upgrade: add web search! (run-llama#40)

1_🏠_Home.py

@@ -31,6 +31,8 @@
     "To build a new agent, please make sure that 'Create a new agent' is selected.",
     icon="ℹ️",
 )
+if "metaphor_key" in st.secrets:
+    st.info("**NOTE**: The ability to add web search is enabled.")
 
 
 add_sidebar()

agent_utils.py

@@ -32,6 +32,9 @@
 from constants import AGENT_CACHE_DIR
 import shutil
 
+from llama_index.callbacks import CallbackManager
+from callback_manager import StreamlitFunctionsCallbackHandler
+
 
 def _resolve_llm(llm_str: str) -> LLM:
     """Resolve LLM."""
@@ -153,9 +156,25 @@ def load_agent(
     """Load agent."""
     extra_kwargs = extra_kwargs or {}
     if isinstance(llm, OpenAI) and is_function_calling_model(llm.model):
+        # TODO: use default msg handler
+        # TODO: separate this from agent_utils.py...
+        def _msg_handler(msg: str) -> None:
+            """Message handler."""
+            st.info(msg)
+            st.session_state.agent_messages.append(
+                {"role": "assistant", "content": msg, "msg_type": "info"}
+            )
+
+        # add streamlit callbacks (to inject events)
+        handler = StreamlitFunctionsCallbackHandler(_msg_handler)
+        callback_manager = CallbackManager([handler])
         # get OpenAI Agent
         agent: BaseChatEngine = OpenAIAgent.from_tools(
-            tools=tools, llm=llm, system_prompt=system_prompt, **kwargs
+            tools=tools,
+            llm=llm,
+            system_prompt=system_prompt,
+            **kwargs,
+            callback_manager=callback_manager,
         )
     else:
         if "vector_index" not in extra_kwargs:
@@ -189,8 +208,12 @@ def load_meta_agent(
     extra_kwargs = extra_kwargs or {}
     if isinstance(llm, OpenAI) and is_function_calling_model(llm.model):
         # get OpenAI Agent
+
         agent: BaseAgent = OpenAIAgent.from_tools(
-            tools=tools, llm=llm, system_prompt=system_prompt, **kwargs
+            tools=tools,
+            llm=llm,
+            system_prompt=system_prompt,
+            **kwargs,
         )
     else:
         agent = ReActAgent.from_tools(
@@ -285,6 +308,66 @@ def construct_agent(
     return agent, extra_info
 
 
+def get_web_agent_tool() -> QueryEngineTool:
+    """Get web agent tool.
+
+    Wrap with our load and search tool spec.
+
+    """
+    from llama_hub.tools.metaphor.base import MetaphorToolSpec
+
+    # TODO: set metaphor API key
+    metaphor_tool = MetaphorToolSpec(
+        api_key=st.secrets.metaphor_key,
+    )
+    metaphor_tool_list = metaphor_tool.to_tool_list()
+
+    # TODO: LoadAndSearch doesn't work yet
+    # The search_and_retrieve_documents tool is the third in the tool list,
+    # as seen above
+    # wrapped_retrieve = LoadAndSearchToolSpec.from_defaults(
+    #     metaphor_tool_list[2],
+    # )
+
+    # NOTE: requires openai right now
+    # We don't give the Agent our unwrapped retrieve document tools
+    # instead passing the wrapped tools
+    web_agent = OpenAIAgent.from_tools(
+        # [*wrapped_retrieve.to_tool_list(), metaphor_tool_list[4]],
+        metaphor_tool_list,
+        llm=BUILDER_LLM,
+        verbose=True,
+    )
+
+    # return agent as a tool
+    # TODO: tune description
+    web_agent_tool = QueryEngineTool.from_defaults(
+        web_agent,
+        name="web_agent",
+        description="""
+            This agent can answer questions by searching the web. \
+Use this tool if the answer is ONLY likely to be found by searching \
+the internet, especially for queries about recent events.
+        """,
+    )
+
+    return web_agent_tool
+
+
+def get_tool_objects(tool_names: List[str]) -> List:
+    """Get tool objects from tool names."""
+    # construct additional tools
+    tool_objs = []
+    for tool_name in tool_names:
+        if tool_name == "web_search":
+            # build web agent
+            tool_objs.append(get_web_agent_tool())
+        else:
+            raise ValueError(f"Tool {tool_name} not recognized.")
+
+    return tool_objs
+
+
 class ParamCache(BaseModel):
     """Cache for RAG agent builder.
 
