ws-bom-robot-app 0.0.104__py3-none-any.whl → 0.0.106__py3-none-any.whl

ws_bom_robot_app/llm/main.py

@@ -1,159 +1,159 @@
-from asyncio import Queue
-import asyncio, json, logging, os, traceback, re
-from fastapi import Request
-from langchain.callbacks.tracers import LangChainTracer
-from langchain_core.callbacks.base import AsyncCallbackHandler
-from langchain_core.messages import BaseMessage, AIMessage, HumanMessage
-from langsmith import Client as LangSmithClient
-from typing import AsyncGenerator, List
-from ws_bom_robot_app.config import config
-from ws_bom_robot_app.llm.agent_description import AgentDescriptor
-from ws_bom_robot_app.llm.agent_handler import AgentHandler, RawAgentHandler
-from ws_bom_robot_app.llm.agent_lcel import AgentLcel
-from ws_bom_robot_app.llm.models.api import InvokeRequest, StreamRequest
-from ws_bom_robot_app.llm.providers.llm_manager import LlmInterface
-from ws_bom_robot_app.llm.tools.tool_builder import get_structured_tools
-from ws_bom_robot_app.llm.nebuly_handler import NebulyHandler
-
-async def invoke(rq: InvokeRequest) -> str:
-  await rq.initialize()
-  _msg: str = rq.messages[-1].content
-  processor = AgentDescriptor(
-      llm=rq.get_llm(),
-      prompt=rq.system_message,
-      mode = rq.mode,
-      rules=rq.rules if rq.rules else None
-  )
-  result: AIMessage = await processor.run_agent(_msg)
-  return {"result": result.content}
-
-def _parse_formatted_message(message: str) -> str:
-  try:
-    text_fragments = []
-    quoted_strings = re.findall(r'"([^"\\]*(?:\\.[^"\\]*)*)"', message)
-    for string in quoted_strings:
-      if not string.startswith(('threadId', 'type')) and len(string) > 1:
-          text_fragments.append(string)
-    result = ''.join(text_fragments)
-    result = result.replace('\\n', '\n')
-  except:
-    result = message
-  return result
-
-async def __stream(rq: StreamRequest, ctx: Request, queue: Queue, formatted: bool = True) -> None:
-  #os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
-
-  # rq initialization
-  await rq.initialize()
-  for tool in rq.app_tools:
-    tool.thread_id = rq.thread_id
-
-  #llm
-  __llm: LlmInterface = rq.get_llm()
-
-  #chat history
-  chat_history: list[BaseMessage] = []
-  for message in rq.messages:
-    if message.role in ["human","user"]:
-      _content = message.content
-      # multimodal content parsing
-      if isinstance(_content, list):
-        try:
-          _content = await __llm.format_multimodal_content(_content)
-        except Exception as e:
-          logging.warning(f"Error parsing multimodal content {_content[:100]}: {e}")
-      chat_history.append(HumanMessage(content=_content))
-    elif message.role in ["ai","assistant"]:
-      message_content = ""
-      if formatted:
-        if '{\"type\":\"string\"' in message.content:
-          try:
-            json_msg = json.loads('[' + message.content[:-1] + ']')
-            for msg in json_msg:
-              if msg.get("content"):
-                message_content += msg["content"]
-          except:
-            message_content = _parse_formatted_message(message.content)
-        elif '{\"type\":\"text\"' in message.content:
-          try:
-            json_msg = json.loads('[' + message.content[:-1] + ']')
-            for msg in json_msg:
-              if msg.get("text"):
-                message_content += msg["text"]
-          except:
-            message_content = _parse_formatted_message(message.content)
-        else:
-          message_content = _parse_formatted_message(message.content)
-      else:
-        message_content = message.content
-      if message_content:
-        chat_history.append(AIMessage(content=message_content))
-
-
-  #agent handler
-  if formatted:
-    agent_handler = AgentHandler(queue, rq.provider, rq.thread_id)
-  else:
-    agent_handler = RawAgentHandler(queue, rq.provider)
-  #TODO: move from os.environ to rq
-  os.environ["AGENT_HANDLER_FORMATTED"] = str(formatted)
-
-  #callbacks
-  ## agent
-  callbacks: List[AsyncCallbackHandler] = [agent_handler]
