tschuehly · Oct 20, 2026
diff --git a/examples-common/src/main/kotlin/com/embabel/example/common/Effort.kt b/examples-common/src/main/kotlin/com/embabel/example/common/Effort.kt
@@ -0,0 +1,36 @@
+/*
+ * Copyright 2024-2026 Embabel Pty Ltd.
+ *
+ * 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.embabel.example.common
+
+import com.embabel.common.ai.model.Thinking
+
+/**
+ * Reasoning effort for the examples' calls.
+ *
+ * Claude 4.7 and later models think adaptively, and Embabel maps a thinking token budget to their
+ * effort level: under 1024 is low, under 4096 is medium. On those models the budget is not a token cap.
+ * Older Claude models receive it as a literal thinking budget instead.
+ *
+ * Effort is set per call because the configured default model carries no effort of its own.
+ */
+object Effort {
+
+    @JvmField
+    val LOW: Thinking = Thinking.withTokenBudget(512)
+
+    @JvmField
+    val MEDIUM: Thinking = Thinking.withTokenBudget(1024)
+}
diff --git a/examples-java/src/main/java/com/embabel/example/factchecker/FactChecker.java b/examples-java/src/main/java/com/embabel/example/factchecker/FactChecker.java
@@ -30,6 +30,7 @@
 import com.embabel.agent.domain.library.InternetResource;
 import com.embabel.common.ai.model.LlmOptions;
 import com.embabel.common.ai.prompt.PromptContributor;
+import com.embabel.example.common.Effort;
 import com.fasterxml.jackson.annotation.JsonPropertyDescription;
 import com.google.common.base.Supplier;
 import org.slf4j.LoggerFactory;
@@ -75,12 +76,13 @@ record FactCheckerProperties(
         List<String> trustedSources,
         List<String> untrustedSources,
         List<String> models,
+        List<String> extractionModels,
         String deduplicationModel,
         int maxConcurrency
 ) {
 
     LlmOptions deduplicationLlm() {
-        return LlmOptions.withModel(deduplicationModel);
+        return LlmOptions.withModel(deduplicationModel).withThinking(Effort.LOW);
     }
 
     /**
@@ -172,7 +174,7 @@ private Stream<Supplier<FactCheck>> factCheckWithSingleLlm(
     }
 
     private Stream<Supplier<DistinctFactualAssertions>> factualAssertionExtractors(UserInput userInput, ActionContext actionContext) {
-        return properties.models().stream()
+        return properties.extractionModels().stream()
                 .map(model -> () -> extractFactualAssertionsWithModel(userInput, actionContext, model));
     }
 
@@ -181,7 +183,7 @@ private Stream<Supplier<DistinctFactualAssertions>> factualAssertionExtractors(U
      */
     private DistinctFactualAssertions extractFactualAssertionsWithModel(UserInput userInput, ActionContext context, String model) {
         return context.ai()
-                .withLlm(model)
+                .withLlm(LlmOptions.withModel(model).withThinking(Effort.LOW))
                 .createObject(
                         """
                                 Extract distinct factual assertions from the following text.

diff --git a/examples-java/src/main/java/com/embabel/example/horoscope/NewsFeedService.java b/examples-java/src/main/java/com/embabel/example/horoscope/NewsFeedService.java
@@ -21,6 +21,7 @@
 import com.embabel.agent.api.tool.ToolCallContext;
 import com.embabel.agent.spi.ToolGroupResolver;
 import com.embabel.common.ai.model.LlmOptions;
+import com.embabel.example.common.Effort;
 import tools.jackson.databind.json.JsonMapper;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
@@ -99,7 +100,7 @@ public Result call(String input, ToolCallContext context) {
                 return result;
             }
         };
-        ai.withLlm(LlmOptions.withModel(defaultLlm).withTimeout(Duration.ofMinutes(3)))
+        ai.withLlm(LlmOptions.withModel(defaultLlm).withThinking(Effort.LOW).withTimeout(Duration.ofMinutes(3)))
                 .withId("fetch_news_feed")
                 .withTool(directFetch)
                 .generateText("Call %s once with url=%s%s. Do not summarize or search."

