Legion: Agentic Code Execution with Pure Elixir
summary
Legion is an Elixir-native agent framework that turns existing modules into executable tools by having LLMs generate real Elixir code which runs in a sandbox environment with timeouts and AST validation. Agents are plain GenServer processes, persist variable bindings across turns, and use Vault for per-request authorization. Built demo: a LiveView conference app where an agent bulk-deletes bookmarks and spawns planner sub-agents to build a day schedule; code execution succeeds in the demo and errors round-trip back to the model for recovery.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:13What is Legion
- 7:50Agents and tools
- 10:19Code execution pipeline
- 12:48Multi-agent systems
- 13:56Demo companion app
- 18:08Live demo
- 18:29Recap
- 19:34Q&A
related talks
27mSelf-Healing Software Using Autonomous LLM Agents
39mLLM Agents Gone Wild (And How to Tame Them with Quarkus)
53mBuilding Agentic Applications w/ Heroku Managed Inference and Agents
01h 00mFrom LLM Theory to Practical Agentic Implementations
41mBuilding LLM Agents – Challenges and Learnings
19mLLM codegen fails and how to stop 'em
15mServerless & Agentic AI: Better Together
01h 43mBuilding LLM Agents (Hackathon)