[EXPERIMENT] claude thinking model support

examples/foundational/14a-function-calling-anthropic.py

@@ -82,6 +82,13 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
     llm = AnthropicLLMService(
         api_key=os.getenv("ANTHROPIC_API_KEY"),
         model="claude-3-7-sonnet-latest",
+        wait_for_all=True,
+        params=AnthropicLLMService.InputParams(
+            max_tokens=16000,
+            extra={
+                "thinking": {"type": "enabled", "budget_tokens": 10000},
+            },
+        ),
     )
     llm.register_function("get_weather", get_weather)
     llm.register_function("get_restaurant_recommendation", fetch_restaurant_recommendation)

src/pipecat/frames/frames.py

@@ -563,6 +563,33 @@ class LLMContextFrame(Frame):
     context: "LLMContext"
 
 
+@dataclass
+class LLMThinkingTextFrame(DataFrame):
+    """Reasoning frame generated by LLM services."""
+
+    thinking: str
+
+    def __post_init__(self):
+        super().__post_init__()
+        # LLM services send text frames with all necessary spaces included
+        self.includes_inter_frame_spaces = True
+
+    def __str__(self):
+        pts = format_pts(self.pts)
+        return f"{self.name}(pts: {pts}, thinking: {self.thinking})"
+
+
+@dataclass
+class LLMThinkingSignatureFrame(DataFrame):
+    """Reasoning signature frame generated by LLM services."""
+
+    signature: str
+
+    def __str__(self):
+        pts = format_pts(self.pts)
+        return f"{self.name}(pts: {pts}, signature: {self.signature})"
+
+
 @dataclass
 class LLMMessagesFrame(DataFrame):
     """Frame containing LLM messages for chat completion.

src/pipecat/processors/aggregators/llm_response_universal.py

@@ -47,6 +47,8 @@
     LLMRunFrame,
     LLMSetToolChoiceFrame,
     LLMSetToolsFrame,
+    LLMThinkingSignatureFrame,
+    LLMThinkingTextFrame,
     SpeechControlParamsFrame,
     StartFrame,
     TextFrame,
@@ -591,6 +593,7 @@ def __init__(
         self._started = 0
         self._function_calls_in_progress: Dict[str, Optional[FunctionCallInProgressFrame]] = {}
         self._context_updated_tasks: Set[asyncio.Task] = set()
+        self._thinking: List[TextPartForConcatenation] = []
 
     @property
     def has_function_calls_in_progress(self) -> bool:
@@ -601,6 +604,11 @@ def has_function_calls_in_progress(self) -> bool:
         """
         return bool(self._function_calls_in_progress)
 
+    async def reset(self):
+        """Reset the aggregation state."""
+        await super().reset()
+        self._thinking = []
+
     async def process_frame(self, frame: Frame, direction: FrameDirection):
         """Process frames for assistant response aggregation and function call management.
 
@@ -619,6 +627,10 @@ async def process_frame(self, frame: Frame, direction: FrameDirection):
             await self._handle_llm_end(frame)
         elif isinstance(frame, TextFrame):
             await self._handle_text(frame)
+        elif isinstance(frame, LLMThinkingTextFrame):
+            await self._handle_thinking(frame)
+        elif isinstance(frame, LLMThinkingSignatureFrame):
+            await self._handle_thinking_signature(frame)
         elif isinstance(frame, LLMRunFrame):
             await self._handle_llm_run(frame)
         elif isinstance(frame, LLMMessagesAppendFrame):
@@ -663,6 +675,14 @@ async def push_aggregation(self):
         timestamp_frame = LLMContextAssistantTimestampFrame(timestamp=time_now_iso8601())
         await self.push_frame(timestamp_frame)
 
+    def thinking_string(self) -> str:
+        """Get the current thinking as a string.
+
+        Returns:
+            The concatenated thinking string.
+        """
+        return concatenate_aggregated_text(self._thinking)
+
     async def _handle_llm_run(self, frame: LLMRunFrame):
         await self.push_context_frame(FrameDirection.UPSTREAM)
 
@@ -824,6 +844,35 @@ async def _handle_text(self, frame: TextFrame):
             )
         )
 
+    async def _handle_thinking(self, frame: LLMThinkingTextFrame):
+        if not self._started:
+            return
+
+        # Make sure we really have text (spaces count, too!)
+        if len(frame.thinking) == 0:
+            return
+
+        self._thinking.append(
+            TextPartForConcatenation(
+                frame.thinking, includes_inter_part_spaces=frame.includes_inter_frame_spaces
+            )
+        )
+
+    async def _handle_thinking_signature(self, frame: LLMThinkingSignatureFrame):
+        if not self._started:
+            return
+
+        thinking = self.thinking_string()
+
+        self._context.add_message(
+            {
+                "role": "assistant",
+                "content": [
+                    {"type": "thinking", "thinking": thinking, "signature": frame.signature},
+                ],
+            }
+        )
+
     def _context_updated_task_finished(self, task: asyncio.Task):
         self._context_updated_tasks.discard(task)
 

src/pipecat/services/anthropic/llm.py

@@ -40,6 +40,8 @@
     LLMFullResponseStartFrame,
     LLMMessagesFrame,
     LLMTextFrame,
+    LLMThinkingSignatureFrame,
+    LLMThinkingTextFrame,
     LLMUpdateSettingsFrame,
     UserImageRawFrame,
 )
@@ -380,6 +382,10 @@ async def _process_context(self, context: OpenAILLMContext | LLMContext):
                         completion_tokens_estimate += self._estimate_tokens(
                             event.delta.partial_json
                         )
+                    elif hasattr(event.delta, "thinking"):
+                        await self.push_frame(LLMThinkingTextFrame(event.delta.thinking))
+                    elif hasattr(event.delta, "signature"):
+                        await self.push_frame(LLMThinkingSignatureFrame(event.delta.signature))
                 elif event.type == "content_block_start":
                     if event.content_block.type == "tool_use":
                         tool_use_block = event.content_block