# KVDesk

KVDesk is a cross-platform desktop AI assistant that runs against local or remote OpenAI-compatible models, including vLLM, Ollama and LM Studio, with task planning, scheduled tasks and sandboxed code execution.

KVDesk is a desktop AI assistant built with React and Tauri that connects to local model runners such as vLLM, Ollama and LM Studio, or to any OpenAI-compatible endpoint, and gives the assistant tools rather than only chat: a visual todo panel with progress tracking, scheduled and recurring reminders with auto-execution, JavaScript and Python code sandboxes, PDF and DOCX text extraction, web search through Tavily or Z.AI, and directory-sandboxed file operations.

Sessions persist across restarts in a local SQLite database, with pinning, search, message editing and resend, and a permission system that can require confirmation before a tool runs. A loop-detection feature stops the assistant automatically after repeated identical tool calls, and full request and response logs are kept per session for debugging. A persistent memory file stores user preferences across sessions, and the interface streams responses at 60 frames per second.

Website: https://github.com/vakovalskii/NeuralDeskApp