Merge pull request #349 from rjesh-git/feat/ant-thinking

Thinking feature for Anthropic and Bedrock LLM Agents

Author: Anthony Bernabeu

docs/src/content/docs/agents/built-in/anthropic-agent.mdx

@@ -339,7 +339,52 @@ agent = AnthropicAgent(AnthropicAgentOptions(
 </Tabs>
 <hr/>
 
-**9. Complete Example with All Options**
+**9. With Reasoning enabled**
+
+<Tabs syncKey="runtime">
+  <TabItem label="TypeScript" icon="seti:typescript" color="blue">
+  ```typescript
+  import { AnthropicAgent } from 'agent-squad';
+
+  const agent = new AnthropicAgent({
+    name: "Tech Agent",
+    description: "Specializes in technology areas including software development, hardware, AI, cybersecurity, blockchain, cloud computing, emerging tech innovations, and pricing/costs related to technology products and services.",
+    inferenceConfig: {
+      maxTokens: 2500,
+      temperature: 1, // 1 for thinking
+      topP: 0.96 // 0.95 or above
+    },
+    modelId: "claude-3-7-sonnet-20250219", // Claude 3.7 or above
+    thinking: {type: "enabled", budget_tokens: 1024},
+    streaming: true,
+    apiKey: 'your-anthropic-api-key',
+  });
+  ```
+  </TabItem>
+  <TabItem label="Python" icon="seti:python">
+  ```python
+
+  agent = AnthropicAgent(
+    AnthropicAgentOptions(
+        name="Tech Agent",
+        api_key='your-anthropic-api-key',
+        streaming=True,
+        description="Specializes in technology areas including software development, hardware, AI, \
+        cybersecurity, blockchain, cloud computing, emerging tech innovations, and pricing/costs \
+        related to technology products and services.",
+        model_id="claude-3-7-sonnet-20250219",
+        callbacks=LLMAgentCallbacks(),
+        inference_config={"maxTokens": 2500, "temperature": 1, "topP": 0.95}, # temperature set to 1 and topP 0.95 or above
+        thinking={"type": "enabled", "budget_tokens": 2000},
+    )
+  )
+  ```
+  </TabItem>
+</Tabs>
+
+<hr/>
+
+**10. Complete Example with All Options**
 
 <Tabs syncKey="runtime">
   <TabItem label="TypeScript" icon="seti:typescript" color="blue">

docs/src/content/docs/agents/built-in/bedrock-llm-agent.mdx

@@ -301,7 +301,48 @@ agent = BedrockLLMAgent(BedrockLLMAgentOptions(
 
 <hr/>
 
-**9. Complete Example with All Options**
+**9. With Thinking enabled**
+
+<Tabs syncKey="runtime">
+  <TabItem label="TypeScript" icon="seti:typescript" color="blue">
+  ```typescript
+  const agent = new BedrockLLMAgent({
+    name: "Tech Agent",
+    modelId: "us.anthropic.claude-3-7-sonnet-20250219-v1:0",
+    description:"Specializes in technology areas including software development, hardware, AI, cybersecurity, \
+    blockchain, cloud computing, emerging tech innovations, and pricing/costs related to technology products and services.",
+    inferenceConfig: {
+      maxTokens: 2500,
+      temperature: 1, // 1 for thinking and unset topP
+    },
+    additional_model_request_fields: {
+      thinking: {type: "enabled", budget_tokens: 1024},
+    },
+    streaming: true,
+  });
+  ```
+  </TabItem>
+  <TabItem label="Python" icon="seti:python">
+  ```python
+  agent = BedrockLLMAgent(
+    BedrockLLMAgentOptions(
+        name="Tech Agent",
+        streaming=False,
+        description="Specializes in technology areas including software development, hardware, AI, \
+        cybersecurity, blockchain, cloud computing, emerging tech innovations, and pricing/costs \
+        related to technology products and services.",
+        model_id="us.anthropic.claude-3-7-sonnet-20250219-v1:0",
+        callbacks=LLMAgentCallbacks(),
+        inference_config={"maxTokens": 2500, "temperature": 1},
+        additional_model_request_fields={"thinking": {"type": "enabled", "budget_tokens": 2000}},
+    )
+  )
+  ```
+  </TabItem>
+</Tabs>
+<hr/>
+
+**10. Complete Example with All Options**
 
 <Tabs syncKey="runtime">
   <TabItem label="TypeScript" icon="seti:typescript" color="blue">
@@ -444,7 +485,7 @@ agent = BedrockLLMAgent(BedrockLLMAgentOptions(
 
 The `BedrockLLMAgent` provides multiple ways to set custom prompts. You can set them either during initialization or after the agent is created, and you can use prompts with or without variables.
 