@@ -338,7 +421,7 @@ def save_to_disk(self, save_dir: str) -> None:
             "file_names": self.file_names,
             "urls": self.urls,
             # TODO: figure out tools
-            # "tools": [],
+            "tools": self.tools,
             "rag_params": self.rag_params.dict(),
             "agent_id": self.agent_id,
         }
@@ -376,11 +459,13 @@ def load_from_disk(
             file_names=cache_dict["file_names"], urls=cache_dict["urls"]
         )
         # load agent from index
+        additional_tools = get_tool_objects(cache_dict["tools"])
         agent, _ = construct_agent(
             cache_dict["system_prompt"],
             cache_dict["rag_params"],
             cache_dict["docs"],
             vector_index=vector_index,
+            additional_tools=additional_tools,
             # TODO: figure out tools
         )
         cache_dict["vector_index"] = vector_index
@@ -505,20 +590,14 @@ def load_data(
         self._cache.urls = urls
         return "Data loaded successfully."
 
-    # NOTE: unused
     def add_web_tool(self) -> str:
         """Add a web tool to enable agent to solve a task."""
         # TODO: make this not hardcoded to a web tool
         # Set up Metaphor tool
-        from llama_hub.tools.metaphor.base import MetaphorToolSpec
-
-        # TODO: set metaphor API key
-        metaphor_tool = MetaphorToolSpec(
-            api_key=os.environ["METAPHOR_API_KEY"],
-        )
-        metaphor_tool_list = metaphor_tool.to_tool_list()
-
-        self._cache.tools.extend(metaphor_tool_list)
+        if "web_search" in self._cache.tools:
+            return "Web tool already added."
+        else:
+            self._cache.tools.append("web_search")
         return "Web tool added successfully."
 
     def get_rag_params(self) -> Dict:
@@ -557,11 +636,13 @@ def create_agent(self, agent_id: Optional[str] = None) -> str:
         if self._cache.system_prompt is None:
             raise ValueError("Must set system prompt before creating agent.")
 
+        # construct additional tools
+        additional_tools = get_tool_objects(self.cache.tools)
         agent, extra_info = construct_agent(
             cast(str, self._cache.system_prompt),
             cast(RAGParams, self._cache.rag_params),
             self._cache.docs,
-            additional_tools=self._cache.tools,
+            additional_tools=additional_tools,
         )
 
         # if agent_id not specified, randomly generate one
@@ -587,6 +668,7 @@ def update_agent(
         chunk_size: Optional[int] = None,
         embed_model: Optional[str] = None,
         llm: Optional[str] = None,
+        additional_tools: Optional[List] = None,
     ) -> None:
         """Update agent.
 
@@ -609,7 +691,6 @@ def update_agent(
         # We call set_rag_params and create_agent, which will
         # update the cache
         # TODO: decouple functions from tool functions exposed to the agent
-
         rag_params_dict: Dict[str, Any] = {}
         if include_summarization is not None:
             rag_params_dict["include_summarization"] = include_summarization
@@ -623,6 +704,11 @@ def update_agent(
             rag_params_dict["llm"] = llm
 
         self.set_rag_params(**rag_params_dict)
+
+        # update tools
+        if additional_tools is not None:
+            self.cache.tools = additional_tools
+
         # this will update the agent in the cache
         self.create_agent()
 
@@ -655,6 +741,33 @@ def update_agent(
 # please make sure to update the LLM above if you change the function below
 
 
+def _get_builder_agent_tools(agent_builder: RAGAgentBuilder) -> List[FunctionTool]:
+    """Get list of builder agent tools to pass to the builder agent."""
+    # see if metaphor api key is set, otherwise don't add web tool
+    # TODO: refactor this later
+
+    if "metaphor_key" in st.secrets:
+        fns: List[Callable] = [
+            agent_builder.create_system_prompt,
+            agent_builder.load_data,
+            agent_builder.add_web_tool,
+            agent_builder.get_rag_params,
+            agent_builder.set_rag_params,
+            agent_builder.create_agent,
+        ]
+    else:
+        fns = [
+            agent_builder.create_system_prompt,
+            agent_builder.load_data,
+            agent_builder.get_rag_params,
+            agent_builder.set_rag_params,
+            agent_builder.create_agent,
+        ]
+
+    fn_tools: List[FunctionTool] = [FunctionTool.from_defaults(fn=fn) for fn in fns]
+    return fn_tools
+
+
 # define agent
 # @st.cache_resource
 def load_meta_agent_and_tools(
@@ -664,15 +777,7 @@ def load_meta_agent_and_tools(
     # think of this as tools for the agent to use
     agent_builder = RAGAgentBuilder(cache)
 
