AnomaMind: Agentic Time Series Anomaly Detection with Tool-Augmented Reasoning

Time series anomaly detection is critical in many real-world applications, where effective solutions must localize anomalous regions and support reliable decision-making under complex settings. However, many existing methods use predefined modeling and scoring pipelines rather than an evidence-driven diagnostic process. This can limit their ability to adapt evidence acquisition to context-dependent anomalies and heterogeneous patterns. To address these challenges, we propose AnomaMind, a supervised agentic time series anomaly detection framework that organizes interval-level detection as a sequential decision-making process. AnomaMind operates through a coarse-to-fine workflow that first localizes suspicious intervals, then constructs diagnostic evidence through tool interaction, and finally refines anomaly decisions through self-reflection. The workflow is supported by reusable tool engines for context-aware diagnostic analysis. A key design is a hybrid inference mechanism: frozen general-purpose models handle tool interaction and self-reflective refinement, while a detection-specific policy is trained through supervised fine-tuning and reinforcement learning with verifiable output-level rewards. This combines task-specific optimization with flexible tool orchestration. Experiments under in-domain and multi-domain joint-training settings show strong detection performance on four benchmarks. The code is available at https://anonymous.4open.science/r/AnomaMind. % The workflow is supported by a toolkit box that combines knowledge memory with numerical diagnostics. Visual anomaly patterns mined from training data and domain knowledge provide contextual guidance, while statistical, value-based, change-based, and region-level operators offer measurable evidence for verification. AnomaMind further adopts a hybrid inference mechanism in which general-purpose models handle flexible reasoning, tool invocation, and refinement, while a detection-specific policy is optimized with detection-specific rewards. Extensive experiments under both in-domain and cross-domain settings demonstrate that AnomaMind consistently improves anomaly detection performance, validating the effectiveness of tool-augmented reasoning for anomaly detection. The code is available at https://anonymous.4open.science/r/AnomaMind.