[BUFG][Reliable Tool Usage]

README.md

@@ -447,46 +447,29 @@ A Plug in and play conversational agent with `GPT4`, `Mixytral`, or any of our m
 - Reliable, this simple system will always provide responses you want.
 
 ```python
-import os
-
-from dotenv import load_dotenv
+from swarms import Agent, Anthropic
 
-from swarms import Conversation, OpenAIChat
 
-conv = Conversation(
-    time_enabled=True,
+## Initialize the workflow
+agent = Agent(
+    agent_name="Transcript Generator",
+    agent_description=(
+        "Generate a transcript for a youtube video on what swarms"
+        " are!"
+    ),
+    llm=Anthropic(),
+    max_loops=3,
+    autosave=True,
+    dashboard=False,
+    streaming_on=True,
+    verbose=True,
+    stopping_token="<DONE>",
+    interactive=True, # Set to True
 )
 
-# Load the environment variables
-load_dotenv()
-
-# Get the API key from the environment
-api_key = os.environ.get("OPENAI_API_KEY")
+# Run the workflow on a task
+agent("Generate a transcript for a youtube video on what swarms are!")
 
-# Initialize the language model
-llm = OpenAIChat(openai_api_key=api_key, model_name="gpt-4")
-
-
-# Run the language model in a loop
-def interactive_conversation(llm):
-    conv = Conversation()
-    while True:
-        user_input = input("User: ")
-        conv.add("user", user_input)
-        if user_input.lower() == "quit":
-            break
-        task = conv.return_history_as_string()  # Get the conversation history
-        out = llm(task)
-        conv.add("assistant", out)
-        print(
-            f"Assistant: {out}",
-        )
-    conv.display_conversation()
-    conv.export_conversation("conversation.txt")
-
-
-# Replace with your LLM instance
-interactive_conversation(llm)
 ```
 
 
@@ -1029,6 +1012,90 @@ autoswarm.run("Analyze these financial data and give me a summary")
 
 ```
 
+## `AgentRearrange`
+Inspired by Einops and einsum, this orchestration techniques enables you to map out the relationships between various agents. For example you specify linear and sequential relationships like `a -> a1 -> a2 -> a3` or concurrent relationships where the first agent will send a message to 3 agents all at once: `a -> a1, a2, a3`. You can customize your workflow to mix sequential and concurrent relationships
+
+```python
+from swarms import Agent, Anthropic, AgentRearrange, 
+
+## Initialize the workflow
+agent = Agent(
+    agent_name="t",
+    agent_description=(
+        "Generate a transcript for a youtube video on what swarms"
+        " are!"
+    ),
+    system_prompt=(
+        "Generate a transcript for a youtube video on what swarms"
+        " are!"
+    ),
+    llm=Anthropic(),
+    max_loops=1,
+    autosave=True,
+    dashboard=False,
+    streaming_on=True,
+    verbose=True,
+    stopping_token="<DONE>",
+)
+
+agent2 = Agent(
+    agent_name="t1",
+    agent_description=(
+        "Generate a transcript for a youtube video on what swarms"
+        " are!"
+    ),
+    llm=Anthropic(),
+    max_loops=1,
+    system_prompt="Summarize the transcript",
+    autosave=True,
+    dashboard=False,
+    streaming_on=True,
+    verbose=True,
+    stopping_token="<DONE>",
+)
+
+agent3 = Agent(
+    agent_name="t2",
+    agent_description=(
+        "Generate a transcript for a youtube video on what swarms"
+        " are!"
+    ),
+    llm=Anthropic(),
+    max_loops=1,
+    system_prompt="Finalize the transcript",
+    autosave=True,
+    dashboard=False,
+    streaming_on=True,
+    verbose=True,
+    stopping_token="<DONE>",
+)
+
+
+# Rearrange the agents
+rearrange = AgentRearrange(
+    agents=[agent, agent2, agent3],
+    verbose=True,
+    # custom_prompt="Summarize the transcript",
+)
+
+# Run the workflow on a task
+results = rearrange(
+    # pattern="t -> t1, t2 -> t2",
+    pattern="t -> t1 -> t2",
+    default_task=(
+        "Generate a transcript for a YouTube video on what swarms"
+        " are!"
