Million-Agent Ruby: Ractor-Local GC in the Age of AI
summary
A Ractor-native agent framework called Ragents runs simulated LLMs on a laptop, claims parallelism gains for CPU-bound tool calls, and shows benchmarks comparing Falcon, Puma, threads, and reactor pools. Messages and contexts are frozen shareable objects with move/copy/share semantics and port ownership to avoid races. Ractor-local GC is still work in progress but promised to cut memory and improve production readiness. A live demo sometimes crashes; code and reproducible benchmarks are on GitHub.
AI-written summary · about
chapters (10)
- 0:00Introduction and welcome
- 1:56Ruby’s concurrency evolution
- 4:03Threads fibers and reactors
- 8:42Ractor ports and agent workflows
- 12:37Sharing data across reactors
- 17:44Agent orchestration and isolation
- 21:20Million-agent Ruby concept
- 23:25Ractor local GC progress
- 25:29Live agent demo walkthrough
- 28:15Closing and acknowledgments
related talks
30mRactor on Ruby 3.4
25mRactor Demonstration
28mCome Discuss Ractors and Concurrency with 'The Dude'!
28mRactors in Ruby 4.0: Message Passing Without the Pain
27mParallel testing with Ractors - Putting CPU's to work
29mUsing Ractors and Making Ractors Usable
21mRubyLLM
34mActors in Ruby! Why let Elixir have all the fun?