Build Multi-Agent LLM Applications with AutoGen

Enable natural language understanding and generation through the LLM component in Autogen, the agent's brain guided by a system message defining personality and style and guiding response format.

The human in the loop sits before auto-reply and LLM components to decide when to intervene. It supports always, terminate, and never modes, with intercept feedback that resets auto-reply counter.

Learn how Autogen's code executors—command line and Jupyter—run code locally or in docker, with independent blocks or a shared kernel, and understand production security considerations.

Define tools as simple Python functions to control agent actions, including web scraping and API calls, register them with agents, and implement an autonomous top three stories scraper for Wikipedia.

Initiate a two-agent chat, summarize it, and carry over the summary for subsequent chats in a sequential workflow with AutoGen. Use adventurer, wizard, and guardian to complete interdependent subtasks efficiently.

Explore group chat with multiple agents using AutoGen. A group chat manager orchestrates turns via auto, round robin, or manual strategies for roles like product manager, scrum master, and software engineer.

Package complex workflows inside a single agent with nested chats, reusing patterns and exposing a single conversational interface. Trigger nested chats by conditions and synthesize final responses from results.

Learn to decompose complex tasks into subtasks assigned to expert agents using group chats and planning functions, with AutoBuild automatically generating a team of agents for a 14-day scrum plan.

Autogen's transform messages capability trims long histories and enforces token limits, and supports a PII remover to redact emails before routing messages to the agent's LLM.

Learn to use non-openai models via proxies with an OpenAI-compatible API to access open source and fine-tuned models in a self-contained, cost-efficient setup, with LM Studio enabling local inference.