Local-first action governance and budget caps framework protecting AI agents from runaways.
Core codebase is fully open source with unlimited local governance and budget caps.
Shotoku is a lightweight, local-first runtime governance and budget ceiling framework designed for autonomous AI agents. As engineering teams empower agents with shell access, local file system mutations, and third-party API integration, the risks of infinite reasoning loops, catastrophic file deletion, and unexpected billing spikes grow exponentially. Shotoku sits transparently between agent planners and tool invocation layers, enforcing strict command sandboxing, path whitelists, and real-time financial expenditure caps. When an agent exhibits hazardous patterns or exceeds predefined thresholds, Shotoku halts execution instantly, keeping autonomous workflows safe and predictable.