-**10. Setting Custom Prompt After Initialization (Without Variables)**
+**11. Setting Custom Prompt After Initialization (Without Variables)**
 
 <Tabs syncKey="runtime">
   <TabItem label="TypeScript" icon="seti:typescript" color="blue">
@@ -495,7 +536,7 @@ When providing business advice:
 
 <hr/>
 
-**11. Setting Custom Prompt After Initialization (With Variables)**
+**12. Setting Custom Prompt After Initialization (With Variables)**
 
 <Tabs syncKey="runtime">
   <TabItem label="TypeScript" icon="seti:typescript" color="blue">
@@ -616,6 +657,7 @@ Choose the approach that best fits your needs:
 | `streaming` | Enables streaming responses for real-time output | Optional |
 | `inferenceConfig` | Fine-tunes the model's output characteristics | Optional |
 | `guardrailConfig` | Applies predefined guardrails to the model's responses | Optional |
+| `reasoningConfig` | Enables thinking and configuration for budget_tokens | Optional | 
 | `retriever` | Integrates a retrieval system for enhanced context | Optional |
 | `toolConfig` | Defines tools the agent can use and how to handle their responses | Optional |
 | `customSystemPrompt` | Defines the agent's system prompt and behavior, with optional variables for dynamic content | Optional |
@@ -632,6 +674,7 @@ Choose the approach that best fits your needs:
 | `streaming` | Enables streaming responses for real-time output | Optional |
 | `inference_config` | Fine-tunes the model's output characteristics | Optional |
 | `guardrail_config` | Applies predefined guardrails to the model's responses | Optional |
+| `additional_model_request_fields` | Additional fields to send to the model, including thinking capability | Optional | 
 | `retriever` | Integrates a retrieval system for enhanced context | Optional |
 | `tool_config` | Defines tools the agent can use and how to handle their responses | Optional |
 | `custom_system_prompt` | Defines the agent's system prompt and behavior, with optional variables for dynamic content | Optional |

examples/local-demo/local-orchestrator.ts

@@ -51,17 +51,25 @@ function createOrchestrator(): AgentSquad {
   // Add weahter agent with tool
   const weatherAgent = new BedrockLLMAgent({
     name: "Weather Agent",
+    modelId:"us.anthropic.claude-3-7-sonnet-20250219-v1:0",
     description:
       "Specialized agent for giving weather condition from a city.",
-    streaming: true,
-    inferenceConfig: {
-      temperature: 0.1,
-    },
+    streaming: false,
     toolConfig: {
       tool: weatherToolDescription,
       useToolHandler: weatherToolHanlder,
       toolMaxRecursions: 5,
-    }
+    },
+    inferenceConfig: {
+        temperature: 1.0,
+        maxTokens:4096,
+      },
+      reasoningConfig:{
+        thinking:{
+          type:'enabled',
+          budget_tokens: 4000,
+        }
+      }
   });
   weatherAgent.setSystemPrompt(WEATHER_PROMPT);
   orchestrator.addAgent(weatherAgent);
@@ -174,6 +182,10 @@ async function runLocalConversation(): Promise<void> {
           for await (const chunk of response.output) {
             if (typeof chunk === "string") {
               process.stdout.write(chunk);
+            }
+            else if (typeof chunk === "object" && chunk.hasOwnProperty("thinking")) {
+              // Print thinking content in cyan color
+              process.stdout.write('\x1b[36m' + chunk.content + '\x1b[0m');
             } else {
               Logger.logger.error("Received unexpected chunk type:", typeof chunk);
             }

examples/python-demo/main-stream.py

@@ -15,8 +15,10 @@
                         AgentStreamResponse,
                         AgentCallbacks)
 from agent_squad.types import ConversationMessage, ParticipantRole
-from agent_squad.utils import AgentTool, AgentTools, AgentToolCallbacks
+from agent_squad.utils import AgentToolCallbacks
+from dotenv import load_dotenv
 
