ws-bom-robot-app 0.0.33__py3-none-any.whl → 0.0.34__py3-none-any.whl

ws_bom_robot_app/config.py

@@ -20,12 +20,21 @@ class Settings(BaseSettings):
     robot_cms_auth: str = ''
     robot_cms_db_folder: str = 'llmVectorDb'
     robot_cms_kb_folder: str ='llmKbFile'
-    robot_debugger_openai_key: str = ''
+    robot_debugger_llm_provider: str = 'openai'
+    robot_debugger_llm_model: str = 'gpt-4o'
+    robot_debugger_llm_key: str = ''
+    robot_debugger_embedding_key: str = ''
+    OPENAI_API_KEY: str = '' # used for dall-e api
+    GOOGLE_APPLICATION_CREDENTIALS: str = './.data/credentials/google.json' # path to google credentials iam file
     model_config = ConfigDict(
         env_file='./.env',
         extra='ignore',
         case_sensitive=False
     )
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        os.environ["OPENAI_API_KEY"] = self.OPENAI_API_KEY
+        os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = self.GOOGLE_APPLICATION_CREDENTIALS
 
     class RuntimeOptions(BaseModel):
         @staticmethod

ws_bom_robot_app/llm/agent_description.py

@@ -1,124 +1,123 @@
-import json, requests, re
-from typing import Any
-from abc import ABC, abstractmethod
-from langchain_openai import ChatOpenAI
-from langchain_core.prompts import ChatPromptTemplate
-from langchain_core.messages import AIMessage
-from langchain_core.runnables import RunnableSerializable
-from langchain_core.runnables import RunnableLambda
-from bs4 import BeautifulSoup
-from ws_bom_robot_app.llm.models.api import LlmRules
-from ws_bom_robot_app.llm.utils.agent_utils import get_rules
-
-# SafeDict helper class
-class SafeDict(dict):
-    def __missing__(self, key):
-        return ''
-
-# Strategy Interface
-class AgentDescriptorStrategy(ABC):
-    @abstractmethod
-    def enrich_prompt(self, prompt: str, input: dict) -> str:
-        pass
-
-    @abstractmethod
-    def rule_input(self, input: dict) -> str:
-        pass
-
-# Concrete Strategy for Default Agent
-class DefaultAgentDescriptor(AgentDescriptorStrategy):
-    def enrich_prompt(self, prompt: str, input: dict) -> str:
-        # Default enrichment logic (could be minimal or no-op)
-        return prompt.format_map(SafeDict(input))
-
-    def rule_input(self, input: dict) -> str:
-        return input.get('content', "")
-
-# Concrete Strategy for URL2Text Agent
-class URL2TextAgentDescriptor(AgentDescriptorStrategy):
-    def enrich_prompt(self, prompt: str, input: dict) -> str:
-        input["context"] = self._get_page_text(input)
-        return prompt.format_map(SafeDict(input))
-
-    def rule_input(self, input: dict) -> str:
-        return input.get('context', "")
-
-    def _get_page_text(self, input: dict) -> str:
-        url = input.get("content", "")
-        exclusions = input.get("exclude", {})
-        response = requests.get(url)
-        response.raise_for_status()
-        soup = BeautifulSoup(response.content, 'html5lib')
-        classes_to_exclude = exclusions.get("classes", [])
-        ids_to_exclude = exclusions.get("ids", [])
-        for class_name in classes_to_exclude:
-            for element in soup.find_all(class_=class_name):
-                element.extract()
-        for id_name in ids_to_exclude:
-            for element in soup.find_all(id=id_name):
-                element.extract()
-        for script in soup(["script", "noscript", "style", "head", "footer", "iframe"]):
-            script.extract()
-        return re.sub(' +', ' ', soup.get_text())
-
-
-class AgentDescriptor:
-    # Dictionary to hold all agent strategies
-    _list: dict[str,AgentDescriptorStrategy] = {
-        "default": DefaultAgentDescriptor(),
-        "url2text": URL2TextAgentDescriptor(),
-    }
-
-    # Functions to manage strategies
-    @staticmethod