diff --git a/examples-java/src/main/java/com/embabel/example/horoscope/StarNewsFinder.java b/examples-java/src/main/java/com/embabel/example/horoscope/StarNewsFinder.java
@@ -25,6 +25,7 @@
 import com.embabel.agent.domain.library.PersonImpl;
 import com.embabel.agent.domain.library.RelevantNewsStories;
 import com.embabel.common.ai.model.LlmOptions;
+import com.embabel.example.common.Effort;
 import org.springframework.beans.factory.annotation.Value;
 
 import java.time.Duration;
@@ -112,7 +113,7 @@ public RelevantNewsStories findNewsStories(
                 .mapToObj(i -> i + ": " + stories.get(i).getTitle() + " — " + stories.get(i).getSummary())
                 .collect(Collectors.joining("\n"));
         var selection = ai
-                .withLlm(LlmOptions.withModel(defaultLlm).withTimeout(Duration.ofMinutes(3)))
+                .withLlm(LlmOptions.withModel(defaultLlm).withThinking(Effort.LOW).withTimeout(Duration.ofMinutes(3)))
                 .withId("select_news_stories")
                 .createObject("""
                         Select between 1 and %d news stories relevant to %s's horoscope.

diff --git a/examples-java/src/main/java/com/embabel/example/injection/InjectedComponent.java b/examples-java/src/main/java/com/embabel/example/injection/InjectedComponent.java
@@ -17,6 +17,7 @@
 
 import com.embabel.agent.api.common.Ai;
 import com.embabel.common.ai.model.LlmOptions;
+import com.embabel.example.common.Effort;
 import org.springframework.stereotype.Component;
 
 
@@ -47,7 +48,7 @@ public record Joke(String leadup, String punchline) {
 
     public String tellJokeAbout(String topic) {
         return ai
-                .withDefaultLlm()
+                .withLlm(LlmOptions.withDefaultLlm().withThinking(Effort.LOW))
                 .generateText("""
                         Tell me a short joke. Reply with the joke only.
                         %s
@@ -57,7 +58,7 @@ public String tellJokeAbout(String topic) {
 
     public Joke createJokeObjectAbout(String topic1, String topic2, String voice) {
         return ai
-                .withLlm(LlmOptions.withDefaultLlm().withTemperature(.8))
+                .withLlm(LlmOptions.withDefaultLlm().withThinking(Effort.LOW))
                 .createObject("""
                                 Tell me a joke about the two topics below, in the voice below.
                                 The joke should have a leadup and a punchline.

diff --git a/examples-java/src/main/java/com/embabel/example/injection/travel/ActivitySummarizer.java b/examples-java/src/main/java/com/embabel/example/injection/travel/ActivitySummarizer.java
@@ -16,6 +16,8 @@
 package com.embabel.example.injection.travel;
 
 import com.embabel.agent.api.common.Ai;
+import com.embabel.common.ai.model.LlmOptions;
+import com.embabel.example.common.Effort;
 import org.springframework.boot.context.properties.ConfigurationProperties;
 import org.springframework.boot.context.properties.bind.DefaultValue;
 import org.springframework.lang.Nullable;
@@ -64,7 +66,7 @@ public TravelerActivityReport summarizeActivity(Long customerId) {
         // We can rely on the record's generated toString in this
         // case to expose the right data, as it will include all fields
         var report = ai
-                .withDefaultLlm()
+                .withLlm(LlmOptions.withDefaultLlm().withThinking(Effort.LOW))
                 .withToolObject(activity)
                 .generateText("""
                         Your job is to help staff of Antechinus Travel understand their customers.

