KunAgent/Kun

Local-first AI agent workspace for coding, writing, design, research, and automation — one runtime for desktop GUI and TUI.

What it solves

Kun addresses the gap between simple AI chat interfaces and the actual delivery of a finished project. It provides a unified workspace where users can move from initial requirement clarification to design, coding, writing, and final acceptance testing without manually transferring context between disparate tools like editors, terminals, and document editors.

How it works

Kun operates as a "local-first" AI Agent workbench. It uses a shared runtime (kun serve) that connects a desktop GUI and a terminal-based TUI, allowing users to switch interfaces while maintaining the same session, plan, and task history. It supports multiple model providers (API, subscriptions, or self-hosted) and offers two primary interaction modes:

  • Direct Mode: For simple questions and quick modifications.
  • Agent Graph: An experimental mode for complex, multi-step tasks where a Lead Agent creates a dependency graph and delegates work to restricted sub-agents, ensuring each step is verified before proceeding.

Who it’s for

It is designed for developers, writers, designers, and researchers who need to transform ideas into verifiable results through AI-assisted coding, writing, prototyping, and automation.

Highlights

  • Unified Workbench: Integrated environments for Code, Write, Design, and Research within one product.
  • GUI + TUI Synergy: A single runtime allows seamless switching between a visual desktop app and a keyboard-centric terminal interface.
  • Model Agnostic: Supports a wide range of providers including OpenAI, Anthropic, DeepSeek, Ollama, and others via API or subscription.
  • Advanced Task Management: Features include /plan for breaking down goals, inline diffs for code review, and automated schedules/loops.
  • Extensibility: Supports MCP (Model Context Protocol), custom Skills, and .kunx extensions.