkNotNull;
+
 import java.time.Duration;
 
 /**
- * Configuration for context caching across all agents in an app.
- *
- * <p>This configuration enables and controls context caching behavior for all LLM agents in an app.
- * When this config is present on an app, context caching is enabled for all agents. When absent
- * (null), context caching is disabled.
- *
- * <p>Context caching can significantly reduce costs and improve response times by reusing
- * previously processed context across multiple requests.
+ * Configuration for context caching across all agents in an app; without it, nothing is cached.
+ * Gemini models cache the stable prefix of an agent's requests from its second request on, once
+ * that prefix reaches the model's minimum: 2048 tokens for Gemini 2.5, 4096 for Gemini 3. Other
+ * models ignore this config, and a cache carries over to the next invocation only with a session
+ * service that stores whole events, such as {@code InMemorySessionService}.
  *
- * @param maxInvocations Maximum number of invocations to reuse the same cache before refreshing it.
- *     Defaults to 10.
- * @param ttl Time-to-live for cache. Defaults to 1800 seconds (30 minutes).
- * @param minTokens Minimum estimated request tokens required to enable caching. This compares
- *     against the estimated total tokens of the request (system instruction + tools + contents).
- *     Context cache storage may have cost. Set higher to avoid caching small requests where
- *     overhead may exceed benefits. Defaults to 0.
+ * @param maxInvocations Maximum number of invocations to reuse the same cache before refreshing it,
+ *     from 1 to 100. Defaults to 10.
+ * @param ttl Time-to-live for cache; a whole number of seconds, at least one. Defaults to 1800
+ *     seconds (30 minutes).
+ * @param minTokens Minimum prompt token count of the agent's previous request needed to create a
+ *     cache; raise it to skip small requests, where cache storage can cost more than it saves. Must
+ *     not be negative. Defaults to 0.
  */
 public record ContextCacheConfig(int maxInvocations, Duration ttl, int minTokens) {
 
+  /**
+   * Validates the config as ADK Python does.
+   *
+   * @throws IllegalArgumentException if a value is out of range
+   * @throws NullPointerException if {@code ttl} is null
+   */
+  public ContextCacheConfig {
+    checkArgument(
+        maxInvocations >= 1 && maxInvocations <= 100,
+        "maxInvocations must be between 1 and 100, but was %s.",
+        maxInvocations);
+    checkNotNull(ttl, "ttl must not be null.");
+    checkArgument(
+        ttl.toSeconds() >= 1 && ttl.toNanosPart() == 0,
+        "ttl must be a whole number of seconds, at least one, but was %s.",
+        ttl);
+    checkArgument(minTokens >= 0, "minTokens must not be negative, but was %s.", minTokens);
+  }
+
   public ContextCacheConfig() {
-    this(10, Duration.ofSeconds(1800), 0);
+    this(10, Duration.ofMinutes(30), 0);
   }
 
-  /** Returns TTL as string format for cache creation. */
+  /** Returns the TTL in the {@code "<seconds>s"} form that the Gemini API uses. */
   public String getTtlString() {
-    return ttl.getSeconds() + "s";
+    return ttl.toSeconds() + "s";
   }
 
   @Override
   public String toString() {
     return "ContextCacheConfig(maxInvocations="
         + maxInvocations
         + ", ttl="
-        + ttl.getSeconds()
+        + ttl.toSeconds()
         + "s, minTokens="
         + minTokens
         + ")";

core/src/main/java/com/google/adk/events/Event.java

@@ -23,6 +23,7 @@
 import com.fasterxml.jackson.annotation.JsonProperty;
 import com.fasterxml.jackson.databind.annotation.JsonDeserialize;
 import com.google.adk.JsonBaseModel;
+import com.google.adk.models.CacheMetadata;
 import com.google.adk.platform.UuidProvider;
 import com.google.common.collect.ImmutableList;
 import com.google.common.collect.Iterables;
@@ -66,6 +67,7 @@ public class Event extends JsonBaseModel {
   private @Nullable String modelVersion;
   private @Nullable Transcription inputTranscription;
   private @Nullable Transcription outputTranscription;
+  private @Nullable CacheMetadata cacheMetadata;
 
   private long timestamp;
 
