Lecture Notes on Agents with Lang Graph

Introduction to Lang Graph

• Lang Graph is built on top of Lang Chain.
• Primary focus is on creating agents or agentic workflows.
• Key functions of Lang Graph:
  • Use of LLM (Large Language Model) to determine a sequence of actions.
  • Combine multiple agents into an agent swarm for data sharing and collaboration.

Comparison with Other Frameworks

• Autogen and Crew AI: Higher-level frameworks with certain merits.
• Lang Graph: Steeper learning curve but provides a deeper understanding and customizable workflows.
• Focus on notes, edges, and state for comprehensive agent behavior control.

Key Concepts

• Notes: Functions or runnables that perform specific tasks; each note processes input and updates state.
• Edges: Connections that define the flow and order of execution between notes.
• State: Represents the data exchanged between notes, enabling dynamic workflows.

Example: Basic Notes and Edges

• Example workflow creation in VS Code:
  • Initial setup includes installing Lang Graph via pip install Lang graph.
  • Usage of an OpenAI API key.
  • Import classes such as MessageGraph.
  • Create nodes (e.g., Branch A, Branch B) and define edges to control the flow.

Visualizing Graphs

• Visualization accessible post-compilation via methods such as get_graph().
• Helpful for understanding complex workflows.

Conditional Edges

• Ability to route actions based on input through conditional edges.
• Example of a simple conditional graph demonstrating this feature.

Cycles in Lang Graph

• Cyclic workflows are possible with Lang Graph but not with mainstream LCL (Lang Chain expression language).
• Example of a cyclic workflow with conditions to repeat tasks.

Real-World Application: Fake Weather API

• Creating an agent that interacts with a simulated weather API.
• Setup includes:
  • Custom state maintaining API call count with a limit on retries.
  • Integration of a weather-checking function utilizing a randomized output.

Workflow Implementation Steps

1. Define messages and tool calls.
2. Use conditional logic to manage API calls based on input messages.
3. Create and invoke the Lang Graph workflow designed to return weather information.

Conclusion

• Understanding how to build agents using Lang Graph leads to complex workflow creation.
• Future discussions will focus on multi-agent workflows.

Main Takeaway: Lang Graph delivers robust flexibility and customizability for building and managing agent interactions and workflows, particularly useful in real-world applications like API integration.