​
FunctionCallResultFrame

Frame containing the result of an LLM function (tool) call.

​
Properties

function_name
str
required

Name of the function that was called

tool_call_id
str
required

Unique identifier for the tool call

arguments
str
required

Arguments passed to the function, typically as JSON string

result
Any
required

Result returned by the function

run_llm
bool
default: "true"

Whether to run the LLM with the function result

​
FunctionCallInProgressFrame

Frame indicating an ongoing function call.

​
Properties

function_name
str
required

Name of the function being called

tool_call_id
str
required

Unique identifier for the tool call

arguments
str
required

Arguments being passed to the function

​
Usage Examples

​
Function Call Implementation

# Define a function handler
async def handle_weather_lookup(location: str) -> FunctionCallResultFrame:
    try:
        weather_data = await get_weather(location)
        return FunctionCallResultFrame(
            function_name="get_weather",
            tool_call_id="call_123",
            arguments=json.dumps({"location": location}),
            result=weather_data,
            run_llm=True  # Let LLM process the result
        )
    except Exception as e:
        return FunctionCallResultFrame(
            function_name="get_weather",
            tool_call_id="call_123",
            arguments=json.dumps({"location": location}),
            result=str(e),
            run_llm=True
        )

​
Progress Tracking

# Signal function execution start
async def start_function_call(name: str, call_id: str, args: dict):
    await pipeline.push_frame(FunctionCallInProgressFrame(
        function_name=name,
        tool_call_id=call_id,
        arguments=json.dumps(args)
    ))

​
Pipeline Integration

# Function calling pipeline
pipeline = Pipeline([
    llm_processor,        # Generates function calls
    function_router,      # Routes calls to handlers
    result_processor,     # Processes function results
    response_generator    # Generates final response
])

​
Frame Flow

​
Notes

  • Function calls are typically initiated by LLM output
  • Tool call IDs should be unique per conversation
  • Arguments should be valid JSON strings
  • Results can be of any type that the LLM can process
  • Progress frames help track long-running operations
LLM FramesSpeech Event Frames