Spaces:
Running
Running
File size: 8,185 Bytes
0646b18 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 |
import json
import uuid
from typing import Literal, Optional, Dict, Callable
from langchain_core.messages import HumanMessage, ToolCall, BaseMessage
from loguru import logger
from cuga.backend.activity_tracker.tracker import ActivityTracker, Step
from cuga.backend.cuga_graph.nodes.shared.base_agent import create_partial
from cuga.backend.cuga_graph.nodes.chat.chat_agent.chat_agent import ChatAgent
from cuga.backend.cuga_graph.nodes.shared.base_node import BaseNode
from cuga.backend.cuga_graph.nodes.human_in_the_loop.followup_model import (
create_flow_approve,
create_new_flow_approve,
)
from cuga.backend.cuga_graph.state.agent_state import AgentState
from cuga.backend.cuga_graph.utils.nodes_names import NodeNames, ActionIds
from langgraph.types import Command
from cuga.config import settings
tracker = ActivityTracker()
ENABLE_SAVE_REUSE = settings.features.save_reuse
class ChatHumanInTheLoopHandler:
"""Handler for chat-specific human-in-the-loop interactions"""
def __init__(self):
self._action_handlers: Dict[str, Callable] = {
ActionIds.FLOW_APPROVE: self._handle_tool_execute,
# Add chat-specific action handlers here
# Example: ActionIds.TOOL_EXECUTE: self._handle_tool_execute,
# ActionIds.CHAT_CONTINUE: self._handle_chat_continue,
}
def handle_human_response(self, state: AgentState, node_name: str) -> Command:
"""Handle any human response based on action_id"""
action_id = state.hitl_response.action_id
if action_id in self._action_handlers:
return self._action_handlers[action_id](state, node_name)
# Default fallback for chat - continue to final answer
return Command(update=state.model_dump(), goto=NodeNames.FINAL_ANSWER_AGENT)
def add_action_handler(self, action_id: str, handler: Callable):
"""Add a custom action handler"""
self._action_handlers[action_id] = handler
def _handle_tool_execute(self, state: AgentState, node_name: str) -> Command:
"""Handle tool execution approval"""
state.sender = node_name
return Command(update=state.model_dump(), goto=NodeNames.WAIT_FOR_RESPONSE)
def _handle_chat_continue(self, state: AgentState, node_name: str) -> Command:
"""Handle continuing chat conversation"""
state.sender = node_name
return Command(update=state.model_dump(), goto=NodeNames.FINAL_ANSWER_AGENT)
class ChatNode(BaseNode):
def __init__(self):
super().__init__()
self.chat_agent: Optional[ChatAgent] = None
self.hitl_handler = ChatHumanInTheLoopHandler()
self._initialized = False
@classmethod
async def create(cls):
"""Factory method to create and initialize the class"""
instance = cls()
instance.chat_agent = ChatAgent()
if settings.features.chat:
await instance.chat_agent.setup()
instance.node = create_partial(
ChatNode.node_handler,
agent=instance.chat_agent,
hitl_handler=instance.hitl_handler,
name=instance.chat_agent.name,
)
instance._initialized = True
return instance
@staticmethod
def format_function_call(func_dict):
name = func_dict["name"]
args = func_dict["args"]
def format_value(v):
if isinstance(v, str):
return f"'{v}'"
elif isinstance(v, (list, dict)):
return repr(v)
else:
return str(v)
arg_strings = [f"{k}={format_value(v)}" for k, v in args.items()]
return f"{name}({', '.join(arg_strings)})"
@staticmethod
async def node_handler(
state: AgentState, agent: ChatAgent, hitl_handler: ChatHumanInTheLoopHandler, name: str
) -> Command[Literal["FinalAnswerAgent", "TaskAnalyzerAgent", "SuggestHumanActions"]]:
# Handle human-in-the-loop responses
if (
state.sender == NodeNames.WAIT_FOR_RESPONSE
and state.hitl_response.action_id == ActionIds.FLOW_APPROVE
):
tool = ToolCall(**state.hitl_response.additional_data.tool)
res = await agent.execute_tool(tool)
parsed_result = res
if isinstance(res, str):
try:
parsed_result = json.loads(res)
except (json.JSONDecodeError, TypeError):
# If parsing fails, keep original string
parsed_result = res
# Get tool details
tool_name = tool.get("name")
tool_args = tool.get("args")
# Add to variable manager
var_name = f"tool_result_{str(uuid.uuid4())[:5]}"
state.variables_manager.add_variable(
parsed_result, var_name, f"Result of tool {tool_name} with args {tool_args}"
)
state.sender = "ChatAgentTool"
state.last_planner_answer = state.variables_manager.present_variable(var_name)
return Command(update=state.model_dump(), goto=NodeNames.FINAL_ANSWER_AGENT)
if (
state.sender == NodeNames.WAIT_FOR_RESPONSE
and state.hitl_response.action_id == ActionIds.NEW_FLOW_APPROVE
):
logger.debug("tool call in chat node")
tool = ToolCall(**state.hitl_response.additional_data.tool)
state.input = tool.get("args").get("user_task")
state.sender = "ChatAgent"
return Command(update=state.model_dump(), goto=NodeNames.TASK_ANALYZER_AGENT)
# If chat feature is disabled, go directly to task analyzer
if not settings.features.chat:
state.sender = name
return Command(update=state.model_dump(), goto=NodeNames.TASK_ANALYZER_AGENT)
# Process chat input
state.sender = name
state.chat_agent_messages.append(HumanMessage(content=state.input))
res: BaseMessage = await agent.invoke(state.chat_agent_messages, state)
state.chat_agent_messages.append(res)
# Handle tool calls - require human approval
if ENABLE_SAVE_REUSE and res.tool_calls and res.tool_calls[0].get("name") == "run_new_flow":
state.final_answer = state.chat_agent_messages[-1].content
state.sender = name
state.hitl_action = create_new_flow_approve(tool=res.tool_calls[0])
return Command(update=state.model_dump(), goto=NodeNames.SUGGEST_HUMAN_ACTIONS)
if ENABLE_SAVE_REUSE and res.tool_calls:
state.final_answer = state.chat_agent_messages[-1].content
state.sender = name
state.hitl_action = create_flow_approve(tool=res.tool_calls[0])
return Command(update=state.model_dump(), goto=NodeNames.SUGGEST_HUMAN_ACTIONS)
if (
not ENABLE_SAVE_REUSE
and res.tool_calls
and len(res.tool_calls) > 0
and res.tool_calls[0].get("name") == "execute_task"
):
logger.debug(f"tool call in chat node {res.tool_calls[0]}")
variables_rel = res.tool_calls[0].get("args").get("relevant_variables")
if variables_rel and len(variables_rel) > 0:
state.input = (
f"task: {res.tool_calls[0].get('args').get('task')}"
+ f"\n relevant variables from history: {res.tool_calls[0].get('args').get('relevant_variables')}"
)
else:
state.input = res.tool_calls[0].get("args").get("task")
return Command(update=state.model_dump(), goto="TaskAnalyzerAgent")
# Regular chat response - add to messages and continue+
res.content = state.variables_manager.replace_variables_placeholders(res.content)
state.messages.append(res)
tracker.collect_step(
step=Step(
name=name,
data=res.content,
current_url=state.url,
)
)
state.final_answer = state.chat_agent_messages[-1].content
return Command(update=state.model_dump(), goto=NodeNames.FINAL_ANSWER_AGENT)
|