-  ## langchain tracing
-  if rq.lang_chain_tracing:
-    client = LangSmithClient(
-      api_key= rq.secrets.get("langChainApiKey", "")
-    )
-    trace = LangChainTracer(project_name=rq.lang_chain_project,client=client,tags=[str(ctx.base_url) if ctx else ''])
-    callbacks.append(trace)
-  ## nebuly tracing
-  if rq.secrets.get("nebulyApiKey","") != "":
-    user_id = rq.system_context.user.id if rq.system_context and rq.system_context.user and rq.system_context.user.id else None
-    nebuly_callback = NebulyHandler(
-        llm_model=__llm.config.model,
-        threadId=rq.thread_id,
-        chat_history=chat_history,
-        url=config.NEBULY_API_URL,
-        api_key=rq.secrets.get("nebulyApiKey", None),
-        user_id=user_id
-      )
-    callbacks.append(nebuly_callback)
-
-  # chain
-  processor = AgentLcel(
-      llm=__llm,
-      sys_message=rq.system_message,
-      sys_context=rq.system_context,
-      tools=get_structured_tools(__llm, tools=rq.app_tools, callbacks=[callbacks], queue=queue),
-      rules=rq.rules,
-      json_schema=rq.output_structure.get("outputFormat") if rq.output_structure and rq.output_structure.get("outputType") == "json" else None
-  )
-  try:
-    await processor.executor.ainvoke(
-        {"chat_history": chat_history},
-        {"callbacks": callbacks},
-    )
-  except Exception as e:
-    _error = f"Agent invoke ex: {e}"
-    logging.warning(_error)
-    if config.runtime_options().debug:
-      _error += f" | {traceback.format_exc()}"
-      await queue.put(_error)
-    await queue.put(None)
-
-  # signal the end of streaming
-  await queue.put(None)
-
-async def stream(rq: StreamRequest, ctx: Request, formatted: bool = True) -> AsyncGenerator[str, None]:
-    queue = Queue()
-    task = asyncio.create_task(__stream(rq, ctx, queue, formatted))
-    try:
-        while True:
-            token = await queue.get()
-            if token is None:  # None indicates the end of streaming
-                break
-            yield token
-    finally:
-        await task
+from asyncio import Queue
+import asyncio, json, logging, os, traceback, re
+from fastapi import Request
+from langchain.callbacks.tracers import LangChainTracer
+from langchain_core.callbacks.base import AsyncCallbackHandler
+from langchain_core.messages import BaseMessage, AIMessage, HumanMessage
+from langsmith import Client as LangSmithClient
+from typing import AsyncGenerator, List
+from ws_bom_robot_app.config import config
+from ws_bom_robot_app.llm.agent_description import AgentDescriptor
+from ws_bom_robot_app.llm.agent_handler import AgentHandler, RawAgentHandler
+from ws_bom_robot_app.llm.agent_lcel import AgentLcel
+from ws_bom_robot_app.llm.models.api import InvokeRequest, StreamRequest
+from ws_bom_robot_app.llm.providers.llm_manager import LlmInterface
+from ws_bom_robot_app.llm.tools.tool_builder import get_structured_tools
+from ws_bom_robot_app.llm.nebuly_handler import NebulyHandler
+
+async def invoke(rq: InvokeRequest) -> str:
+  await rq.initialize()
+  _msg: str = rq.messages[-1].content
+  processor = AgentDescriptor(
+      llm=rq.get_llm(),
+      prompt=rq.system_message,
+      mode = rq.mode,
+      rules=rq.rules if rq.rules else None
+  )
+  result: AIMessage = await processor.run_agent(_msg)
+  return {"result": result.content}
+
+def _parse_formatted_message(message: str) -> str:
+  try:
+    text_fragments = []
+    quoted_strings = re.findall(r'"([^"\\]*(?:\\.[^"\\]*)*)"', message)
+    for string in quoted_strings:
+      if not string.startswith(('threadId', 'type')) and len(string) > 1:
+          text_fragments.append(string)
+    result = ''.join(text_fragments)
+    result = result.replace('\\n', '\n')
+  except:
+    result = message
+  return result
+
+async def __stream(rq: StreamRequest, ctx: Request, queue: Queue, formatted: bool = True) -> None:
+  #os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
+
+  # rq initialization
+  await rq.initialize()
+  for tool in rq.app_tools:
