Hi, are you using something like Autogen Studio to import the workflow file? Can you document where and how to use the output?

I got this issue after pulling the latest version (6.6.6)

Plz don't use dated gpt model pointers like gpt-4-1106-preview as those get outdated.

(See https://platform.openai.com/docs/models/gpt-4-turbo-and-gpt-4)

If you could put the agents in a agents.yaml file and include a task.yaml file, you'd be legendary

Hi, tnx first of all for doing what I was hoping somebody would do with a similar vision of building a meta agent framework.

How about adding some configurable reflection cycles for each agent and history management allowing to intervene anywhere to finetune steps taken and letting it continue from there?

Not a major issue. The tool is working perfectly fine but when I saw the videos, I realized that the JSON files are named according to the role where as the ones that my setup create are just agent_0, agent_1, agent_2 and so on. Is it as intended or am I missing something?

The various prompts actually used (the skills, the agents, etc etc) should be in at least a user config file so it can be edited and/or add some UI to edit and save.

And while on the topic, using 'Autogen Studio python skill' in the skill creation prompt gets better results than 'python skill', and giving an example showing the format (which autogen actually can (and does?) read for context) helps with ensuring the formatting is correct, and comments and arguments are correctly spec'ed

It's a best practice to use .gitignore to remove the pycache directories, which are both autocreated, and user specific (version, machine, etc etc)

Also, provide a sample config file to copy before first use, and remove the actual config from the repo, otherwise any git update balks due to file changes made to that config file.

81bffd9

you reverted the autogen db location back to your personal config.

PLEASE, fix the config.py problem. There are ways to make it work with streamlit.
The config.py should either be user changed and NOT in the repo, or in the repo, and get values from some other file, NOT in the repo. The moment the user changes the config.py, the next time (and every time) they gitpull, they have a problem they have to deal with.

Was just going back and forth with the agents and this occurred. Had a similar error earlier.

NameError: name 'url_content' is not defined
Traceback:
File "/home/zeus/miniconda3/envs/cloudspace/lib/python3.10/site-packages/streamlit/runtime/scriptrunner/script_runner.py", line 537, in _run_script
self._session_state.on_script_will_rerun(rerun_data.widget_states)
File "/home/zeus/miniconda3/envs/cloudspace/lib/python3.10/site-packages/streamlit/runtime/state/safe_session_state.py", line 68, in on_script_will_rerun
self._state.on_script_will_rerun(latest_widget_states)
File "/home/zeus/miniconda3/envs/cloudspace/lib/python3.10/site-packages/streamlit/runtime/state/session_state.py", line 500, in on_script_will_rerun
self._call_callbacks()
File "/home/zeus/miniconda3/envs/cloudspace/lib/python3.10/site-packages/streamlit/runtime/state/session_state.py", line 513, in _call_callbacks
self._new_widget_state.call_callback(wid)
File "/home/zeus/miniconda3/envs/cloudspace/lib/python3.10/site-packages/streamlit/runtime/state/session_state.py", line 260, in call_callback
callback(*args, **kwargs)
File "/teamspace/studios/this_studio/AutoGroq/AutoGroq/agent_management.py", line 22, in callback
process_agent_interaction(agent_index)
File "/teamspace/studios/this_studio/AutoGroq/AutoGroq/agent_management.py", line 222, in process_agent_interaction
request += f" Additional input: {user_input}. Reference URL content: {url_content}."

When attempting to use the 'gpt-4o' parameter, the system still uses 'gpt-4'.

Your single page app or whatever streamlit is doing can be given urls (pages) that will reflect the content in the address bar, allowing us to see what we are doing.

It is now not clear whether you allow to edit the main request in "Enter your request" when editing an agent config. By offering to edit the data there users might think its a copy that needs to be edited to get better results but when doing that it is actually the main input ;(

So please implement simple UX practices: one url to edit a resource (did NOT duplicate resources/form(parts) in other locations

Hi J,

I downloaded Autogroq to test it out.

I had an issue with it not connecting to the Autogenstudio database.

I see you had it hard coded in config.py.

here is the update to make that dynamic.

#get user home directory
import os
home_dir = os.path.expanduser("~")

Database path

AUTOGEN_DB_PATH = f'{home_dir}/.autogenstudio/database.sqlite'

I enter a request like "make a crm" and proceeds to make the re engineered prompt but not agents.

Seeing a lot of \_ in generated skills...
this breaks both name recognition code and most of the python code, requiring hand edits.
I added a quick .replace("\\_","_") to fix it, but unsure of the best answer (unsure why it's escaping those)

Also here's my current improved prompt for better skills, based on best practices from Autogen Studio examples:

def get_generate_skill_prompt(rephrased_skill_request):
    return f'''
                Based on the rephrased skill request below, please do the following:

                1. Do step-by-step reasoning and think to understand the request better.
                2. Code the best Autogen Studio Python skill as per the request as a [skill_name].py file.
                3. Return only the skill file, no commentary, intro, or other extra text. If there ARE any non-code lines, please pre-pend them with a '#' symbol to comment them out.
                4. A proper skill will have these parts:
                   a. Imports (import libraries needed for the skill)
                   b. Function definition AND docstrings (this helps the LLM understand what the function does and how to use it)
                   c. Function body (the actual code that implements the function)
                   d. (optional) Example usage - ALWAYS commented out
                   Here is an example of a well formatted skill:

                   # skill filename: save_file_to_disk.py
                   # Import necessary module(s)
                   import os

                   def save_file_to_disk(contents, file_name):
                   # docstrings
                   """
                   Saves the given contents to a file with the given file name.