+    ),
+    t="Generate a transcript for a YouTube video on what swarms are!",
+    # t2="Summarize the transcript",
+    # t3="Finalize the transcript",
+)
+# print(results)
+
+
+```
+
+
 ---
 
 ## Documentation

devin.py

@@ -7,11 +7,13 @@ Plan -> act in a loop until observation is met
 - Text Editor
 - Browser
 """
-from swarms import Agent, OpenAIChat, tool
+from swarms import Agent, Anthropic, tool
 import subprocess
 
 # Model
-llm = OpenAIChat()
+llm = Anthropic(
+    temperature=0.1,
+)
 
 
 # Tools
@@ -37,7 +39,7 @@ def terminal(
 @tool
 def browser(query: str):
     """
-    Search the query in the browser.
+    Search the query in the browser with the `browser` tool.
 
     Args:
         query (str): The query to search in the browser.
@@ -58,10 +60,10 @@ agent = Agent(
     system_prompt=(
         "Autonomous agent that can interact with humans and other"
         " agents. Be Helpful and Kind. Use the tools provided to"
-        " assist the user."
+        " assist the user. Return all code in markdown format."
     ),
     llm=llm,
-    max_loops=4,
+    max_loops="auto",
     autosave=True,
     dashboard=False,
     streaming_on=True,
@@ -73,5 +75,5 @@ agent = Agent(
 )
 
 # Run the agent
-out = agent("What is the weather today in palo alto?")
+out = agent("What is the weather today in palo alto use the browser tool to search for the weather?")
 print(out)

pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "poetry.core.masonry.api"
 
 [tool.poetry]
 name = "swarms"
-version = "4.7.3"
+version = "4.7.4"
 description = "Swarms - Pytorch"
 license = "MIT"
 authors = ["Kye Gomez <kye@apac.ai>"]

swarms/agents/worker_agent.py

@@ -138,16 +138,13 @@ class Worker(Agent):
                 ai_role=self.role,
                 tools=self.tools,
                 llm=self.llm,
-                memory=self.vectorstore.as_retriever(
-                    search_kwargs=self.search_kwargs
-                ),
+                # memory = None,
                 human_in_the_loop=self.human_in_the_loop,
             )
 
         except Exception as error:
             raise RuntimeError(f"Error setting up agent: {error}")
 
-    # @log_decorator
     @error_decorator
     @timing_decorator
     def run(self, task: str = None, *args, **kwargs):
@@ -166,7 +163,6 @@ class Worker(Agent):
         except Exception as error:
             raise RuntimeError(f"Error while running agent: {error}")
 
-    # @log_decorator
     @error_decorator
     @timing_decorator
     def __call__(self, task: str = None, *args, **kwargs):

swarms/structs/agent.py

@@ -24,7 +24,8 @@ from swarms.utils.code_interpreter import SubprocessCodeInterpreter
 from swarms.utils.data_to_text import data_to_text
 from swarms.utils.parse_code import extract_code_from_markdown
 from swarms.utils.pdf_to_text import pdf_to_text
-
+from swarms.tools.exec_tool import execute_tool_by_name
+from swarms.tools.function_util import process_tool_docs
 
 # Utils
 # Custom stopping condition
@@ -320,8 +321,17 @@ class Agent:
                 memory=self.short_memory.return_history_as_string(),
             )
 
-            # Append the tools prompt to the sop
-            self.sop = f"{self.sop}\n{tools_prompt}"
+            # Append the tools prompt to the short_term_memory
+            self.short_memory.add(
+                role=self.agent_name, content=tools_prompt
+            )
+            
+            # And, add the tool documentation to the memory
+            for tool in self.tools:
+                tool_docs = process_tool_docs(tool)
+                self.short_memory.add(
+                    role=self.agent_name, content=tool_docs
+                )
 
         # If the long term memory is provided then set the long term memory prompt
 
@@ -607,6 +617,19 @@ class Agent:
                             role=self.agent_name, content=response
                         )
 
+                        if self.tools:
+                            # Extract code from markdown
+                            response = extract_code_from_markdown(
+                                response
+                            )
+
+                            # Execute the tool by name
+                            execute_tool_by_name(
+                                response,
+                                self.tools,
+                                stop_token=self.stopping_token,
