Local Agents, Remote Brains: connecting OpenCode to LUMI

Artificial intelligence is changing how we write software, and coding agents are at the center of this shift. Instead of simple autocomplete, they can plan, write, debug, and run code—but they also raise new questions about safety, infrastructure, and how to use shared compute responsibly.

In this blog, machine learning specialists Artur Vojt-Antal, Mai Nguyen and Anis Rahman from LUMI AI Factory/CSC show how to run a custom OpenCode agent locally in Docker while using LUMI as its “remote brain” via a Qwen3-Coder model and vLLM. You’ll get a high-level view of the agent architecture, a minimal setup recipe, and practical guidelines for using LUMI’s supercomputing resources efficiently and in line with HPC principles.

Read the full blog on LUMI supercomputer website here