Skip to content

feat(demos): add AG2 multi-agent research team with Plano orchestration#841

Open
faridun-ag2 wants to merge 1 commit intokatanemo:mainfrom
faridun-ag2:feat/ag2-agent-orchestration-demo
Open

feat(demos): add AG2 multi-agent research team with Plano orchestration#841
faridun-ag2 wants to merge 1 commit intokatanemo:mainfrom
faridun-ag2:feat/ag2-agent-orchestration-demo

Conversation

@faridun-ag2
Copy link
Copy Markdown

Summary

  • Adds a new demo demos/agent_orchestration/multi_agent_ag2/ showcasing AG2 multi-agent orchestration via Plano
  • AG2 runs an internal GroupChat (researcher + analyst) behind a single Plano endpoint — Plano sees one agent, while multi-agent collaboration happens inside
  • Key differentiator vs the existing CrewAI/LangChain demo: there, each framework = one agent = one endpoint; here, AG2 runs a team of agents behind one endpoint

What's included

  • ag2/research_agent.py — FastAPI server with OpenAI-compatible /v1/chat/completions (streaming + non-streaming) and /health; AG2 GroupChat with researcher + analyst agents
  • config.yaml — Plano config registering the agent on port 10530 with plano_orchestrator_v1 router
  • run_demo.sh / start_agents.sh — one-command startup scripts
  • pyproject.toml — dependencies (ag2[openai], fastapi, uvicorn, openai)
  • openai_protocol.py — shared streaming utility (copied from existing demo)
  • README.md — full documentation with architecture diagram and quickstart

Test plan

  • GET /health returns {"status":"healthy","agent":"ag2_research_team"}
  • POST /v1/chat/completions streaming works end-to-end through Plano
  • POST /v1/chat/completions non-streaming works end-to-end through Plano
  • AG2 GroupChat runs exactly 2 LLM calls (researcher → analyst) with no extra round-trips
  • pre-commit hooks pass (black, yaml, secrets scan)

AG2 (formerly AutoGen) integration demonstrating:
- Multi-agent GroupChat (researcher + analyst) behind a single Plano endpoint
- OpenAI-compatible streaming HTTP endpoint via FastAPI
- Internal multi-agent collaboration transparent to Plano routing
- Unified observability through Plano's tracing layer
@faridun-ag2
Copy link
Copy Markdown
Author

Hey @adilhafeez, I've added an AG2 multi-agent demo to the repo!

This shows a unique pattern - AG2's GroupChat (researcher + analyst) running behind a single Plano endpoint. Unlike the existing CrewAI/LangChain demo where each framework = one agent, here AG2 runs an internal team that Plano sees as one agent.

Would love your feedback when you get a chance!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant