Show HN: Kaset — Embed Coding Agents Directly in Your Web App (medium.com)

Kaset is an open-source toolkit designed to embed coding agents directly within web applications, running entirely client-side using a browser-native filesystem (OPFS). This innovative approach transforms the traditional developer-driven workflow by enabling users—without any coding experience—to create and modify plugins or app features on the fly. By representing app components such as user data, configuration, and UI state as files, Kaset leverages a filesystem interface that supports searchable, diff-able, and version-controlled edits, ensuring all changes are transparent, reviewable, and reversible. The significance for the AI/ML community lies in Kaset’s novel integration of coding agents with built-in guardrails: scoped workspaces, user approval for mutations, and version control via isomorphic-git. Running agents locally reduces infrastructure costs, cuts latency, and enhances privacy by keeping sensitive modifications on-device. This contrasts with existing RPC-based coding models like MCP by focusing on flexible, incremental “micro-edits,” perfectly suited to applications demanding extensive user customization, such as dashboards, knowledge tools, and vertical SaaS products. Kaset’s modular design, governed by a simple agents.md file outlining permitted actions, also allows developers to swap agents without redesigning the app environment, encouraging experimentation with different AI assistants. Currently in early development, Kaset invites community feedback to refine its secure defaults and expand use cases. Its client-side coding agents mark a shift toward more dynamic, user-empowering interfaces that could redefine how web apps adapt and evolve through AI-driven collaboration.