Roadmap for releasing a new web application based on autogen.

[pcdeadeasy]

No description provided.

[Rockets2TheMoon]

Very excited for this! How can I contribute?

[KenDingel]

^ Same here!

[pcdeadeasy]

Thanks for your interest. The goal is to release a research assistant that allows you to do q & a on papers, aggregate results, create visualizations/artifacts, skills that can be reused.
Tagging @sonichi @victordibia
@gagb @afourney

[binary-husky]

I working on implementing AutoGen in GPT-Academic: https://github.com/binary-husky/gpt_academic/

I need some help: what is the best way to re-route the message console output from autogen elsewhere?

[sonichi]

I know two options: register a callback function using register_reply or use web sockets. @victordibia do you have a better answer?

[binary-husky]

In #322 , I remap callback by :

agent_handle._print_received_message = lambda a,b: self.gpt_academic_print_override(agent_kwargs, a, b)
if agent_kwargs['name'] == 'user_proxy':
    agent_handle.get_human_input = lambda a: self.gpt_academic_get_human_input(user_proxy, a)
    user_proxy = agent_handle
if agent_kwargs['name'] == 'assistant': assistant = agent_handle

[binary-husky]

In #322 , I remap callback by :

agent_handle._print_received_message = lambda a,b: self.gpt_academic_print_override(agent_kwargs, a, b)
if agent_kwargs['name'] == 'user_proxy':
    agent_handle.get_human_input = lambda a: self.gpt_academic_get_human_input(user_proxy, a)
    user_proxy = agent_handle
if agent_kwargs['name'] == 'assistant': assistant = agent_handle

register_reply is a good idea, however, several flag variables are probably required inside autogen code.

[sonichi]

It looks to be on the right track. What flag variables are you referring to?

[ragyabraham]

Hey @sonichi @pcdeadeasy @binary-husky @victordibia , we have just released a web app based on autogen. You can find the repo here. It is still early days, but we intend to build out a lot of features to help accelerate uptake and also to help non-technical users get the most out of autogen

[pcdeadeasy]

@ragyabraham, I looked at the code base for a bit. It will be great if you create an architecture diagram to show the data workflow of the different components of the application (frontend, agent-backend etc.)

[ragyabraham]

Sure @pcdeadeasy so you're after an arch diagram to show how the messages are being passed from agent-backend to webapp?

[GoldenWind8]

Where can we make contributions and what tech stack is going to be used?

[sonichi]

@GoldenWind8 would you like to review #509 ?

[gagb]

New relevant PR: pcdeadeasy#2

[victordibia]

An initial commit on this has been made to the autogen repo in /samples/apps/autogen-assistant .

Some next steps

### Requested Features
- [ ] Make agent selection persistent and based on db entries.
- [ ] Change subtitle to by Multi-Agents
- [ ] Save/export logs from interactions.
- [ ] Use function calling to register and execute functions instead of putting them in prompt
- [ ] Stream messages from agents to UI as they are generated
- [ ] Add other contributed agents: teachable agent, compressible agent, gptassistant agent

@gagb

[gagb]

Moving the task list to new issue here: #737

[victordibia]

Closing this due to inactivity.

References

• AutoGen studio is the emerging reference example of using AutoGen with a web interface.
• Roadmap for AutoGen Studio is here [Roadmap] AutoGen Studio #737