dify/api/services/agent_service.py
2024-05-04 16:17:15 +08:00

131 lines
5.2 KiB
Python

import pytz
from flask_login import current_user
from core.app.app_config.easy_ui_based_app.agent.manager import AgentConfigManager
from core.tools.tool_manager import ToolManager
from extensions.ext_database import db
from models.account import Account
from models.model import App, Conversation, EndUser, Message, MessageAgentThought
class AgentService:
@classmethod
def get_agent_logs(cls, app_model: App,
conversation_id: str,
message_id: str) -> dict:
"""
Service to get agent logs
"""
conversation: Conversation = db.session.query(Conversation).filter(
Conversation.id == conversation_id,
Conversation.app_id == app_model.id,
).first()
if not conversation:
raise ValueError(f"Conversation not found: {conversation_id}")
message: Message = db.session.query(Message).filter(
Message.id == message_id,
Message.conversation_id == conversation_id,
).first()
if not message:
raise ValueError(f"Message not found: {message_id}")
agent_thoughts: list[MessageAgentThought] = message.agent_thoughts
if conversation.from_end_user_id:
# only select name field
executor = db.session.query(EndUser, EndUser.name).filter(
EndUser.id == conversation.from_end_user_id
).first()
else:
executor = db.session.query(Account, Account.name).filter(
Account.id == conversation.from_account_id
).first()
if executor:
executor = executor.name
else:
executor = 'Unknown'
timezone = pytz.timezone(current_user.timezone)
result = {
'meta': {
'status': 'success',
'executor': executor,
'start_time': message.created_at.astimezone(timezone).isoformat(),
'elapsed_time': message.provider_response_latency,
'total_tokens': message.answer_tokens + message.message_tokens,
'agent_mode': app_model.app_model_config.agent_mode_dict.get('strategy', 'react'),
'iterations': len(agent_thoughts),
},
'iterations': [],
'files': message.files,
}
agent_config = AgentConfigManager.convert(app_model.app_model_config.to_dict())
agent_tools = agent_config.tools
def find_agent_tool(tool_name: str):
for agent_tool in agent_tools:
if agent_tool.tool_name == tool_name:
return agent_tool
for agent_thought in agent_thoughts:
tools = agent_thought.tools
tool_labels = agent_thought.tool_labels
tool_meta = agent_thought.tool_meta
tool_inputs = agent_thought.tool_inputs_dict
tool_outputs = agent_thought.tool_outputs_dict
tool_calls = []
for tool in tools:
tool_name = tool
tool_label = tool_labels.get(tool_name, tool_name)
tool_input = tool_inputs.get(tool_name, {})
tool_output = tool_outputs.get(tool_name, {})
tool_meta_data = tool_meta.get(tool_name, {})
tool_config = tool_meta_data.get('tool_config', {})
if tool_config.get('tool_provider_type', '') != 'dataset-retrieval':
tool_icon = ToolManager.get_tool_icon(
tenant_id=app_model.tenant_id,
provider_type=tool_config.get('tool_provider_type', ''),
provider_id=tool_config.get('tool_provider', ''),
)
if not tool_icon:
tool_entity = find_agent_tool(tool_name)
if tool_entity:
tool_icon = ToolManager.get_tool_icon(
tenant_id=app_model.tenant_id,
provider_type=tool_entity.provider_type,
provider_id=tool_entity.provider_id,
)
else:
tool_icon = ''
tool_calls.append({
'status': 'success' if not tool_meta_data.get('error') else 'error',
'error': tool_meta_data.get('error'),
'time_cost': tool_meta_data.get('time_cost', 0),
'tool_name': tool_name,
'tool_label': tool_label,
'tool_input': tool_input,
'tool_output': tool_output,
'tool_parameters': tool_meta_data.get('tool_parameters', {}),
'tool_icon': tool_icon,
})
result['iterations'].append({
'tokens': agent_thought.tokens,
'tool_calls': tool_calls,
'tool_raw': {
'inputs': agent_thought.tool_input,
'outputs': agent_thought.observation,
},
'thought': agent_thought.thought,
'created_at': agent_thought.created_at.isoformat(),
'files': agent_thought.files,
})
return result