Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OpenDerisk: An Industrial Framework for AI-Driven SRE, with Design, Implementation, and Case Studies

Peng Di, Faqiang Chen, Xiao Bai, Hongjun Yang, Qingfeng Li, Ganglin Wei, Jian Mou, Feng Shi, Keting Chen, Peng Tang, Zhitao Shen, Zheng Li, Wenhui Shi, Junwei Guo, Hang Yu

TL;DR

The paper addresses the heavy cognitive burden on SREs by introducing OpenDerisk, an open-source, diagnostic-native, multi-agent framework for AI-driven SRE. It combines a pluggable Reasoning Engine, a Knowledge Engine (K-Engine), a Model Context Protocol (MCP), and adaptive multi-agent collaboration to enable domain-specialist agents to perform long-running RCA tasks. Through large-scale industrial deployment at Ant Group and rigorous comparisons against monolithic baselines, it demonstrates that a Multi-Specialist Agent framework significantly improves accuracy and robustness, with model-agnostic adaptability and scalability. The work provides detailed design, implementation, and empirical validation, and outlines a path toward autonomous remediation and self-healing knowledge for enterprise SRE environments.

Abstract

The escalating complexity of modern software imposes an unsustainable operational burden on Site Reliability Engineering (SRE) teams, demanding AI-driven automation that can emulate expert diagnostic reasoning. Existing solutions, from traditional AI methods to general-purpose multi-agent systems, fall short: they either lack deep causal reasoning or are not tailored for the specialized, investigative workflows unique to SRE. To address this gap, we present OpenDerisk, a specialized, open-source multi-agent framework architected for SRE. OpenDerisk integrates a diagnostic-native collaboration model, a pluggable reasoning engine, a knowledge engine, and a standardized protocol (MCP) to enable specialist agents to collectively solve complex, multi-domain problems. Our comprehensive evaluation demonstrates that OpenDerisk significantly outperforms state-of-the-art baselines in both accuracy and efficiency. This effectiveness is validated by its large-scale production deployment at Ant Group, where it serves over 3,000 daily users across diverse scenarios, confirming its industrial-grade scalability and practical impact. OpenDerisk is open source and available at https://github.com/derisk-ai/OpenDerisk/

OpenDerisk: An Industrial Framework for AI-Driven SRE, with Design, Implementation, and Case Studies

TL;DR

The paper addresses the heavy cognitive burden on SREs by introducing OpenDerisk, an open-source, diagnostic-native, multi-agent framework for AI-driven SRE. It combines a pluggable Reasoning Engine, a Knowledge Engine (K-Engine), a Model Context Protocol (MCP), and adaptive multi-agent collaboration to enable domain-specialist agents to perform long-running RCA tasks. Through large-scale industrial deployment at Ant Group and rigorous comparisons against monolithic baselines, it demonstrates that a Multi-Specialist Agent framework significantly improves accuracy and robustness, with model-agnostic adaptability and scalability. The work provides detailed design, implementation, and empirical validation, and outlines a path toward autonomous remediation and self-healing knowledge for enterprise SRE environments.

Abstract

The escalating complexity of modern software imposes an unsustainable operational burden on Site Reliability Engineering (SRE) teams, demanding AI-driven automation that can emulate expert diagnostic reasoning. Existing solutions, from traditional AI methods to general-purpose multi-agent systems, fall short: they either lack deep causal reasoning or are not tailored for the specialized, investigative workflows unique to SRE. To address this gap, we present OpenDerisk, a specialized, open-source multi-agent framework architected for SRE. OpenDerisk integrates a diagnostic-native collaboration model, a pluggable reasoning engine, a knowledge engine, and a standardized protocol (MCP) to enable specialist agents to collectively solve complex, multi-domain problems. Our comprehensive evaluation demonstrates that OpenDerisk significantly outperforms state-of-the-art baselines in both accuracy and efficiency. This effectiveness is validated by its large-scale production deployment at Ant Group, where it serves over 3,000 daily users across diverse scenarios, confirming its industrial-grade scalability and practical impact. OpenDerisk is open source and available at https://github.com/derisk-ai/OpenDerisk/
Paper Structure (52 sections, 9 figures, 2 tables)

This paper contains 52 sections, 9 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. SRE Automation.
  4. LLM-based Software Engineering.
  5. Multi-Agent Frameworks.
  6. Context Engineering and Optimization.
  7. OpenDerisk System
  8. Design Philosophy
  9. System Workflow and Components
  10. Perception Layer: Sensing the Environment
  11. The DeRisk System: Core Decision and Execution
  12. Multi-Agent System and Collaboration Paradigms
  13. Reasoning Engine
  14. Knowledge Engine (K-Engine)
  15. Tools and Model Context Protocol (MCP)
  16. ...and 37 more sections

Figures (9)

  • Figure 1: The web-based user interface for OpenDerisk is open source and available at https://github.com/derisk-ai/OpenDerisk/
  • Figure 2: The end-to-end architecture of the OpenDerisk system, illustrating the flow from Perception to Analysis, centered around the core Decision and Execution engine, and augmented by Human-in-the-Loop feedback.
  • Figure 3: Multi-Agent Collaboration workflow in OpenDerisk. The framework dynamically selects a collaboration paradigm (Single-Agent, TeamMode, or GroupMode) based on the application scenario, supported by a multi-layered memory module.
  • Figure 4: The five-stage knowledge processing pipeline of the K-Engine, which transforms raw data into a structured, indexed, and actively maintained knowledge base.
  • Figure 5: The designs of Emergency Time-Series Trend Agent. Left is basic version with a two-agents team, right is an advanced design with multi-agents paradigm.
  • ...and 4 more figures