-    def add_strategy(name: str, strategy: AgentDescriptorStrategy):
-        """_summary_
-            add a new strategy to the dictionary
-        Args:
-            name (str): name of the strategy, in lowercase
-            strategy (AgentDescriptorStrategy): class implementing the strategy
-        Examples:
-            AgentDescriptor.add_strategy("custom_agent_descriptor", CustomAgentDescriptor())
-        """
-        AgentDescriptor._list[name.lower()] = strategy
-
-    @staticmethod
-    def get_strategy(name: str) -> AgentDescriptorStrategy:
-        return AgentDescriptor._list.get(name.lower(), DefaultAgentDescriptor())
-
-    def __init__(self, api_key: str, prompt: str, mode: str, rules: LlmRules = None):
-        self.__prompt = prompt
-        self.__llm = ChatOpenAI(model="gpt-4o", temperature=0, api_key=api_key)  # type: ignore
-        self.api_key = api_key
-        self.rules= rules
-        self.strategy = self.get_strategy(mode)  # Selects the strategy from the dictionary
-
-    async def __create_prompt(self, input_dict: dict):
-        input_data = json.loads(input_dict.get("input", {}))
-        system = self.strategy.enrich_prompt(self.__prompt, input_data)
-        if self.rules:
-            rule_input = self.strategy.rule_input(input_data)
-            rules_prompt = await get_rules(self.rules,self.api_key, rule_input)
-            system += rules_prompt
-        return ChatPromptTemplate.from_messages(
-            [
-                ("system", system),
-                ("user", input_data.get("content", ""))
-            ]
-        )
-
-    def __create_agent_descriptor(self, content) -> RunnableSerializable[Any, Any]:
-        content = json.loads(content)
-        agent = (
-            {
-                "input": lambda x: x["input"],
-            }
-            | RunnableLambda(self.__create_prompt)
-            | self.__llm
-        )
-        return agent
-
-    async def run_agent(self, content) -> Any:
-        agent_descriptor = self.__create_agent_descriptor(content)
-        response: AIMessage = await agent_descriptor.ainvoke({"input": content})
-        return response
+import json, requests, re
+from typing import Any
+from abc import ABC, abstractmethod
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.messages import AIMessage
+from langchain_core.runnables import RunnableSerializable
+from langchain_core.runnables import RunnableLambda
+from bs4 import BeautifulSoup
+from ws_bom_robot_app.llm.models.api import LlmRules
+from ws_bom_robot_app.llm.providers.llm_manager import LlmInterface
+from ws_bom_robot_app.llm.utils.agent_utils import get_rules
+
+# SafeDict helper class
+class SafeDict(dict):
+    def __missing__(self, key):
+        return ''
+
+# Strategy Interface
+class AgentDescriptorStrategy(ABC):
+    @abstractmethod
+    def enrich_prompt(self, prompt: str, input: dict) -> str:
+        pass
+
+    @abstractmethod
+    def rule_input(self, input: dict) -> str:
+        pass
+
+# Concrete Strategy for Default Agent
+class DefaultAgentDescriptor(AgentDescriptorStrategy):
+    def enrich_prompt(self, prompt: str, input: dict) -> str:
+        # Default enrichment logic (could be minimal or no-op)
+        return prompt.format_map(SafeDict(input))
+
+    def rule_input(self, input: dict) -> str:
+        return input.get('content', "")
+
+# Concrete Strategy for URL2Text Agent
+class URL2TextAgentDescriptor(AgentDescriptorStrategy):
+    def enrich_prompt(self, prompt: str, input: dict) -> str:
+        input["context"] = self._get_page_text(input)
+        return prompt.format_map(SafeDict(input))
+
+    def rule_input(self, input: dict) -> str:
+        return input.get('context', "")
+
+    def _get_page_text(self, input: dict) -> str:
+        url = input.get("content", "")
+        exclusions = input.get("exclude", {})