                   Parameters:
                   contents (str): The string contents to save to the file.
                   file_name (str): The name of the file, including its extension.

                   Returns:
                   str: A message indicating the success of the operation.
                   """

                   # Body of skill

                   # Ensure the directory exists; create it if it doesn't
                   directory = os.path.dirname(file_name)
                   if directory and not os.path.exists(directory):
                      os.makedirs(directory)

                   # Write the contents to the file
                   with open(file_name, 'w') as file:
                      file.write(contents)
    
                   return f"File file_name has been saved successfully."

                   # Example usage:
                   # contents_to_save = "Hello, world!"
                   # file_name = "example.txt"
                   # print(save_file_to_disk(contents_to_save, file_name))

                Rephrased skill request: "{rephrased_skill_request}"
                '''

I got this error message while trying to run this command "streamlit run AutoGroq/main.py". How do I set up the api key?

No secrets files found. Valid paths for a secrets.toml file are: C:\Users\camer.streamlit\secrets.toml, C:\Users\camer\Documents\AutoGroq.streamlit\secrets.toml

FileNotFoundError: No secrets files found. Valid paths for a secrets.toml file are: C:\Users\camer.streamlit\secrets.toml, C:\Users\camer\Documents\AutoGroq.streamlit\secrets.toml
Traceback:
File "C:\Users\camer\Documents\AutoGroq\autogroq_venv\Lib\site-packages\streamlit\runtime\scriptrunner\script_runner.py", line 600, in _run_script
exec(code, module.dict)
File "C:\Users\camer\Documents\AutoGroq\AutoGroq\main.py", line 82, in
main()
File "C:\Users\camer\Documents\AutoGroq\AutoGroq\main.py", line 23, in main
api_key = get_api_key()
^^^^^^^^^^^^^
File "C:\Users\camer\Documents\AutoGroq\AutoGroq\auth_utils.py", line 24, in get_api_key
api_key = st.secrets.get(api_key_name)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "", line 774, in get
File "C:\Users\camer\Documents\AutoGroq\autogroq_venv\Lib\site-packages\streamlit\runtime\secrets.py", line 305, in getitem
value = self._parse(True)[key]
^^^^^^^^^^^^^^^^^
File "C:\Users\camer\Documents\AutoGroq\autogroq_venv\Lib\site-packages\streamlit\runtime\secrets.py", line 214, in _parse
raise FileNotFoundError(err_msg)

Like the title says....create a tool that will select the correct model from the list if provided ;)

Long time listener first time caller. Just installed AutoGroq, set the config_local.py for LmStudio and received an error that Api_url couldn't be imported when I tried to run Autogroq. The following is the error:

ImportError: cannot import name 'API_URL' from 'config' (C:\Users\shake\Ag\Autogroq\AutoGroq\config.py)
Traceback:
File "C:\Users\shake.conda\envs\Ag\Lib\site-packages\streamlit\runtime\scriptrunner\script_runner.py", line 600, in _run_script
exec(code, module.dict)
File "C:\Users\shake\Ag\AutoGroq\AutoGroq\main.py", line 5, in
from agent_management import display_agents
File "C:\Users\shake\Ag\Autogroq\AutoGroq\agent_management.py", line 10, in
from utils.ui_utils import get_llm_provider, regenerate_json_files_and_zip, update_discussion_and_whiteboard
File "C:\Users\shake\Ag\Autogroq\AutoGroq\utils\ui_utils.py", line 13, in
from config import API_URL, LLM_PROVIDER, MAX_RETRIES, MODEL_TOKEN_LIMITS, RETRY_DELAY

This is my config_local.py but all I changed was the llm provider

# User-specific configurations

LLM_PROVIDER = "LMSTUDIO"
GROQ_API_URL = "https://api.groq.com/openai/v1/chat/completions"
LMSTUDIO_API_URL = "http://localhost:1234/v1/chat/completions"
OLLAMA_API_URL = "http://127.0.0.1:11434/api/generate"
OPENAI_API_KEY = "your_openai_api_key"
OPENAI_API_URL = "https://api.openai.com/v1/chat/completions"

Not sure what I could be doing wrong. LmStudio is up and running as a server but I don't think that should really matter for the purpose of the error. Love what you're doin here by the way, keep it up.

I added to: AutoGroq/Autogroq/config.py

A new line: (as a new third line from top in the file)

GROQ_API_KEY = "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"

[yes, not my real key here in the above, only the length is same.]

