Integration with AutoGen workflow

[QZGao]

Construct AutoGen config workflow with create_memgpt_autogen_agent_from_config.

If human_input_mode == "ALWAYS", it's basically a wrapped create_autogen_memgpt_agent;

If human_input_mode == "TERMINATE" or "NEVER", it's an auto-running group chat of memgpt agent and user proxy agent.

[cpacker]

@QZGao could you add a small example (can be a small diff off of examples/agent_groupchat.py) that demonstrates how the new functionality works?

I tested the existing script and nothing broke, so once you add an example + paste the outputs in a comment here, I think this is ready to merge.

[cpacker]

Example on how to use new feature + example output

[QZGao]

The original agent_groupchat.py shall generate the same result with USE_AUTOGEN_WORKFLOW = True.

The new agent_autoreply.py shall generate the following result:

User_proxy (to MemGPT_coder):

I want to design an app to make me one million dollars in one month. Tell me all the details, then try out every steps.

--------------------------------------------------------------------------------
MemGPT_coder (to MemGPT_coder):

Your goal is ambitious, but not impossible. I'd suggest three phases: 1) Research and Planning, 2) App Design and Development, and 3) Marketing & Monetization. Let's dig deep into these, starting with Research and Planning.

--------------------------------------------------------------------------------
MemGPT_coder (to MemGPT_coder):

In the Research and Planning phase, we define the app's purpose and make sure it offers value. This involves market research, understanding the target audience, and analyzing competitors. Then, choosing the right platform (iOS or Android, or both) and determining the app's features and functionality. It's also a good time to decide on the technology stack for development. Finally, we set a timeline and budget for the project. Now let's elaborate on these points.

--------------------------------------------------------------------------------
MemGPT_coder (to MemGPT_coder):

Step 1: Market Research - Explore app stores, study user reviews, identify successful apps and analyze their business models, target audience, and technology. Step 2: Identify the Target Audience - Understand their preferences, demographics, and mobile behavior. Step 3: Competitor Analysis - Evaluate similar apps, analyze their features, and identify what you can offer differently. Now, let's talk about selecting the right platform.

--------------------------------------------------------------------------------
MemGPT_coder (to MemGPT_coder):

Choosing a platform: If the audience is largely in the U.S or earns a high income, iOS might be better. Globally, Android dominates. Features and functionality: In the first version, only include necessary features. As the app grows, you can add more based on feedback. Tech stack: Depending on complexity, you can choose native development (Swift for iOS, Java for Android), hybrid development (React Native, Flutter), or progressive web apps. Now let's move onto the App Design and Development phase.

--------------------------------------------------------------------------------

[cpacker]

LGTM - @vivi , should we add a README in the examples folder?

TODO add example