@@ -306,6 +308,19 @@ public void setOutputTranscription(@Nullable Transcription outputTranscription)
     this.outputTranscription = outputTranscription;
   }
 
+  /**
+   * Context cache state of the LLM response this event carries. The next request of the same agent
+   * reads it to reuse or refresh the cache.
+   */
+  @JsonProperty("cacheMetadata")
+  public Optional<CacheMetadata> cacheMetadata() {
+    return Optional.ofNullable(cacheMetadata);
+  }
+
+  public void setCacheMetadata(@Nullable CacheMetadata cacheMetadata) {
+    this.cacheMetadata = cacheMetadata;
+  }
+
   /** The timestamp of the event. */
   @JsonProperty("timestamp")
   public long timestamp() {
@@ -415,6 +430,7 @@ public static class Builder {
     private @Nullable String modelVersion;
     private @Nullable Transcription inputTranscription;
     private @Nullable Transcription outputTranscription;
+    private @Nullable CacheMetadata cacheMetadata;
     private @Nullable Long timestamp;
 
     @JsonCreator
@@ -592,6 +608,13 @@ public Builder outputTranscription(@Nullable Transcription value) {
       return this;
     }
 
+    @CanIgnoreReturnValue
+    @JsonProperty("cacheMetadata")
+    public Builder cacheMetadata(@Nullable CacheMetadata value) {
+      this.cacheMetadata = value;
+      return this;
+    }
+
     public Event build() {
       Event event = new Event();
       event.setId(id);
@@ -616,6 +639,7 @@ public Event build() {
           timestamp().orElseGet(() -> InstantSource.system().instant().toEpochMilli()));
       event.setInputTranscription(inputTranscription);
       event.setOutputTranscription(outputTranscription);
+      event.setCacheMetadata(cacheMetadata);
       return event;
     }
   }
@@ -653,6 +677,7 @@ public Builder toBuilder() {
             .modelVersion(this.modelVersion)
             .inputTranscription(this.inputTranscription)
             .outputTranscription(this.outputTranscription)
+            .cacheMetadata(this.cacheMetadata)
             .timestamp(this.timestamp);
     return builder;
   }
@@ -685,7 +710,8 @@ public boolean equals(Object obj) {
         && Objects.equals(customMetadata, other.customMetadata)
         && Objects.equals(modelVersion, other.modelVersion)
         && Objects.equals(inputTranscription, other.inputTranscription)
-        && Objects.equals(outputTranscription, other.outputTranscription);
+        && Objects.equals(outputTranscription, other.outputTranscription)
+        && Objects.equals(cacheMetadata, other.cacheMetadata);
   }
 
   @Override
@@ -716,6 +742,7 @@ public int hashCode() {
         modelVersion,
         inputTranscription,
         outputTranscription,
+        cacheMetadata,
         timestamp);
   }
 }

core/src/main/java/com/google/adk/flows/llmflows/BaseLlmFlow.java

@@ -885,7 +885,8 @@ private Event buildModelResponseEvent(
             .usageMetadata(llmResponse.usageMetadata().orElse(null))
             .modelVersion(llmResponse.modelVersion().orElse(null))
             .inputTranscription(llmResponse.inputTranscription().orElse(null))
-            .outputTranscription(llmResponse.outputTranscription().orElse(null));
+            .outputTranscription(llmResponse.outputTranscription().orElse(null))
+            .cacheMetadata(llmResponse.cacheMetadata().orElse(null));
 
     Event event = eventBuilder.build();
 

core/src/main/java/com/google/adk/flows/llmflows/ContextCacheRequestProcessor.java