+                            )
+
                         if self.code_interpreter:
                             extracted_code = (
                                 extract_code_from_markdown(response)
@@ -676,8 +699,7 @@ class Agent:
 
                 # Check stopping conditions
                 if (
-                    self.stopping_token
-                    and self.stopping_token in response
+                    self.stopping_token in response
                 ):
                     break
                 elif (
@@ -685,13 +707,13 @@ class Agent:
                     and self._check_stopping_condition(response)
                 ):
                     break
-                elif self.stopping_func and self.stopping_func(
+                elif self.stopping_func is not None and self.stopping_func(
                     response
                 ):
                     break
 
                 if self.interactive:
-                    user_input = input("You: ")
+                    user_input = colored(input("You: "), "red")
 
                     # User-defined exit command
                     if (

swarms/tools/exec_tool.py

@@ -1,4 +1,5 @@
 import json
+import concurrent.futures
 import re
 from abc import abstractmethod
 from typing import Dict, List, NamedTuple
@@ -101,7 +102,46 @@ def execute_tool_by_name(
     if action.name in tools:
         tool = tools[action.name]
         try:
-            observation = tool.run(action.args)
+            # Check if multiple tools are used
+            tool_names = [name for name in tools if name in text]
+            if len(tool_names) > 1:
+                # Execute tools concurrently
+                with concurrent.futures.ThreadPoolExecutor() as executor:
+                    futures = []
+                    for tool_name in tool_names:
+                        futures.append(
+                            executor.submit(
+                                tools[tool_name].run, action.args
+                            )
+                        )
+
+                    # Wait for all futures to complete
+                    concurrent.futures.wait(futures)
+
+                    # Get results from completed futures
+                    results = [
+                        future.result()
+                        for future in futures
+                        if future.done()
+                    ]
+
+                    # Process results
+                    for result in results:
+                        # Handle errors
+                        if isinstance(result, Exception):
+                            result = (
+                                f"Error: {str(result)},"
+                                f" {type(result).__name__}, args:"
+                                f" {action.args}"
+                            )
+                        # Handle successful execution
+                        else:
+                            result = (
+                                f"Command {tool.name} returned:"
+                                f" {result}"
+                            )
+            else:
+                observation = tool.run(action.args)
         except ValidationError as e:
             observation = (
                 f"Validation Error in args: {str(e)}, args:"
@@ -121,5 +161,4 @@ def execute_tool_by_name(
             "Please refer to the 'COMMANDS' list for available "
             "commands and only respond in the specified JSON format."
         )
-
     return result

swarms/tools/function_util.py

@@ -0,0 +1,25 @@
+import inspect
+
+def process_tool_docs(item):
+    """
+    Process the documentation for a given item.
+
+    Args:
+        item: The item to process the documentation for.
+
+    Returns:
+        metadata: The processed metadata containing the item's name, documentation, and source code.
+    """
+    # If item is an instance of a class, get its class
+    if not inspect.isclass(item) and hasattr(item, '__class__'):
+        item = item.__class__
+
+    doc = inspect.getdoc(item)
+    source = inspect.getsource(item)
+    is_class = inspect.isclass(item)
+    item_type = "Class Name" if is_class else "Function Name"
+    metadata = f"{item_type}: {item.__name__}\n\n"
+    if doc:
+        metadata += f"Documentation:\n{doc}\n\n"
+    metadata += f"\n{source}"
+    return metadata