+        response = requests.get(url)
+        response.raise_for_status()
+        soup = BeautifulSoup(response.content, 'html5lib')
+        classes_to_exclude = exclusions.get("classes", [])
+        ids_to_exclude = exclusions.get("ids", [])
+        for class_name in classes_to_exclude:
+            for element in soup.find_all(class_=class_name):
+                element.extract()
+        for id_name in ids_to_exclude:
+            for element in soup.find_all(id=id_name):
+                element.extract()
+        for script in soup(["script", "noscript", "style", "head", "footer", "iframe"]):
+            script.extract()
+        return re.sub(' +', ' ', soup.get_text())
+
+
+class AgentDescriptor:
+    # Dictionary to hold all agent strategies
+    _list: dict[str,AgentDescriptorStrategy] = {
+        "default": DefaultAgentDescriptor(),
+        "url2text": URL2TextAgentDescriptor(),
+    }
+
+    # Functions to manage strategies
+    @staticmethod
+    def add_strategy(name: str, strategy: AgentDescriptorStrategy):
+        """_summary_
+            add a new strategy to the dictionary
+        Args:
+            name (str): name of the strategy, in lowercase
+            strategy (AgentDescriptorStrategy): class implementing the strategy
+        Examples:
+            AgentDescriptor.add_strategy("custom_agent_descriptor", CustomAgentDescriptor())
+        """
+        AgentDescriptor._list[name.lower()] = strategy
+
+    @staticmethod
+    def get_strategy(name: str) -> AgentDescriptorStrategy:
+        return AgentDescriptor._list.get(name.lower(), DefaultAgentDescriptor())
+
+    def __init__(self, llm: LlmInterface, prompt: str, mode: str, rules: LlmRules = None):
+        self.__prompt = prompt
+        self.__llm = llm,
+        self.rules= rules
+        self.strategy = self.get_strategy(mode)  # Selects the strategy from the dictionary
+
+    async def __create_prompt(self, input_dict: dict):
+        input_data = json.loads(input_dict.get("input", {}))
+        system = self.strategy.enrich_prompt(self.__prompt, input_data)
+        if self.rules:
+            rule_input = self.strategy.rule_input(input_data)
+            rules_prompt = await get_rules(self.__llm.get_embeddings(), self.rules, rule_input)
+            system += rules_prompt
+        return ChatPromptTemplate.from_messages(
+            [
+                ("system", system),
+                ("user", input_data.get("content", ""))
+            ]
+        )
+
+    def __create_agent_descriptor(self, content) -> RunnableSerializable[Any, Any]:
+        content = json.loads(content)
+        agent = (
+            {
+                "input": lambda x: x["input"],
+            }
+            | RunnableLambda(self.__create_prompt)
+            | self.__llm.get_llm()
+        )
+        return agent
+
+    async def run_agent(self, content) -> Any:
+        agent_descriptor = self.__create_agent_descriptor(content)
+        response: AIMessage = await agent_descriptor.ainvoke({"input": content})
+        return response

ws_bom_robot_app/llm/agent_handler.py

@@ -1,167 +1,180 @@
-from asyncio import Queue
-from langchain_core.agents import AgentFinish
-from langchain_core.outputs import ChatGenerationChunk, GenerationChunk
-from langchain.callbacks.base import AsyncCallbackHandler
-from ws_bom_robot_app.llm.utils.print import printJson, printString
-from typing import Any, Dict, List, Optional, Union
-from uuid import UUID
-import ws_bom_robot_app.llm.settings as settings
-from langchain_core.callbacks.base import AsyncCallbackHandler
-from langchain_core.outputs import ChatGenerationChunk, GenerationChunk
-from langchain_core.messages import BaseMessage, HumanMessage, AIMessage
-import json
-
-# Here is a custom handler that will print the tokens to stdout.