+load_dotenv()
 class LLMAgentCallbacks(AgentCallbacks):
     async def on_agent_start(
         self,
@@ -105,10 +107,23 @@ async def handle_request(_orchestrator: AgentSquad, _user_input:str, _user_id:st
         async for chunk in response.output:
             if isinstance(chunk, AgentStreamResponse):
                 if response.streaming:
-                    print(chunk.text, end='', flush=True)
+                    if (chunk.thinking):
+                        print(f"\033[34m{chunk.thinking}\033[0m", end='', flush=True)
+                    elif (chunk.text):
+                        print(chunk.text, end='', flush=True)
     else:
         if isinstance(response.output, ConversationMessage):
             print(response.output.content[0]['text'])
+
+            # Safely extract thinking content from response
+            thinking_content = None
+            for content_item in response.output.content:
+                if isinstance(content_item, dict) and 'reasoningContent' in content_item:
+                    thinking_content = content_item['reasoningContent']
+                    break
+
+            if thinking_content:
+                print(f"\nThinking: {thinking_content}")
         elif isinstance(response.output, str):
             print(response.output)
         else:
@@ -159,26 +174,50 @@ def custom_output_payload_decoder(response: dict[str, Any]) -> Any:
     ))
     orchestrator.add_agent(tech_agent)
 
+
     # Add some agents
     tech_agent = BedrockLLMAgent(BedrockLLMAgentOptions(
         name="Health Agent",
-        streaming=False,
+        streaming=True,
+        inference_config={
+            "maxTokens": 4096,
+            "temperature":1.0
+        },
         description="Specializes in health and well being.",
-        model_id="anthropic.claude-3-sonnet-20240229-v1:0",
+        model_id="us.anthropic.claude-3-7-sonnet-20250219-v1:0",
+        additional_model_request_fields={
+            "thinking": {
+                "type": "enabled",
+                "budget_tokens": 4000
+            }
+        }
     ))
     orchestrator.add_agent(tech_agent)
 
     # Add a Anthropic weather agent with a tool in anthropic's tool format
     # weather_agent = AnthropicAgent(AnthropicAgentOptions(
-    #     api_key='api-key',
+    #     api_key=os.getenv('ANTHROPIC_API_KEY', None),
     #     name="Weather Agent",
-    #     streaming=False,
+    #     streaming=True,
+    #     model_id="claude-3-7-sonnet-20250219",
     #     description="Specialized agent for giving weather condition from a city.",
     #     tool_config={
     #         'tool': [tool.to_claude_format() for tool in weather_tool.weather_tools.tools],
     #         'toolMaxRecursions': 5,
     #         'useToolHandler': weather_tool.anthropic_weather_tool_handler
     #     },
+    #     inference_config={
+    #         "maxTokens": 4096,
+    #         "temperature":1.0,
+    #         "topP":1.0
+    #     }
+    #     ,
+    #     additional_model_request_fields = {
+    #         "thinking": {
+    #             "type": "enabled",
+    #             "budget_tokens": 4000
+    #         }
+    #     },
     #     callbacks=LLMAgentCallbacks()
     # ))
 
@@ -210,13 +249,24 @@ def custom_output_payload_decoder(response: dict[str, Any]) -> Any:
     # Add a Bedrock weather agent with custom handler and bedrock's tool format
     weather_agent = BedrockLLMAgent(BedrockLLMAgentOptions(
         name="Weather Agent",
-        streaming=False,
+        streaming=True,
+        model_id="us.anthropic.claude-3-7-sonnet-20250219-v1:0",
         description="Specialized agent for giving weather condition from a city.",
         tool_config={
             'tool': [tool.to_bedrock_format() for tool in weather_tool.weather_tools.tools],
             'toolMaxRecursions': 5,
             'useToolHandler': weather_tool.bedrock_weather_tool_handler
-        }
+        },
+        additional_model_request_fields={
+            "thinking": {
+                "type": "enabled",
+                "budget_tokens": 4000
+            }
+        },
+        inference_config={
+            "maxTokens": 4096,
+            "temperature":1.0
+        },
     ))
 
 
@@ -236,5 +286,6 @@ def custom_output_payload_decoder(response: dict[str, Any]) -> Any:
             print("Exiting the program. Goodbye!")
             sys.exit()
 
-        # Run the async function
-        asyncio.run(handle_request(orchestrator, user_input, USER_ID, SESSION_ID))
+        if user_input != '':
+            # Run the async function
+            asyncio.run(handle_request(orchestrator, user_input, USER_ID, SESSION_ID))