+    tool.thread_id = rq.thread_id
+
+  #llm
+  __llm: LlmInterface = rq.get_llm()
+
+  #chat history
+  chat_history: list[BaseMessage] = []
+  for message in rq.messages:
+    if message.role in ["human","user"]:
+      _content = message.content
+      # multimodal content parsing
+      if isinstance(_content, list):
+        try:
+          _content = await __llm.format_multimodal_content(_content)
+        except Exception as e:
+          logging.warning(f"Error parsing multimodal content {_content[:100]}: {e}")
+      chat_history.append(HumanMessage(content=_content))
+    elif message.role in ["ai","assistant"]:
+      message_content = ""
+      if formatted:
+        if '{\"type\":\"string\"' in message.content:
+          try:
+            json_msg = json.loads('[' + message.content[:-1] + ']')
+            for msg in json_msg:
+              if msg.get("content"):
+                message_content += msg["content"]
+          except:
+            message_content = _parse_formatted_message(message.content)
+        elif '{\"type\":\"text\"' in message.content:
+          try:
+            json_msg = json.loads('[' + message.content[:-1] + ']')
+            for msg in json_msg:
+              if msg.get("text"):
+                message_content += msg["text"]
+          except:
+            message_content = _parse_formatted_message(message.content)
+        else:
+          message_content = _parse_formatted_message(message.content)
+      else:
+        message_content = message.content
+      if message_content:
+        chat_history.append(AIMessage(content=message_content))
+
+
+  #agent handler
+  if formatted:
+    agent_handler = AgentHandler(queue, rq.provider, rq.thread_id)
+  else:
+    agent_handler = RawAgentHandler(queue, rq.provider)
+  #TODO: move from os.environ to rq
+  os.environ["AGENT_HANDLER_FORMATTED"] = str(formatted)
+
+  #callbacks
+  ## agent
+  callbacks: List[AsyncCallbackHandler] = [agent_handler]
+  ## langchain tracing
+  if rq.lang_chain_tracing:
+    client = LangSmithClient(
+      api_key= rq.secrets.get("langChainApiKey", "")
+    )
+    trace = LangChainTracer(project_name=rq.lang_chain_project,client=client,tags=[str(ctx.base_url) if ctx else ''])
+    callbacks.append(trace)
+  ## nebuly tracing
+  if rq.secrets.get("nebulyApiKey","") != "":
+    user_id = rq.system_context.user.id if rq.system_context and rq.system_context.user and rq.system_context.user.id else None
+    nebuly_callback = NebulyHandler(
+        llm_model=__llm.config.model,
+        threadId=rq.thread_id,
+        chat_history=chat_history,
+        url=config.NEBULY_API_URL,
+        api_key=rq.secrets.get("nebulyApiKey", None),
+        user_id=user_id
+      )
+    callbacks.append(nebuly_callback)
+
+  # chain
+  processor = AgentLcel(
+      llm=__llm,
+      sys_message=rq.system_message,
+      sys_context=rq.system_context,
+      tools=get_structured_tools(__llm, tools=rq.app_tools, callbacks=[callbacks], queue=queue),
+      rules=rq.rules,
+      json_schema=rq.output_structure.get("outputFormat") if rq.output_structure and rq.output_structure.get("outputType") == "json" else None
+  )
+  try:
+    await processor.executor.ainvoke(
+        {"chat_history": chat_history},
+        {"callbacks": callbacks},
+    )
+  except Exception as e:
+    _error = f"Agent invoke ex: {e}"
+    logging.warning(_error)
+    if config.runtime_options().debug:
+      _error += f" | {traceback.format_exc()}"
+      await queue.put(_error)
+    await queue.put(None)
+
+  # signal the end of streaming
+  await queue.put(None)
+
+async def stream(rq: StreamRequest, ctx: Request, formatted: bool = True) -> AsyncGenerator[str, None]:
+    queue = Queue()
+    task = asyncio.create_task(__stream(rq, ctx, queue, formatted))
+    try:
+        while True:
+            token = await queue.get()
+            if token is None:  # None indicates the end of streaming
+                break
+            yield token
+    finally:
+        await task

