🤖 AI Agents for Feed Building & Social Prompting

🤖 MBD Agent for Feed Building

This project showcases several AI agents built using LangGraph and Groq Inference, designed to enhance interaction with the Farcaster decentralized social network. These agents leverage the powerful mbd API to enable dynamic feed building and social prompting.

🔧 Building blocks

Using structured LangGraph workflows, we implement the two following agents. Both agents maintain a contextual memory summary of the conversation, allowing them to interpret ambiguous user inputs, remember past user preferences throughout the session and reduce token usage. All results are returned in a human-readable Markdown format.

1 – Feed Builder

Retrieve content from Farcaster network. It can use several tools, including:

• Personalized feed (based on user ID)
• Trending casts
• Popular casts
• Semantic casts (based on user query)

2 – Discover New Users

Discover relevant people in the Farcaster network. It can use several tools, including:

• Similar users (based on user ID)
• Semantic users (based on user query)
• Suggested users (based on user ID)

✏️ Graph Design

🧠 Powered By

• Groq LLMs
• Langfuse: Integrated for tracing and observability
• mbd APIs: For fetching real-time data from Farcaster

---

This agent combines three distinct functionalities using a unified LangGraph architecture:

1. Feed Builder
2. New Users
3. Small Talk : Engages in general conversation when the user is making casual inquiries.

✏️ Graph Design

The agent evaluates user input and decides which action to take based on the user's query.

The planner node helps determine whether to:

• Trigger the feed-builder node (for querying Farcaster content),
• Activate the new-user node (to suggest users), or
• Engage the small-talk node (for conversational interaction).

🤖 MBD Agent for Social Prompting

🔧 Building blocks

Using structured LangGraph workflows, we implement the two following agents: one reads the recent activity of the user, and based on that extrapolate keywords and topics to do recommendation; the other is an icebreaker bot that is activated when the user's profile is quite inactive, so the bot tries to capture user's interests trough a multi-turn conversation.

✏️ Graph Design

🚀 Reproducibility and Usage

1. Clone the Repository

git clone https://github.com/msilver22/agentic_mbd.git
cd agentic_mbd

2. Set Up Docker Environment

Make sure to run the repository within a Docker container.

3. Install Python Dependencies

pip install -r requirements.txt

4. Set up the API keys

Make sure to set the API keys into .envfile. It's needed:

• Groq API key
• mbd API key
• Langfuse public/private key (optional for tracking)

5. Explore the Agents

The repository is organized as follows:

• src/ : This folder contains the core implementation of the agents. You can explore the code here to modify or test the underlying models and functionalities.

• streamlit/ : This folder contains the Streamlit-based UI for the agents. Each agent is designed to be easily interactable through the web interface provided by Streamlit.

6. Running the Agents with Streamlit UI

To run any of the agents with the Streamlit UI, navigate to the streamlit directory.

cd streamlit

Run the desired agent using Streamlit.

streamlit run agent_name.py

Replace agent_name.py with the appropriate script name for the agent you want to run (e.g., feed_builder.py, social_prompter.py, mbd_agent.py).

The agent will open in your web browser, where you can interact with it via the UI.