Archi: Agentic Operations at the CMS Experiment

Archi is an open-source framework that deploys AI agents to manage scientific data and operations for CERN's CMS experiment. Since February 2026, it has successfully supported the Computing Operations team by retrieving and reasoning over documentation, historical data, and live monitoring systems using locally-hosted models that maintain data privacy.

Archi represents a significant shift in how large scientific collaborations leverage AI for operational support. Rather than relying on cloud-based solutions that expose sensitive institutional data, the framework demonstrates that open-weight, locally-hosted models can deliver competitive performance while maintaining full privacy control—a critical requirement for organizations handling classified or proprietary research data.

The deployment at CERN's CMS experiment, one of the world's largest scientific collaborations, validates the practical viability of agentic systems in complex, high-stakes operational environments. The system's ability to synthesize heterogeneous data sources (documentation, historical records, live systems) and provide actionable intelligence to technical operators addresses a real production need. This reflects broader industry recognition that AI agents excel at information retrieval and reasoning tasks when properly scoped.

The emphasis on local model deployment has significant implications for enterprise and research institutions evaluating AI infrastructure. By proving that open-weight models perform competitively without sacrificing capability, Archi challenges the assumption that organizations must accept cloud dependency and potential data exposure for advanced AI functionality. This enables risk-averse sectors—scientific research, government, finance—to adopt agentic AI while maintaining institutional control.

Looking ahead, the evaluation methodology (operator feedback combined with human and automated grading) establishes a replicable template for assessing agent performance in production environments. Success here could accelerate adoption of similar systems across other CERN experiments and scientific facilities, potentially establishing new standards for how large institutions deploy and evaluate AI agents.

• →Archi successfully deploys AI agents for CERN's CMS computing operations, handling real-world technical queries since February 2026.
• →Locally-hosted, open-weight models deliver competitive performance while enabling full data privacy and institutional control.
• →The framework integrates heterogeneous data sources (documentation, historical data, live monitoring) to provide operators with comprehensive analytical support.
• →Evaluation based on production usage and operator feedback demonstrates practical effectiveness beyond benchmark testing.
• →The approach establishes a replicable model for enterprise and research institutions to adopt agentic AI without cloud dependency.