Open source rule based guardrails for coding agents (github.com)

A new open-source project called Prempti has been launched to provide customizable guardrails for AI coding agents, enhancing the safety and oversight of their operations. Built on Falco, Prempti allows users to enforce rules that can block or ask for confirmation on potentially harmful actions taken by coding agents, such as executing shell commands or modifying files. The tool operates in two modes: guardrails mode, which actively enforces these rules, and monitor mode, which logs actions without interference, making it ideal for observing agent behavior and tuning rules. This initiative is significant for the AI/ML community as it addresses safety concerns surrounding the increasingly autonomous actions of coding agents. By offering customizable YAML rules that cater to specific workflows and providing a full audit trail of all tool calls, Prempti empowers developers, product managers, and designers to establish clear boundaries for their agents. While it does not replace necessary security measures like sandboxing or system hardening, Prempti complements these with real-time tool-call interception and user-friendly interaction through a CLI. This tool not only enhances operational awareness but also fosters a cooperative environment for developing safer AI applications.