A zero-dependency Python loop detector for AI agents (github.com)

SAVI has launched a zero-dependency Python library named savi-loop-guard, designed to detect when AI agents fall into loops—either by repetitively calling the same tool or making calls at an unrealistic pace. This library stands out as it requires no external accounts, API keys, or additional dependencies, making it incredibly easy for developers to integrate into their agent code. Users can track agent calls in real-time and check for looping patterns either after each call or on specific intervals. The library distinguishes between two types of loops: velocity loops, where an agent makes too many calls in a short time, and structural loops, which occur when a tool is called multiple times, even if the names slightly vary. The significance of this tool lies in its ability to proactively prevent looping issues, rather than merely observing them after they occur—a feature not offered by existing loop detection libraries. The savi-loop-guard’s flexible configuration empowers developers to tailor the detection parameters to their specific needs, allowing for customized thresholds on call counts and time windows. This ensures that developers can effectively manage AI behavior, leading to more efficient workflows and ultimately enhancing the reliability of AI systems.