Verified project record

KunAgent/Kun

kunagent/kun is a local-first AI agent workspace that combines coding, writing, design, research, and automation into a single shared runtime accessible through both a desktop GUI and a terminal TUI. It is designed to keep tasks visible, controllable, and traceable as users advance ideas into deliverables.

Project overview

The project addresses the friction of manually moving context across disconnected tools by offering one continuous runtime shared between a desktop GUI and terminal TUI, with experimental Agent Graph orchestration for supervised dependency management in complex tasks.

Project type: AI Agent · AI Coding
Use cases: Coding & Development
Deployment: Refer to project documentation
License: License pending

Best for

Developers, creators, and researchers who work across coding, writing, design, research, and automation and want a single controllable workspace.
Users who prefer switching between a desktop GUI and a terminal TUI while maintaining continuous task state.

Key capabilities

Provides a single `kun serve` runtime shared by the desktop GUI and terminal TUI to operate tasks continuously.
Searches code, edits files, executes commands, manages plans, and views diffs and tests for code changes and reviews.
Handles writing, polishing, data organization, and editing to export Markdown, HTML, PDF, DOCX, and editable PPTX.
Explores visual directions, generates interactive prototypes, and builds design systems for implementation.
Reads PDFs and images, extracts evidence, and organizes structured conclusions and research notes.
Coordinates cross-file and cross-stage tasks using dependency graphs with supervised sub-agents and verification. This capability is experimental.
Supports one-time or periodic schedules, visual loops, and hooks for continuous execution.
Manages multiple subscriptions, plans, APIs, and compatible services in one entry to select models by task.

Limitations and risks

Strictly limited to learning, research, and non-commercial use without separate authorization.
Choosing a cloud provider will send prompts, attachments, and task context to that provider. Users should evaluate this data boundary before selecting a cloud model provider.

Getting started

Download the installer from GitHub Releases, install the application, then open the desktop GUI and configure a model Provider. First launch requires choosing a language, logging in to a model, and opening a project. At least one supported subscription, API, or custom Provider is required.

Evidence and sources

Open original GitHub project