Agentic AI at RCAC¶
Researchers on RCAC's clusters are already using agentic coding tools — Claude Code, OpenAI Codex, Gemini CLI, opencode, and Warp — to write Slurm scripts, debug failing jobs, move data, and explore software. RCAC's stance is proactive engagement, not prohibition. Rather than forbidding these tools, we shape the context in which you use them so their output is more likely to be correct for our systems, and we give you what you need to verify that output rather than simply trust it.
These tools are genuinely useful, but they are also capable of producing confident, plausible, and wrong answers: a Slurm script for a scheduler we don't run, a module name that doesn't exist, a path on the wrong filesystem. The aim of this section is to make the tools work well on RCAC by giving them accurate, cluster-specific context, and to give you the judgment to check what they produce. Treat an agent as augmenting your expertise, not outsourcing it — ask why, not just what.
Early and evolving
This is new, actively-developed work. The MCP servers, the shared context files, and the per-harness settings documented here are prototypes that will change over time. We publish them openly and invite your feedback — see the shared-context pages for how to send corrections.
In this section¶
-
Acceptable Use & Etiquette
The rules for running agents on RCAC systems, consistent with Purdue's Acceptable Use Policy.
-
Best Practices & Limitations
Using an agent for research versus for operations; context engineering; verifying output; and the caution/blast-radius risks.
-
MCP Servers
RCAC's MCP servers (
rcac-mcp,globus-mcp,rcac-docs-mcp) and how they give an agent context that knows our clusters. -
Running Agents
Set up your harness on the cluster (login nodes) or locally, targeting the cluster over SSH — for all five harnesses.
-
Shared Context & Settings
The actual context files and per-harness settings each cluster deploys, published verbatim as the canonical source of truth — plus how to send feedback.
For cluster-specific setup pointers, see the Using AI Agents chapter in each cluster's user guide — for example Gautschi, Negishi, or Gilbreth.
RCAC's approach to agentic computing is described in the PEARC'26 paper "Hello Computer: HPC in the Agentic Era" (Lentner & Ashish, 2026) — see the replication package.