Bian Que: An Agentic Framework with Flexible Skill Arrangement for Online System Operations

Bian Que is a new agentic framework from KuaiShou, China's major short-video platform, designed to automate the operation and maintenance (O&M) of large-scale online engine systems like search, recommendation, and advertising. The key innovation is Flexible Skill Arrangement, where each Skill specifies which data (metrics, logs, change events) and knowledge (handbook rules, practitioner experience) to retrieve for a given business-module context. These Skills can be automatically generated and updated by LLMs or iteratively refined through natural-language instructions from on-call engineers, solving the bottleneck of manually curating event-to-data mappings for dozens of daily releases.

The framework abstracts O&M into three canonical patterns: release interception, proactive inspection, and alert root cause analysis. A unified self-evolving mechanism drives two parallel pathways from a single correction signal: case-memory-to-knowledge distillation and targeted Skill refinement. Deployed on KuaiShou's e-commerce search engine, Bian Que reduces alert volume by 75%, achieves 80% root-cause analysis accuracy, and cuts mean time to resolution by over 50%. The framework also achieves a 99.0% pass rate on offline evaluations, and its code is publicly available on GitHub.