diff --git a/examples-java/src/main/java/com/embabel/example/repeatuntil/GoatWriterConfig.java b/examples-java/src/main/java/com/embabel/example/repeatuntil/GoatWriterConfig.java
@@ -19,6 +19,8 @@
 import com.embabel.agent.api.common.workflow.loop.TextFeedback;
 import com.embabel.agent.api.models.OpenAiModels;
 import com.embabel.agent.core.Agent;
+import com.embabel.common.ai.model.LlmOptions;
+import com.embabel.example.common.Effort;
 import org.springframework.context.annotation.Bean;
 import org.springframework.context.annotation.Configuration;
 
@@ -50,7 +52,7 @@ public Agent goatWriter() {
                                     Goat.class);
                 })
                 .withEvaluator(context ->
-                        context.ai().withDefaultLlm().createObject(
+                        context.ai().withLlm(LlmOptions.withDefaultLlm().withThinking(Effort.LOW)).createObject(
                                 """
                                         Evaluate the goat description for creativity and detail.
                                         The goat must have a name.

diff --git a/examples-java/src/main/java/com/embabel/example/secured/MarketIntelligenceAgent.java b/examples-java/src/main/java/com/embabel/example/secured/MarketIntelligenceAgent.java
@@ -23,6 +23,8 @@
 import com.embabel.agent.core.CoreToolGroups;
 import com.embabel.agent.domain.io.UserInput;
 import com.embabel.agent.mcpserver.security.SecureAgentTool;
+import com.embabel.common.ai.model.LlmOptions;
+import com.embabel.example.common.Effort;
 
 import java.util.stream.Collectors;
 
@@ -66,7 +68,7 @@ public class MarketIntelligenceAgent {
     @Action
     public AnalysisSubject parseSubject(UserInput userInput, OperationContext context) {
         return context.ai()
-                .withDefaultLlm()
+                .withLlm(LlmOptions.withDefaultLlm().withThinking(Effort.LOW))
                 .createObject(
                         """
                         Extract the subject of market analysis from the following user input.
@@ -93,7 +95,7 @@ public AnalysisSubject parseSubject(UserInput userInput, OperationContext contex
     @Action
     public String gatherIntelligence(AnalysisSubject subject, OperationContext context) {
         return context.ai()
-                .withDefaultLlm()
+                .withLlm(LlmOptions.withDefaultLlm().withThinking(Effort.LOW))
                 .withToolGroup(CoreToolGroups.WEB)
                 .createObject(
                         """
@@ -147,7 +149,7 @@ public MarketIntelligenceReport synthesiseReport(
             OperationContext context
     ) {
         SwotEntryList swotList = context.ai()
-                .withDefaultLlm()
+                .withLlm(LlmOptions.withDefaultLlm().withThinking(Effort.LOW))
                 .createObject(
                         """
                         Based on the following market intelligence, produce a SWOT analysis for %s.
@@ -161,7 +163,7 @@ Each entry must have a category (Strength, Weakness, Opportunity, or Threat)
                 );
 
         CompetitorInsightList competitorList = context.ai()
-                .withDefaultLlm()
+                .withLlm(LlmOptions.withDefaultLlm().withThinking(Effort.LOW))
                 .createObject(
                         """
                         Based on the following market intelligence, identify the top 4 competitors
@@ -175,7 +177,7 @@ Each entry must have a category (Strength, Weakness, Opportunity, or Threat)
                 );
 
         KeyTrendList trendList = context.ai()
-                .withDefaultLlm()
+                .withLlm(LlmOptions.withDefaultLlm().withThinking(Effort.LOW))
                 .createObject(
                         """
                         Based on the following market intelligence, extract the 5 most significant
@@ -189,7 +191,7 @@ Each entry must have a category (Strength, Weakness, Opportunity, or Threat)
                 );
 
