Wandesk is an AI desktop environment where users can create custom applications simply by describing what they need in natural language. The core purpose is to move beyond AI conversation as the only interface, providing a persistent workspace where AI-generated software has shape and stays, enabling users to build tools like calorie trackers, reading lists, or invoice generators directly on their local machine. Key features include the ability to build apps through natural language description, support for plugging in various AI models (Claude Code, Codex, DeepSeek, OpenAI, Kimi, Qwen, or any OpenAI-compatible API), and shared context across all applications where the AI remembers user preferences and information. The platform operates 100% locally with no signup required, keeping all apps, data, files, and memory on the user's machine. Apps are generated with clear, layered structure including UI, logic, data layers, and an APP.md file that serves as a source of truth for future edits. The system works by having users describe their app needs in natural language, with the AI generating applications iteratively rather than in one shot. When descriptions are vague, the AI makes reasonable assumptions and builds a version 1 that users can refine through chat. The platform maintains one shared memory store above all applications, allowing context to be ambient and persistent without requiring per-pair wiring between apps. Every action is visible to users, with commands shown in chat, and the AI asks before performing destructive operations like deletions or overwrites. Benefits include the ability to prototype useful local tools quickly without spinning up full projects, complete data privacy since everything stays on the local machine, and editable application files that users can adjust manually or have the AI modify specific parts. Use cases range from developers wanting faster prototyping methods to non-technical users needing end-to-end solutions without touching a terminal, with examples including productivity trackers, fitness applications, invoice generators, and bill splitters. The target users initially are developers and builders who want to prototype local tools quickly, but the long-term direction aims to serve non-technical users who need to create applications without coding. The platform is available on macOS and Windows, is open source, and can integrate with MCP (Model Context Protocol) when needing to reach external services. Technical details include local storage in workspace folders with per-app SQLite databases, memory stored in a memories table, and applications structured as real editable files on disk.