kun

Kun is a local-first AI Agent workbench aimed at advancing AI from "answering questions" to "completing work." It offers two main modes: Code, focused on software delivery with an integrated Design canvas in the same task, and Work, for writing, information organization, document analysis, and presentation output. Agents can read workspace context, create plans, call tools, modify files, run validations, and leave evidence alongside tasks. The desktop GUI suits observation, review, and process control, while the terminal TUI suits keyboard-focused work. Both share threads, goals, plans, approvals, and background tasks through the same local `kun serve` runtime, rather than being separate, disconnected sessions. Key capabilities: - Code mode: Project context, file editing, terminal, Git/Worktree, Diff, testing, and review. - Design canvas: Capture prototypes, design systems, and Design-to-Code context within the same Code task. - Work mode: Edit Markdown, preview, reference, and analyze PDF/Office documents, analyze spreadsheets, and create presentations from outlines; Office files remain read-only. - Automation: Scheduled tasks, Loops, Hooks, MCP, Skills, and installable extensions. - Model integration: Subscriptions, plans, APIs, OpenAI/Anthropic-compatible services, and self-hosted models can be connected via Provider configuration, with presets covering ecosystems like ChatGPT/Codex, Claude, Gemini, Cursor, Ollama, DeepSeek, Kimi, GLM, Qwen, MiniMax, and Xiaomi MiMo. The workflow is: clarify goals → form a plan → execute and collaborate → check evidence → deliver or continue. Requirements and plans are saved in the project by default, facilitating version control, code review, and later recovery. Local-first means sessions, preferences, logs, and runtime data are stored locally by default; when choosing cloud models, prompts, attachments, and task context are sent to the selected provider. Tool permissions, sensitive operations, and extension permissions are clearly presented in the interface, leaving authorization to the user. Installation and running: Installers for macOS (.dmg/.zip, Apple Silicon/Intel), Windows (.exe, x64), and Linux (.AppImage/.deb, x64) can be downloaded from GitHub Releases. Running from source requires Node.js 22.19+, npm, and at least one working model connection, with common commands including `npm run dev`, `npm run dev:tui`, `npm run typecheck`, `npm run lint`, `npm run test`, `npm run build`, and platform-specific packaging commands. The license is PolyForm Noncommercial License 1.0.0, for learning, research, and non-commercial use only; commercial use, commercial distribution, SaaS/hosted services, resale, or integration into commercial products requires separate written authorization from the author.