         String summary = context.ai()
-                .withDefaultLlm()
+                .withLlm(LlmOptions.withDefaultLlm().withThinking(Effort.LOW))
                 .createObject(
                         """
                         Write a 4-sentence executive summary for a market intelligence report on

diff --git a/examples-java/src/main/java/com/embabel/example/secured/NewsDigestAgent.java b/examples-java/src/main/java/com/embabel/example/secured/NewsDigestAgent.java
@@ -23,6 +23,8 @@
 import com.embabel.agent.core.CoreToolGroups;
 import com.embabel.agent.domain.io.UserInput;
 import com.embabel.agent.mcpserver.security.SecureAgentTool;
+import com.embabel.common.ai.model.LlmOptions;
+import com.embabel.example.common.Effort;
 
 import java.util.stream.Collectors;
 
@@ -61,7 +63,7 @@ public class NewsDigestAgent {
     @Action
     public NewsTopic extractTopic(UserInput userInput, OperationContext context) {
         return context.ai()
-                .withDefaultLlm()
+                .withLlm(LlmOptions.withDefaultLlm().withThinking(Effort.LOW))
                 .createObject(
                         """
                         Extract a research topic from the following user input.
@@ -106,7 +108,7 @@ public NewsDigest produceDigest(NewsTopic topic, OperationContext context) {
                 : "";
 
         DigestItemList digestItemList = context.ai()
-                .withDefaultLlm()
+                .withLlm(LlmOptions.withDefaultLlm().withThinking(Effort.LOW))
                 .withToolGroup(CoreToolGroups.WEB)
                 .createObject(
                         """
@@ -118,7 +120,7 @@ public NewsDigest produceDigest(NewsTopic topic, OperationContext context) {
                 );
 
         String narrative = context.ai()
-                .withDefaultLlm()
+                .withLlm(LlmOptions.withDefaultLlm().withThinking(Effort.LOW))
                 .createObject(
                         """
                         Write a 3-sentence editorial summary of these news items about %s.

diff --git a/examples-java/src/main/resources/application.properties b/examples-java/src/main/resources/application.properties
@@ -1,7 +1,8 @@
 embabel.agent.application.name=embabel-examples
 
-embabel.models.default-llm=gpt-4.1-mini
-embabel.models.llms.cheapest=gpt-6-luna
+# Calls on the default model and the cheapest role ask for low effort in code (see Effort)
+embabel.models.default-llm=claude-haiku-5-5
+embabel.models.llms.cheapest=claude-haiku-5-5
 
 embabel.models.llms.best=gpt-6-astra
 
@@ -20,10 +21,12 @@ embabel.agent-platform.ranking.llm=gpt-4.1-mini
 
 
 embabel.fact-checker.models=gpt-4.1-mini,gpt-6-luna,gpt-6.1-sol
+# Smaller models extract opinions as facts
+embabel.fact-checker.extraction-models=claude-sonnet-5-5
 embabel.fact-checker.max-concurrency=8
 
-# Use a good LLM for this
-embabel.fact-checker.deduplication-model=gpt-4.1
+# Use a good LLM for this: smaller models merge an assertion with its negation
+embabel.fact-checker.deduplication-model=claude-sonnet-5-5
 embabel.fact-checker.reasoning-word-count=50
 embabel.fact-checker.trusted-sources=Wikipedia,Wikidata,Britannica,BBC,Reuters,NY Times,ABC Australia
 embabel.fact-checker.untrusted-sources=reddit,4chan,twitter

diff --git a/examples-java/src/test/java/com/embabel/example/injection/InjectedComponentTest.java b/examples-java/src/test/java/com/embabel/example/injection/InjectedComponentTest.java
@@ -16,9 +16,11 @@
 package com.embabel.example.injection;
 
 import com.embabel.agent.test.unit.FakeOperationContext;
+import com.embabel.example.common.Effort;
 import org.junit.jupiter.api.Test;
 
 import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertSame;
 import static org.junit.jupiter.api.Assertions.assertTrue;
 