@@ -0,0 +1,101 @@
+/*
+ * Copyright 2026 Google LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package com.google.adk.flows.llmflows;
+
+import com.google.adk.agents.ContextCacheConfig;
+import com.google.adk.agents.InvocationContext;
+import com.google.adk.events.Event;
+import com.google.adk.models.CacheMetadata;
+import com.google.adk.models.LlmRequest;
+import com.google.common.base.Strings;
+import com.google.common.collect.ImmutableList;
+import com.google.genai.types.GenerateContentResponseUsageMetadata;
+import io.reactivex.rxjava3.core.Single;
+import java.util.Optional;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+/**
+ * {@link RequestProcessor} that enables context caching when the app configures it. It puts the
+ * config, the agent's latest cache metadata and its previous prompt token count on the request; the
+ * model creates, reuses and deletes the caches.
+ */
+final class ContextCacheRequestProcessor implements RequestProcessor {
+
+  private static final Logger logger = LoggerFactory.getLogger(ContextCacheRequestProcessor.class);
+
+  @Override
+  public Single<RequestProcessingResult> processRequest(
+      InvocationContext context, LlmRequest request) {
+    Optional<ContextCacheConfig> cacheConfig = context.contextCacheConfig();
+    if (cacheConfig.isEmpty()) {
+      return Single.just(RequestProcessingResult.create(request, ImmutableList.of()));
+    }
+
+    String agentName = context.agent().name();
+    CacheMetadata cacheMetadata = null;
+    Integer previousTokenCount = null;
+    for (Event event : context.session().immutableEvents().reverse()) {
+      if (!agentName.equals(event.author())) {
+        continue;
+      }
+      if (cacheMetadata == null && event.cacheMetadata().isPresent()) {
+        cacheMetadata = countInvocation(event, context.invocationId());
+      }
+      if (previousTokenCount == null) {
+        previousTokenCount =
+            event
+                .usageMetadata()
+                .flatMap(GenerateContentResponseUsageMetadata::promptTokenCount)
+                .orElse(null);
+      }
+      if (cacheMetadata != null && previousTokenCount != null) {
+        break;
+      }
+    }
+    if (cacheMetadata != null) {
+      logger.debug("Found cache metadata for agent {}: {}", agentName, cacheMetadata);
+    }
+    if (previousTokenCount != null) {
+      logger.debug(
+          "Found previous prompt token count for agent {}: {}", agentName, previousTokenCount);
+    }
+    logger.debug("Context caching enabled for agent {}", agentName);
+
+    LlmRequest updatedRequest =
+        request.toBuilder()
+            .cacheConfig(cacheConfig.get())
+            .cacheMetadata(cacheMetadata)
+            .cacheableContentsTokenCount(previousTokenCount)
+            .build();
+    return Single.just(RequestProcessingResult.create(updatedRequest, ImmutableList.of()));
+  }
+
+  /**
+   * Returns the event's cache metadata, counting one more use when it names an active cache from an
+   * earlier invocation.
+   */
+  private static CacheMetadata countInvocation(Event event, String invocationId) {
+    CacheMetadata metadata = event.cacheMetadata().get();
+    if (Strings.isNullOrEmpty(event.invocationId())
+        || event.invocationId().equals(invocationId)
+        || metadata.cacheName().isEmpty()) {
+      return metadata;
+    }
+    return metadata.toBuilder().invocationsUsed(metadata.invocationsUsed().get() + 1).build();
+  }
+}

core/src/main/java/com/google/adk/flows/llmflows/SingleFlow.java

@@ -33,6 +33,7 @@ public class SingleFlow extends BaseLlmFlow {
           new Identity(),
           new Compaction(),
           new Contents(),
+          new ContextCacheRequestProcessor(),
           CodeExecution.requestProcessor);
 
   protected static final ImmutableList<ResponseProcessor> RESPONSE_PROCESSORS =