-# Instead of printing to stdout you can send the data elsewhere; e.g., to a streaming API response
-
-
-class AgentHandler(AsyncCallbackHandler):
-
-    def __init__(self, queue: Queue, threadId: str = None) -> None:
-        super().__init__()
-        self._threadId = threadId
-        self.json_block = ""
-        self.is_json_block = False
-        self.backtick_count = 0  # Conteggio dei backticks per il controllo accurato
-        self.queue = queue
-
-    async def on_llm_start(
-        self,
-        serialized: Dict[str, Any],
-        prompts: List[str],
-        *,
-        run_id: UUID,
-        parent_run_id: UUID = None,
-        tags: List[str] = None,
-        metadata: Dict[str, Any] = None,
-        **kwargs: Any,
-    ) -> None:
-        firstChunk = {
-            "type": "info",
-            "threadId": self._threadId,
-        }
-        await self.queue.put(printString(firstChunk))
-
-    """async def on_chat_model_start(self, serialized: Dict[str, Any], messages: List[List[BaseMessage]], *, run_id: UUID = None, parent_run_id = None, tags = None, metadata = None, **kwargs: Any) -> Any:
-        pass"""
-
-    async def on_tool_end(self, output: Any, *, run_id: UUID, parent_run_id: UUID = None, tags: List[str] = None, **kwargs: Any) -> None:
-      pass
-
-    async def on_llm_new_token(
-        self,
-        token: str,
-        *,
-        chunk: Optional[Union[GenerationChunk, ChatGenerationChunk]] = None,
-        run_id: UUID,
-        parent_run_id: Optional[UUID] = None,
-        tags: Optional[List[str]] = None,
-        **kwargs: Any,
-    ) -> None:
-        """Gestisce i nuovi token durante lo streaming."""
-
-        if token != "":
-            self.backtick_count += token.count("`")
-
-            if self.backtick_count >= 3:
-                if not self.is_json_block:
-                    self.is_json_block = True
-                    self.json_block = ""
-                else:
-                    self.is_json_block = False
-                    self.json_block += token.replace("```json", '')
-                    await self.process_json_block(self.json_block)
-                    self.json_block = ""
-                self.backtick_count = 0
-            elif self.is_json_block:
-                self.json_block += token
-            else:
-                await self.queue.put(printString(token))
-        pass
-
-    async def on_agent_finish(
-        self,
-        finish: AgentFinish,
-        *,
-        run_id: UUID,
-        parent_run_id: UUID = None,
-        tags: List[str] = None,
-        **kwargs: Any,
-    ) -> None:
-        settings.chat_history.extend(
-            [
-                AIMessage(content=finish.return_values["output"]),
-            ]
-        )
-        finalChunk = {"type": "end"}
-        await self.queue.put(printJson(finalChunk))
-        await self.queue.put(None)
-
-    async def process_json_block(self, json_block: str):
-      """Processa il blocco JSON completo."""
-      # Rimuove il delimitatore iniziale '```json' se presente, e spazi vuoti
-      json_block_clean = json_block.replace('```', '').replace('json', '').strip()
-      # Verifica che il blocco non sia vuoto prima di tentare il parsing
-      if json_block_clean:
-          try:
-              # Prova a fare il parsing del JSON
-              parsed_json = json.loads(json_block_clean)
-              await self.queue.put(printJson(parsed_json))
-          except json.JSONDecodeError as e:
-              # Se il JSON è malformato, logga l'errore
-              raise e
-
-class RawAgentHandler(AsyncCallbackHandler):
-
-    def __init__(self,queue: Queue) -> None:
-        super().__init__()
-        self.queue = queue
-
-    async def on_llm_start(
-        self,
-        serialized: Dict[str, Any],
-        prompts: List[str],
-        *,
-        run_id: UUID,
-        parent_run_id: UUID = None,
-        tags: List[str] = None,
-        metadata: Dict[str, Any] = None,
-        **kwargs: Any,
-    ) -> None:
-        pass
-
-    """async def on_chat_model_start(self, serialized: Dict[str, Any], messages: List[List[BaseMessage]], *, run_id: UUID = None, parent_run_id = None, tags = None, metadata = None, **kwargs: Any) -> Any:
-        pass"""
-
-    async def on_tool_end(self, output: Any, *, run_id: UUID, parent_run_id: UUID = None, tags: List[str] = None, **kwargs: Any) -> None:
-      pass
-
-    async def on_llm_new_token(
-        self,
-        token: str,
-        *,
-        chunk: Optional[Union[GenerationChunk, ChatGenerationChunk]] = None,
-        run_id: UUID,
-        parent_run_id: Optional[UUID] = None,
-        tags: Optional[List[str]] = None,
-        **kwargs: Any,
-    ) -> None:
-        """Handles new tokens during streaming."""