But; every time when I go to: http://localhost:8501

AutoGroq still asks to input the api key as;

Enter the GROQ API Key:

After entering it the output is: API Key entered successfully.

And now as far as I understand everything works.

However the issue: "The GROQ API Key must be entered every time" when going to http://localhost:8501 still remains.

For example when going to new tab (in the same browser) the asking is:

Enter the GROQ API Key:
The actual input field here
Please enter the GROQ API Key to use the app.

Unless one of you beats me to it, I'll try incorporating RAG next...

https://github.com/AllAboutAI-YT/easy-local-rag

I'm confused on how you are using Skills vs Tools. While you prompt for both, and make json for both...
Skills are specific files of python code, to do 'this action'.
Tools are used by OpenAI, to do 'this action'
But in the actual Autogen Studio code, 'tools' are used to call skills.

This seems redundant, and I'm curious if why we need both created. Skills is the right answer, "tools" is only how skills get called in the final code.

(I'm trying to write a better set of prompts)

I attempted to run the application and got the following error related to html5lib

I have attempted to update html5lib without any errors but it still fails

I have uninstalled html5lib and then reinstalled it without any errors and it still failed

I am running this on Windows 10 currently

Any suggestions?

Here is the error:

PS D:__AI-Projects\AutoGroq\AutoGroq> streamlit run .\main.py

You can now view your Streamlit app in your browser.

Local URL: http://localhost:8501
Network URL: http://192.168.1.126:8501

2024-05-17 22:47:20.577 Uncaught app exception
Traceback (most recent call last):
File "C:\Users\%USERNAME%\AppData\Local\Programs\Python\Python312\Lib\site-packages\streamlit\runtime\scriptrunner\script_runner.py", line 565, in _run_script
exec(code, module.dict)
File "D:_AI-Projects\AutoGroq\AutoGroq\main.py", line 6, in
from agent_management import display_agents
File "D:_AI-Projects\AutoGroq\AutoGroq\agent_management.py", line 8, in
from bs4 import BeautifulSoup
File "C:\Users\%USERNAME%\AppData\Local\Programs\Python\Python312\Lib\site-packages\bs4_init.py", line 30, in
from .builder import builder_registry, ParserRejectedMarkup
File "C:\Users\%USERNAME%\AppData\Local\Programs\Python\Python312\Lib\site-packages\bs4\builder_init.py", line 314, in
from . import _html5lib
File "C:\Users\%USERNAME%\AppData\Local\Programs\Python\Python312\Lib\site-packages\bs4\builder_html5lib.py", line 70, in
class TreeBuilderForHtml5lib(html5lib.treebuilders._base.TreeBuilder):
^^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: module 'html5lib.treebuilders' has no attribute '_base'. Did you mean: 'base'?
Stopping...
PS D:__AI-Projects\AutoGroq\AutoGroq>

Getting this error when first trying to generate agents:

Error making API request: 'str' object has no attribute 'get'
Debug: AutoGen Agents: []
Debug: CrewAI Agents: []
Error: No agents created.

line 197 - the call to "zip_files_in_memory" has 3 arguments and the method only accepts 1.

I changed the call to model the other calls in that same file and it seemed to work.

I don't (yet) know how to make PRs in github so I thought I would just put this here.

Hello,

Everything generates in AutoGroq fine, imports into AutoGen database just fine. I change the LLM to a local LLM running on LMStudio. Everytime I try to run the agents and workflow generated by AutoGroq I get an API key error, if I manually generate the agents it works fine.

Any suggestions?

Thank you!

After running the application we get this following error at the top of the web browser screen above GROQ API Key

CSS file not found: D:__AI-Projects\AutoGroq\AutoGroq\AutoGroq\style.css

I believe it is looking for the style.css in the wrong location as it is stored within the project here:

D:__AI-Projects\AutoGroq\AutoGroq\style.css

Hi,
I used autogroq to create autogen agents and downloaded the json files.
There are files for agents, and the workflow. however, the agent files have skills that don't exist, and there are no skill definition files in the files I downloaded.

Is this a bug? if not, how do I add the skills?

{ "type": "assistant", "config": { "name": "codebase_analyst", "llm_config": { "config_list": [ { "model": "gpt-4" } ], "temperature": 0.1, "timeout": 600, "cache_seed": 42 }, "human_input_mode": "NEVER", "max_consecutive_auto_reply": 8, "system_message": "You are a helpful assistant that can act as Codebase Analyst who Analyzes local codebases, identifies areas for improvement, and defines refactoring strategies." }, "description": "Analyzes local codebases, identifies areas for improvement, and defines refactoring strategies", "skills": [ "code_analysis", "software_architecture", "refactoring_techniques" ], "tools": [ "sonarqube", "codeheat", "codemetrics" ] }

when I try to add this through autogenstudio, I get an "unprocessable entity" error.

Looking inside the skill directory:

Where is the image generation skill .py file?

It would be very handy to have the ability to connect to a local LLM server like OLLAMA, to have a truly local solution capable of generating these agents.