Building Scalable AI Agents with Ray
summary
Ragent is a Ray-based framework that distributes agents as actors and tasks, uses Ray Core for scheduling and Ray Serve for real-time endpoints, and runs indexing and heavy CPU work as RData jobs on the same cluster. It already hosts 500 agent apps, integrates LangChain/AutoGen/MetaGPT, and demonstrates a multi-agent “snake game” workflow that completes design, code and unit tests with GPT‑4 after several iterative rounds. Main wins: dynamic GPU/CPU placement and multi-agent orchestration; ongoing work adds an agent protocol and online/offline fusion.
AI-written summary · about
chapters (7)
- 0:00Introduction to scalable AI agents
- 5:49Challenges in agent production
- 8:01Why Ray for agent scaling
- 13:26Building a single agent
- 15:17Multi-agent collaboration demo
- 17:04Future work on agent protocols
- 18:30Q&A session
related talks
15mRay: A Distributed Compute Engine for AI
22mRay, a Unified Distributed Framework for the Modern AI Stack
31mLeveraging Agent Skills for Ray Workloads
28mScaling AI & Machine Learning Workloads With Ray on AWS, Kubernetes, & BERT
17mFrom Agents to Action: Scaling Agentic AI with Ray on Intel Xeon
09m[Opening Keynote] State of Ray
24mScalable GenAI in Production
25mFrontiers of Agentic AI