-        if token:  # Only process non-empty tokens
-            await self.queue.put(token)
-
-    async def on_agent_finish(
-        self,
-        finish: AgentFinish,
-        *,
-        run_id: UUID,
-        parent_run_id: UUID = None,
-        tags: List[str] = None,
-        **kwargs: Any,
-    ) -> None:
-        settings.chat_history.extend(
-            [
-                AIMessage(content=finish.return_values["output"]),
-            ]
-        )
-        await self.queue.put(None)
+from asyncio import Queue
+from langchain_core.agents import AgentFinish
+from langchain_core.outputs import ChatGenerationChunk, GenerationChunk
+from langchain.callbacks.base import AsyncCallbackHandler
+from ws_bom_robot_app.llm.utils.print import printJson, printString
+from typing import Any, Dict, List, Optional, Union
+from uuid import UUID
+import ws_bom_robot_app.llm.settings as settings
+from langchain_core.callbacks.base import AsyncCallbackHandler
+from langchain_core.outputs import ChatGenerationChunk, GenerationChunk
+from langchain_core.messages import BaseMessage, HumanMessage, AIMessage
+import json
+
+# Here is a custom handler that will print the tokens to stdout.
+# Instead of printing to stdout you can send the data elsewhere; e.g., to a streaming API response
+
+
+def _parse_token(llm:str,token: str) -> str:
+    """Parses the token based on the LLM provider."""
+    if llm == "anthropic" and isinstance(token, list):
+      first = token[0]
+      if 'text' in first:
+        token = first['text']
+      else:
+        #[{'id': 'toolu_01GGLwJcrQ8PvFMUkQPGu8n7', 'input': {}, 'name': 'document_retriever_xxx', 'type': 'tool_use', 'index': 1}]
+        token = ""
+    return token
+
+class AgentHandler(AsyncCallbackHandler):
+
+    def __init__(self, queue: Queue, llm:str, threadId: str = None) -> None:
+        super().__init__()
+        self._threadId = threadId
+        self.json_block = ""
+        self.is_json_block = False
+        self.backtick_count = 0  # Conteggio dei backticks per il controllo accurato
+        self.queue = queue
+        self.llm = llm
+
+    async def on_llm_start(
+        self,
+        serialized: Dict[str, Any],
+        prompts: List[str],
+        *,
+        run_id: UUID,
+        parent_run_id: UUID = None,
+        tags: List[str] = None,
+        metadata: Dict[str, Any] = None,
+        **kwargs: Any,
+    ) -> None:
+        firstChunk = {
+            "type": "info",
+            "threadId": self._threadId,
+        }
+        await self.queue.put(printString(firstChunk))
+
+    """async def on_chat_model_start(self, serialized: Dict[str, Any], messages: List[List[BaseMessage]], *, run_id: UUID = None, parent_run_id = None, tags = None, metadata = None, **kwargs: Any) -> Any:
+        pass"""
+
+    async def on_tool_end(self, output: Any, *, run_id: UUID, parent_run_id: UUID = None, tags: List[str] = None, **kwargs: Any) -> None:
+      pass
+
+    async def on_llm_new_token(
+        self,
+        token: str,
+        *,
+        chunk: Optional[Union[GenerationChunk, ChatGenerationChunk]] = None,
+        run_id: UUID,
+        parent_run_id: Optional[UUID] = None,
+        tags: Optional[List[str]] = None,
+        **kwargs: Any,
+    ) -> None:
+        """Gestisce i nuovi token durante lo streaming."""
+
+        if token:
+            token = _parse_token(self.llm,token)
+            if token:
+              self.backtick_count += token.count("`")
+
+              if self.backtick_count >= 3:
+                  if not self.is_json_block:
+                      self.is_json_block = True
+                      self.json_block = ""
+                  else:
+                      self.is_json_block = False
+                      self.json_block += token.replace("```json", '')
+                      await self.process_json_block(self.json_block)
+                      self.json_block = ""
+                  self.backtick_count = 0
+              elif self.is_json_block:
+                  self.json_block += token
+              else:
+                  await self.queue.put(printString(token))
+
+    async def on_agent_finish(
+        self,
+        finish: AgentFinish,
+        *,
+        run_id: UUID,
+        parent_run_id: UUID = None,
+        tags: List[str] = None,
+        **kwargs: Any,
+    ) -> None:
+        settings.chat_history.extend(
+            [
+                AIMessage(content=_parse_token(self.llm,finish.return_values["output"])),
+            ]
+        )
+        finalChunk = {"type": "end"}
+        await self.queue.put(printJson(finalChunk))
+        await self.queue.put(None)
+
+    async def process_json_block(self, json_block: str):
+      """Processa il blocco JSON completo."""
+      # Rimuove il delimitatore iniziale '```json' se presente, e spazi vuoti
+      json_block_clean = json_block.replace('```', '').replace('json', '').strip()
+      # Verifica che il blocco non sia vuoto prima di tentare il parsing
+      if json_block_clean:
+          try:
+              # Prova a fare il parsing del JSON
+              parsed_json = json.loads(json_block_clean)
+              await self.queue.put(printJson(parsed_json))
+          except json.JSONDecodeError as e:
+              # Se il JSON è malformato, logga l'errore
+              raise e
+
+class RawAgentHandler(AsyncCallbackHandler):
+
+    def __init__(self,queue: Queue, llm: str) -> None:
+        super().__init__()
+        self.queue = queue
+        self.llm = llm
+    async def on_llm_start(
+        self,
+        serialized: Dict[str, Any],
+        prompts: List[str],
+        *,
+        run_id: UUID,
+        parent_run_id: UUID = None,
+        tags: List[str] = None,
+        metadata: Dict[str, Any] = None,
+        **kwargs: Any,
+    ) -> None:
+        pass
+
+    """async def on_chat_model_start(self, serialized: Dict[str, Any], messages: List[List[BaseMessage]], *, run_id: UUID = None, parent_run_id = None, tags = None, metadata = None, **kwargs: Any) -> Any:
+        pass"""
+
+    async def on_tool_end(self, output: Any, *, run_id: UUID, parent_run_id: UUID = None, tags: List[str] = None, **kwargs: Any) -> None:
+      pass
+
+    async def on_llm_new_token(
+        self,
+        token: str,
+        *,
+        chunk: Optional[Union[GenerationChunk, ChatGenerationChunk]] = None,
+        run_id: UUID,
+        parent_run_id: Optional[UUID] = None,
+        tags: Optional[List[str]] = None,
+        **kwargs: Any,
+    ) -> None:
+        """Handles new tokens during streaming."""
+        if token:  # Only process non-empty tokens
+            await self.queue.put(_parse_token(self.llm,token))
+
+    async def on_agent_finish(
+        self,
+        finish: AgentFinish,
+        *,
+        run_id: UUID,
+        parent_run_id: UUID = None,
+        tags: List[str] = None,
+        **kwargs: Any,
+    ) -> None:
+        settings.chat_history.extend(
+            [
+                AIMessage(content=_parse_token(self.llm,finish.return_values["output"]))
+            ]
+        )
+        await self.queue.put(None)