mirror of
https://github.com/NousResearch/atropos.git
synced 2026-04-19 12:57:58 +00:00
61fdc37f61
- Update pre-commit config to use ruff with --select=I for imports only - Apply ruff import sorting to fix pre-commit issues - Ruff and black work together without conflicts
121 lines
4.5 KiB
Python
121 lines
4.5 KiB
Python
import logging # Added logging
|
|
import os
|
|
from typing import List, Optional # Add Optional & List import
|
|
|
|
from livekit.agents import ( # Changed import; Add ChatContext import
|
|
ChatContext,
|
|
JobContext,
|
|
WorkerOptions,
|
|
cli,
|
|
mcp, # Corrected import for mcp
|
|
tts, # Corrected import for tts module
|
|
)
|
|
from livekit.agents.llm import ( # Added function_tool for delegate_to_router_agent if it were defined here
|
|
ChatChunk,
|
|
function_tool,
|
|
)
|
|
from livekit.agents.types import NOT_GIVEN # Corrected import for NOT_GIVEN
|
|
from livekit.agents.voice import Agent, AgentSession
|
|
from livekit.plugins import deepgram, openai, silero
|
|
|
|
# Removed: from mcp_client import MCPServerStdio
|
|
# Removed: from mcp_client.agent_tools import MCPToolsIntegration
|
|
from livekit.plugins.turn_detector.multilingual import ( # Added from official example
|
|
MultilingualModel,
|
|
)
|
|
|
|
logger = logging.getLogger("agent-math-official") # Added logger
|
|
|
|
mcp_script_path = os.path.abspath(
|
|
os.path.join(
|
|
os.path.dirname(__file__), "..", "tools", "mcp", "calc", "calc_server.py"
|
|
)
|
|
)
|
|
|
|
|
|
class CalculatorAgent(Agent):
|
|
"""A LiveKit agent that uses MCP tools from one or more MCP servers."""
|
|
|
|
def __init__(
|
|
self,
|
|
chat_ctx: ChatContext,
|
|
instructions: Optional[str] = None,
|
|
mcp_servers: Optional[list[mcp.MCPServer]] = None,
|
|
tts: Optional[tts.TTS] = NOT_GIVEN,
|
|
tools: Optional[List[function_tool]] = None,
|
|
): # Added tools parameter
|
|
|
|
final_instructions = (
|
|
instructions
|
|
if instructions is not None
|
|
else """
|
|
You are a specialist Math assistant. Your expertise is in solving mathematical problems,
|
|
performing calculations, arithmetic, and answering questions about numbers.
|
|
You have two calculation tools: 'multiply' and 'add'.
|
|
When your current math task is complete, or if the user asks for something not related to math,
|
|
you MUST use the 'delegate_to_router_agent' tool to return to the main assistant.
|
|
"""
|
|
)
|
|
|
|
# Combine passed tools with any class-defined tools if necessary (none here for now)
|
|
all_tools = tools if tools is not None else []
|
|
|
|
super().__init__(
|
|
instructions=final_instructions,
|
|
chat_ctx=chat_ctx,
|
|
allow_interruptions=True,
|
|
mcp_servers=[
|
|
mcp.MCPServerStdio(
|
|
command="python",
|
|
args=[mcp_script_path],
|
|
)
|
|
# MODIFIED: Removed chat_ctx=chat_ctx argument
|
|
],
|
|
tools=all_tools, # Pass the tools to the parent Agent class
|
|
)
|
|
# MCP tools are automatically integrated by AgentSession if mcp_servers is configured.
|
|
# No need for MCPToolsIntegration or manually adding tools here.
|
|
|
|
async def llm_node(self, chat_ctx, tools, model_settings):
|
|
"""Override the llm_node to say a message when a tool call is detected."""
|
|
tool_call_detected = False
|
|
|
|
async for chunk in super().llm_node(chat_ctx, tools, model_settings):
|
|
if (
|
|
isinstance(chunk, ChatChunk)
|
|
and chunk.delta
|
|
and chunk.delta.tool_calls
|
|
and not tool_call_detected
|
|
):
|
|
tool_call_detected = True
|
|
# Example: if self.tts: self.session.say("Working on the math problem.")
|
|
# Currently, Math agent does not say anything here.
|
|
yield chunk
|
|
|
|
async def on_enter(self):
|
|
# when the agent is added to the session, we'll initiate the conversation by
|
|
# using the LLM to generate a reply
|
|
self.session.generate_reply()
|
|
|
|
|
|
async def entrypoint(ctx: JobContext):
|
|
"""Main entrypoint for the LiveKit agent application."""
|
|
await ctx.connect() # Connect earlier as in official example
|
|
|
|
# Directly configure AgentSession with mcp_servers
|
|
session = AgentSession(
|
|
vad=silero.VAD.load(), # Redundant if agent has it, but official example does this
|
|
stt=deepgram.STT(model="nova-2", language="en-US"), # Consistent with agent
|
|
llm=openai.LLM(model="gpt-4o"), # Consistent with agent
|
|
tts=openai.TTS(voice="alloy"), # Consistent with agent
|
|
turn_detection=MultilingualModel(), # Consistent with agent
|
|
)
|
|
|
|
# Instantiate the agent
|
|
agent = CalculatorAgent(chat_ctx=session._chat_ctx)
|
|
|
|
await session.start(agent=agent, room=ctx.room)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
cli.run_app(WorkerOptions(entrypoint_fnc=entrypoint))
|