ws_bom_robot_app/llm/models/api.py

@@ -66,6 +66,7 @@ class LlmAppToolChainSettings(BaseModel):
 
 class LlmAppToolDbSettings(BaseModel):
   connection_string: Optional[str] = Field(None, validation_alias=AliasChoices("connectionString","connection_string"))
+  additionalPrompt: Optional[str] = Field(None, validation_alias=AliasChoices("additionalPrompt","additional_prompt"))
 
 class LlmAppTool(BaseModel):
   id: Optional[str] = None

ws_bom_robot_app/llm/models/feedback.py

@@ -1,30 +1,30 @@
-from pydantic import BaseModel, Field
-
-class NebulyFeedbackAction(BaseModel):
-    """
-    FeedbackAction is a model that represents the action taken by the user
-    in response to the feedback provided by the LLM.
-    """
-    slug: str = Field("rating", description="A string identifier for the feedback action",
-             enum=["thumbs_up", "thumbs_down", "copy_input", "copy_output", "paste", "rating"])
-    text: str = Field(..., description="The text content of the feedback")
-    value: int = Field(..., description="A numeric value associated with the feedback")
-
-class NebulyFeedbackMetadata(BaseModel):
-  """
-  FeedbackMetadata is a model that represents the metadata associated with user feedback.
-  This includes information about the interaction and the user who provided feedback.
-  """
-  input: str = Field(None, description="The input of the interactions to which the action refers to")
-  output: str = Field(None, description="The output of the interactions to which the action refers to")
-  end_user: str = Field(..., description="The identifier used for the end-user")
-  timestamp: str = Field(..., description="The timestamp of the action event")
-  anonymize: bool = Field(False, description="Boolean flag. If set to true, PII will be removed from the text field")
-
-class NebulyFeedbackPayload(BaseModel):
-  """
-  NebulyFeedback is a model that combines feedback action and metadata.
-  It represents a complete feedback entry from a user interaction with the LLM.
-  """
-  action: NebulyFeedbackAction = Field(..., description="The action taken by the user as feedback")
-  metadata: NebulyFeedbackMetadata = Field(..., description="Metadata associated with the feedback")
+from pydantic import BaseModel, Field
+
+class NebulyFeedbackAction(BaseModel):
+    """
+    FeedbackAction is a model that represents the action taken by the user
+    in response to the feedback provided by the LLM.
+    """
+    slug: str = Field("rating", description="A string identifier for the feedback action",
+             enum=["thumbs_up", "thumbs_down", "copy_input", "copy_output", "paste", "rating"])
+    text: str = Field(..., description="The text content of the feedback")
+    value: int = Field(..., description="A numeric value associated with the feedback")
+
+class NebulyFeedbackMetadata(BaseModel):
+  """
+  FeedbackMetadata is a model that represents the metadata associated with user feedback.
+  This includes information about the interaction and the user who provided feedback.
+  """
+  input: str = Field(None, description="The input of the interactions to which the action refers to")
+  output: str = Field(None, description="The output of the interactions to which the action refers to")
+  end_user: str = Field(..., description="The identifier used for the end-user")
+  timestamp: str = Field(..., description="The timestamp of the action event")
+  anonymize: bool = Field(False, description="Boolean flag. If set to true, PII will be removed from the text field")
+
+class NebulyFeedbackPayload(BaseModel):
+  """
+  NebulyFeedback is a model that combines feedback action and metadata.
+  It represents a complete feedback entry from a user interaction with the LLM.
+  """
+  action: NebulyFeedbackAction = Field(..., description="The action taken by the user as feedback")
+  metadata: NebulyFeedbackMetadata = Field(..., description="Metadata associated with the feedback")