🐛 Fix functions

next/src/components/AppTitle.tsx

@@ -8,7 +8,7 @@ const AppTitle = () => {
 
   return (
     <div id="title" className="relative flex flex-col items-center font-mono">
-      <div className="flex flex-row items-start shadow-2xl">
+      <div className="flex flex-row items-start">
         <span className="text-4xl font-bold text-[#C0C0C0] xs:text-5xl sm:text-6xl">Agent</span>
         <span className="text-4xl font-bold text-white xs:text-5xl sm:text-6xl">GPT</span>
         <PopIn delay={0.5}>
@@ -31,4 +31,4 @@ const AppTitle = () => {
   );
 };
 
-export default AppTitle;
+export default AppTitle;

next/src/stores/agentStore.ts

@@ -67,6 +67,9 @@ export const useAgentStore = createSelectors(
       {
         name: "agent-storage-v2",
         storage: createJSONStorage(() => localStorage),
+        partialize: (state) => ({
+          tools: state.tools,
+        }),
       }
     )
   )

platform/reworkd_platform/web/api/agent/agent_service/open_ai_agent_service.py

@@ -7,7 +7,7 @@
 from loguru import logger
 
 from reworkd_platform.web.api.agent.agent_service.agent_service import AgentService
-from reworkd_platform.web.api.agent.analysis import Analysis
+from reworkd_platform.web.api.agent.analysis import Analysis, AnalysisArguments
 from reworkd_platform.web.api.agent.helpers import (
     call_model_with_handling,
     openai_error_handler,
@@ -19,9 +19,11 @@
     start_goal_prompt,
 )
 from reworkd_platform.web.api.agent.task_output_parser import TaskOutputParser
-from reworkd_platform.web.api.agent.tools.open_ai_function import analysis_function
+from reworkd_platform.web.api.agent.tools.open_ai_function import get_tool_function
 from reworkd_platform.web.api.agent.tools.tools import (
+    get_default_tool,
     get_tool_from_name,
+    get_tool_name,
     get_user_tools,
 )
 from reworkd_platform.web.api.errors import OpenAIError
@@ -67,15 +69,19 @@ async def analyze_task_agent(
                 task=task,
                 language=self.language,
             ).to_messages(),
-            functions=[analysis_function(get_user_tools(tool_names))],
+            functions=list(map(get_tool_function, get_user_tools(tool_names))),
         )
 
         function_call = message.additional_kwargs.get("function_call", {})
         completion = function_call.get("arguments", "")
 
         try:
-            pydantic_parser = PydanticOutputParser(pydantic_object=Analysis)
-            return parse_with_handling(pydantic_parser, completion)
+            pydantic_parser = PydanticOutputParser(pydantic_object=AnalysisArguments)
+            analysis_arguments = parse_with_handling(pydantic_parser, completion)
+            return Analysis(
+                action=function_call.get("name", get_tool_name(get_default_tool())),
+                **analysis_arguments.dict(),
+            )
         except OpenAIError:
             return Analysis.get_default_analysis()
 

platform/reworkd_platform/web/api/agent/analysis.py

@@ -1,11 +1,18 @@
 from pydantic import BaseModel, validator
 
 
-class Analysis(BaseModel):
+class AnalysisArguments(BaseModel):
+    """
+    Arguments for the analysis function of a tool. OpenAI functions will resolve these values but leave out the action.
+    """
+
     reasoning: str
-    action: str
     arg: str
 
+
+class Analysis(AnalysisArguments):
+    action: str
+
     @validator("action")
     def action_must_be_valid_tool(cls, v: str) -> str:
         # TODO: Remove circular import

platform/reworkd_platform/web/api/agent/model_settings.py

@@ -28,7 +28,7 @@ def create_model(
         max_tokens=model_settings.max_tokens,
         streaming=streaming,
         max_retries=5,
-        user=user.email,
+        model_kwargs={"user": user.email},
     )
 
 

platform/reworkd_platform/web/api/agent/prompts.py

@@ -24,10 +24,11 @@
     High level objective: "{goal}"
     Current task: "{task}"
 
-    Based on this information, use the "analyze" function to return an object for what specific 'tool' to use.
-    Select the correct tool by being smart and efficient. Provide concrete reasoning for the tool choice detailing 
-    your overall plan and any concerns you may have. Your reasoning should be no more than three sentences.
-    Ensure "reasoning" and only "reasoning" is in the {language} language.
+    Based on this information, use the best function to make progress or accomplish the task entirely.
+    Select the correct function by being smart and efficient. Ensure "reasoning" and only "reasoning" is in the 
+    {language} language.
+
+    Note you MUST select a function.
     """,
     input_variables=["goal", "task", "language"],
 )
@@ -57,8 +58,8 @@
     the following overall objective `{goal}` and the following sub-task, `{task}`.
 