 /**
@@ -40,6 +42,7 @@ void tellJokeAboutReturnsTheJoke() {
 
         assertEquals(terribleJoke, joke);
         assertTrue(promptOf(context).contains("Topic: frogs"));
+        assertSame(Effort.LOW, context.getLlmInvocations().getFirst().getInteraction().getLlm().getThinking());
     }
 
     @Test

diff --git a/examples-java/src/test/java/com/embabel/example/injection/travel/ActivitySummarizerTest.java b/examples-java/src/test/java/com/embabel/example/injection/travel/ActivitySummarizerTest.java
@@ -18,14 +18,18 @@
 import com.embabel.agent.api.common.Ai;
 import com.embabel.agent.api.common.PromptRunner;
 import com.embabel.agent.test.unit.FakeOperationContext;
+import com.embabel.common.ai.model.LlmOptions;
+import com.embabel.example.common.Effort;
 import org.junit.jupiter.api.Test;
 import org.mockito.Mockito;
 
 import java.time.Instant;
 import java.util.List;
 
 import static org.junit.jupiter.api.Assertions.*;
+import static org.mockito.ArgumentMatchers.any;
 import static org.mockito.ArgumentMatchers.anyString;
+import static org.mockito.ArgumentMatchers.argThat;
 import static org.mockito.Mockito.when;
 
 class ActivitySummarizerTest {
@@ -55,7 +59,7 @@ void testSummarizeActivityLlmThrowsException() {
 
         var travellerActivity = createSampleTravellerActivity();
         when(mockReportingService.report(1L)).thenReturn(travellerActivity);
-        when(mockAi.withDefaultLlm()).thenReturn(mockPromptRunner);
+        when(mockAi.withLlm(any(LlmOptions.class))).thenReturn(mockPromptRunner);
         when(mockPromptRunner.withToolObject(travellerActivity)).thenReturn(mockPromptRunner);
         when(mockPromptRunner.generateText(anyString())).thenThrow(new RuntimeException("LLM service unavailable"));
 
@@ -64,7 +68,7 @@ void testSummarizeActivityLlmThrowsException() {
         assertThrows(RuntimeException.class, () -> activitySummarizer.summarizeActivity(1L));
 
         Mockito.verify(mockReportingService).report(1L);
-        Mockito.verify(mockAi).withDefaultLlm();
+        Mockito.verify(mockAi).withLlm(argThat((LlmOptions llm) -> llm.getThinking() == Effort.LOW));
         Mockito.verify(mockPromptRunner).withToolObject(travellerActivity);
     }
 
@@ -79,7 +83,7 @@ void testSummarizeActivityValidReturn() {
         var expectedSummary = "John Doe is a frequent traveler who has taken 2 trips to Paris and Tokyo, spending $3000 total. As a high spender, he enjoys premium destinations.";
 
         when(mockReportingService.report(1L)).thenReturn(travellerActivity);
-        when(mockAi.withDefaultLlm()).thenReturn(mockPromptRunner);
+        when(mockAi.withLlm(any(LlmOptions.class))).thenReturn(mockPromptRunner);
         when(mockPromptRunner.withToolObject(travellerActivity)).thenReturn(mockPromptRunner);
         when(mockPromptRunner.generateText(anyString())).thenReturn(expectedSummary);
 
@@ -90,7 +94,7 @@ void testSummarizeActivityValidReturn() {
         assertEquals(travellerActivity, result.activity());
 
         Mockito.verify(mockReportingService).report(1L);
-        Mockito.verify(mockAi).withDefaultLlm();
+        Mockito.verify(mockAi).withLlm(argThat((LlmOptions llm) -> llm.getThinking() == Effort.LOW));
         Mockito.verify(mockPromptRunner).withToolObject(travellerActivity);
         Mockito.verify(mockPromptRunner).generateText(anyString());
     }