Langgraph - The Agent Orchestrator
AI Summary
Summary: Langra - An Agent Orchestrator
- Introduction
- Langra is a package from Langin, developed by Aris Chase and his team.
- It serves as an agent orchestrator.
- Recommended to read a Medium blog post explaining orchestration before the recording.
- Background
- Orchestration and choreography are service interaction patterns from microservices.
- Orchestration can be applied to micro agents to create multi-agent interaction systems.
- Motivation for Langra
- Existing multi-agent frameworks are often monolithic.
- Traditional frameworks define agents and orchestration within the same component.
- This can be problematic if changes to orchestration affect agent definitions.
- Langra’s Approach
- Decouples agent definition from orchestration.
- Allows development of agents separately and orchestration on top.
- Similar to microservices, where atomic services are orchestrated by tools like Camunda, Pega, or Flowable.
- Capabilities of Langra
- Combines traditional microservices with micro agents.
- Supports agents from different LLM providers and frameworks (e.g., Gemini, Bedrock, OpenAI GPT, Autogen, Crei, Langroid).
- Enables hybrid techniques for service interaction.
- Use Case Example
- An application that presents regional news to geographically distributed users.
- Atomic agents for BBC, CNN, Fox, and NDTV collect news.
- Langra orchestrates these agents to filter news for users from India, USA, and EMEA.
- Demonstration
- For simplicity, responses from agents are hardcoded instead of calling actual services.
- Four wrappers created for BBC, CNN, NDTV, and Fox agents.
- Langra uses a shared state (messages list) accessible by all agents.
- Workflow defined with entry points, edges, and a finish point.
- Compilation of workflow generates a runnable interface with invoke functions.
- Example run collects and prints news from all four agents.
- Conclusion
- The demonstration simplifies Langra’s concept without using real LLM agents.
- Future recordings will explore more complex use cases and integration with traditional microservices.