     Perform the task by understanding the problem, extracting variables, and being smart
-    and efficient. Provide a descriptive response, make decisions yourself when
-    confronted with choices and provide reasoning for ideas / decisions.
+    and efficient. Write a detailed response that address the task.
+    When confronted with choices, make a decision yourself with reasoning.
     """,
     input_variables=["goal", "language", "task"],
 )

platform/reworkd_platform/web/api/agent/tools/code.py

@@ -6,10 +6,7 @@
 
 
 class Code(Tool):
-    description = (
-        "Useful for writing, reviewing, and refactoring code. Can also fix bugs, "
-        "and explain programming concepts."
-    )
+    description = "Should only be used to write code, refactor code, fix code bugs, and explain programming concepts."
     public_description = "Write and review code."
 
     async def call(

platform/reworkd_platform/web/api/agent/tools/image.py

@@ -50,12 +50,13 @@ async def get_open_ai_image(input_str: str) -> str:
 
 
 class Image(Tool):
-    description = (
-        "Used to sketch, draw, or generate an image. The input string "
-        "should be a detailed description of the image touching on image "
-        "style, image focus, color, etc"
-    )
+    description = "Used to sketch, draw, or generate an image."
     public_description = "Generate AI images."
+    arg_description = (
+        "The input prompt to the image generator. "
+        "This should be a detailed description of the image touching on image "
+        "style, image focus, color, etc."
+    )
 
     async def call(
         self, goal: str, task: str, input_str: str

platform/reworkd_platform/web/api/agent/tools/open_ai_function.py

@@ -1,4 +1,4 @@
-from typing import List, Type, TypedDict
+from typing import Type, TypedDict
 
 from reworkd_platform.web.api.agent.tools.tool import Tool
 from reworkd_platform.web.api.agent.tools.tools import get_tool_name
@@ -15,37 +15,29 @@ class FunctionDescription(TypedDict):
     """The parameters of the function."""
 
 
-def analysis_function(tools: List[Type[Tool]]) -> FunctionDescription:
-    """A function that will return the tool specifications from OpenAI"""
-    tool_names = [get_tool_name(tool) for tool in tools]
-    tool_name_to_description = [
-        f"{get_tool_name(tool)}: {tool.description}" for tool in tools
-    ]
+def get_tool_function(tool: Type[Tool]) -> FunctionDescription:
+    """A function that will return the tool's function specification"""
+    name = get_tool_name(tool)
 
     return {
-        "name": "analysis",
-        "description": (
-            "Return an object for what specific 'action'/'tool' to call based on their descriptions:\n"
-            f"{tool_name_to_description}"
-        ),
+        "name": name,
+        "description": tool.description,
         "parameters": {
             "type": "object",
             "properties": {
                 "reasoning": {
                     "type": "string",
                     "description": (
-                        f"You must use one of the tools available to you: {tool_names}"
-                        "This reasoning should be how you will accomplish the task with the provided action."
+                        f"Reasoning is how the task will be accomplished with the current function. "
                         "Detail your overall plan along with any concerns you have."
                         "Ensure this reasoning value is in the user defined langauge "
                     ),
                 },
-                "action": {"type": "string", "enum": tool_names},
                 "arg": {
                     "type": "string",
-                    "description": "The appropriate action argument based on the action type",
+                    "description": tool.arg_description,
                 },
             },
-            "required": ["reasoning", "action", "arg"],
+            "required": ["reasoning", "arg"],
         },
     }

platform/reworkd_platform/web/api/agent/tools/search.py

@@ -36,9 +36,9 @@ class Search(Tool):
     description = (
         "Search Google for short up to date searches for simple questions "
         "news and people.\n"
-        "The input arg should be the search query. Ensure this value is NOT empty."
     )
     public_description = "Search google for information about current events."
+    arg_description = "The search query. Ensure this value is NOT empty."
 
     @staticmethod
     def available() -> bool:

platform/reworkd_platform/web/api/agent/tools/tool.py

@@ -7,6 +7,8 @@
 class Tool(ABC):
     description: str = ""
     public_description: str = ""
+    arg_description: str = "Always leave as an empty string"
+
     model: BaseChatModel
     language: str
 

platform/reworkd_platform/web/api/agent/tools/utils.py

@@ -13,6 +13,12 @@ class CitedSnippet:
     text: str
     url: str
 
+    def __repr__(self):
+        """
+        The string representation the AI model will see
+        """
+        return f"{{i: {self.index}, text: {self.text}, url: {self.url}}}"
+
 
 def summarize(
     model: BaseChatModel,
@@ -25,6 +31,11 @@ def summarize(
 
     chain = LLMChain(llm=model, prompt=summarize_prompt)
 
+    print(
+        summarize_prompt.format_prompt(
+            goal=goal, query=query, snippets=snippets, language=language
+        )
+    )
     return StreamingResponse.from_chain(
         chain,
         {

platform/reworkd_platform/web/api/agent/tools/wikipedia_search.py

@@ -9,15 +9,16 @@ class Wikipedia(Tool):
     description = (
         "Search Wikipedia for information about historical people, companies, events, "
         "places or research. This should be used over search for broad overviews of "
-        "specific nouns.\n The argument should be a simple query of just the noun."
+        "specific nouns."
     )
     public_description = "Search Wikipedia for historical information."
+    arg_description = "A simple query string of just the noun in question."
 
     async def call(self, goal: str, task: str, input_str: str) -> StreamingResponse:
         wikipedia_client = WikipediaAPIWrapper(
             wiki_client=None,  # Meta private value but mypy will complain its missing
         )
-        
+
         # TODO: Make the below async
         wikipedia_search = wikipedia_client.run(input_str)
         # return summarize(self.model, self.language, goal, task, [wikipedia_search])