Kun is a local-first AI workbench designed to transition artificial intelligence from simple conversational responses to autonomous task execution across real-world projects. The platform operates via a unified local runtime (kun serve) that synchronizes state between a desktop GUI and a terminal-based TUI, ensuring continuity for tasks, approvals, and evidentiary logs. The system features two primary modes: "Code" mode, which facilitates software delivery through file editing, terminal access, Git/Worktree management, and a dedicated Design canvas for prototyping, and "Work" mode, which handles document analysis, Markdown editing, and presentation generation from outlines. The agentic workflow follows a structured pipeline: goal clarification, plan formulation, execution via tool calls and file modifications, and final delivery with verifiable evidence such as diffs and test results. Kun supports a wide array of model providers, including OpenAI, Anthropic, Google Gemini, Ollama, and DeepSeek, allowing users to configure via API, subscriptions, or self-hosted models. Architecturally, it implements a memory system inspired by the separation of threads and memory for source tracking and hybrid retrieval. The project is built using TypeScript and requires Node.js 22.19+ for development, supporting macOS, Windows, and Linux architectures.