-    fns: List[Callable] = [
-        agent_builder.create_system_prompt,
-        agent_builder.load_data,
-        # add_web_tool,
-        agent_builder.get_rag_params,
-        agent_builder.set_rag_params,
-        agent_builder.create_agent,
-    ]
-    fn_tools = [FunctionTool.from_defaults(fn=fn) for fn in fns]
+    fn_tools = _get_builder_agent_tools(agent_builder)
 
     builder_agent = load_meta_agent(
         fn_tools, llm=BUILDER_LLM, system_prompt=RAG_BUILDER_SYS_STR, verbose=True

callback_manager.py

@@ -0,0 +1,70 @@
+"""Streaming callback manager."""
+from llama_index.callbacks.base_handler import BaseCallbackHandler
+from llama_index.callbacks.schema import CBEventType
+
+from typing import Optional, Dict, Any, List, Callable
+
+STORAGE_DIR = "./storage"  # directory to cache the generated index
+DATA_DIR = "./data"  # directory containing the documents to index
+
+
+class StreamlitFunctionsCallbackHandler(BaseCallbackHandler):
+    """Callback handler that outputs streamlit components given events."""
+
+    def __init__(self, msg_handler: Callable[[str], Any]) -> None:
+        """Initialize the base callback handler."""
+        self.msg_handler = msg_handler
+        super().__init__([], [])
+
+    def on_event_start(
+        self,
+        event_type: CBEventType,
+        payload: Optional[Dict[str, Any]] = None,
+        event_id: str = "",
+        parent_id: str = "",
+        **kwargs: Any,
+    ) -> str:
+        """Run when an event starts and return id of event."""
+        if event_type == CBEventType.FUNCTION_CALL:
+            if payload is None:
+                raise ValueError("Payload cannot be None")
+            arguments_str = payload["function_call"]
+            tool_str = payload["tool"].name
+            print_str = f"Calling function: {tool_str} with args: {arguments_str}\n\n"
+            self.msg_handler(print_str)
+        else:
+            pass
+        return event_id
+
+    def on_event_end(
+        self,
+        event_type: CBEventType,
+        payload: Optional[Dict[str, Any]] = None,
+        event_id: str = "",
+        **kwargs: Any,
+    ) -> None:
+        """Run when an event ends."""
+        pass
+        # TODO: currently we don't need to do anything here
+        # if event_type == CBEventType.FUNCTION_CALL:
+        #     response = payload["function_call_response"]
+        #     # Add this to queue
+        #     print_str = (
+        #         f"\n\nGot output: {response}\n"
+        #         "========================\n\n"
+        #     )
+        # elif event_type == CBEventType.AGENT_STEP:
+        #     # put response into queue
+        #     self._queue.put(payload["response"])
+
+    def start_trace(self, trace_id: Optional[str] = None) -> None:
+        """Run when an overall trace is launched."""
+        pass
+
+    def end_trace(
+        self,
+        trace_id: Optional[str] = None,
+        trace_map: Optional[Dict[str, List[str]]] = None,
+    ) -> None:
+        """Run when an overall trace is exited."""
+        pass

pages/2_⚙️_RAG_Config.py

@@ -24,6 +24,7 @@ def update_agent() -> None:
         "config_agent_builder" in st.session_state.keys()
         and st.session_state.config_agent_builder is not None
     ):
+        additional_tools = st.session_state.additional_tools_st.split(",")
         agent_builder = cast(RAGAgentBuilder, st.session_state.config_agent_builder)
         ### Update the agent
         agent_builder.update_agent(
@@ -34,6 +35,7 @@ def update_agent() -> None:
             chunk_size=st.session_state.chunk_size_st,
             embed_model=st.session_state.embed_model_st,
             llm=st.session_state.llm_st,
+            additional_tools=additional_tools,
         )
 
         # Update Radio Buttons: update selected agent to the new id
@@ -114,6 +116,14 @@ def delete_agent() -> None:
         value=rag_params.include_summarization,
         key="include_summarization_st",
     )
+
+    # add web tool
+    additional_tools_st = st.text_input(
+        "Additional tools (currently only supports 'web_search')",
+        value=",".join(agent_builder.cache.tools),
+        key="additional_tools_st",
+    )
+
     top_k_st = st.number_input("Top K", value=rag_params.top_k, key="top_k_st")
     chunk_size_st = st.number_input(
         "Chunk Size", value=rag_params.chunk_size, key="chunk_size_st"