Ctrlb团队发布了开源日志分析工具ctrlb-decompose，可将原始日志行转化为结构化模式、统计数据和异常标志[1]。该工具通过时间戳提取、CLP编码、Drain3聚类、变量提取与分类、统计累积、异常检测及评分与关联等处理流程实现数据规模缩减[1]。功能演示表明，1,247,831行日志可被压缩为43个模式，实现99.9%的压缩率[1]。
ctrlb-decompose支持多种运行方式和输出格式[1]。用户可将其作为命令行工具、WASM浏览器版本或Rust库使用，并可通过brew install、dpkg、git clone或Claude Code插件等方式安装[1]。该工具支持人类可读的ANSI终端输出、针对大语言模型优化的markdown格式以及JSON格式[1]。底层技术采用DDSketch量分位数、HyperLogLog++基数估计、储层采样和Drain3前缀树聚类等算法[1]。
Ctrlb-decompose, an open-source log analysis tool, has been released to compress massive volumes of raw log data into structured patterns, statistical summaries, and anomaly flags [1]. The tool achieves a 99.9% reduction in data scale, enabling more efficient processing of logs before sending them to large language models [1].
The tool operates through a multi-stage processing pipeline that extracts timestamps, applies CLP encoding, performs Drain3 clustering, extracts and categorizes variables, accumulates statistics, detects anomalies, and generates relevance scores [1]. A demonstration showed the tool compressing 1,247,831 log lines into just 43 patterns, achieving the stated compression rate [1]. Users can access ctrlb-decompose as a command-line interface, a WebAssembly browser version, or a Rust library [1]. Output formats include human-readable ANSI terminal display, LLM-optimized markdown, and JSON [1]. Installation options include Homebrew, dpkg, git clone, or integration as a Claude Code plugin [1].
The tool leverages advanced statistical techniques including DDSketch quantile sketching, HyperLogLog++ cardinality estimation, and reservoir sampling to efficiently process large log datasets [1]. These technologies, combined with Drain3 prefix tree clustering, enable the rapid identification and consolidation of similar log entries [1].