Kun: Local-First AI Agent Workspace

Overview

Kun is a desktop and terminal workspace for directing AI agents through practical work, from editing code to preparing documents. It aims to keep planning, tool use, approvals, and evidence tied to the task instead of splitting them across unrelated chat sessions.

It may suit people who want an agent workspace they can inspect and control locally, with options for individual work and multi-agent collaboration. It is more involved than a standalone chat client, and its value depends on having a compatible model provider configured.

Key Features

• Shared local runtime for the desktop GUI and terminal TUI, so tasks and approvals can continue across interfaces.
• Code workflows for project context, file edits, terminal commands, Git and worktrees, diffs, tests, and reviews.
• Design canvas integrated into Code tasks, connecting design work with implementation context.
• Work mode for Markdown, document analysis, spreadsheets, presentations, and whiteboard-style planning.
• Rooms for agent discussions, delegated work, proposals, and review, with execution tasks isolated in Git worktrees.
• Automation and extension options including scheduled tasks, Loops, Hooks, MCP, Skills, and installable extensions.
• Provider configuration for subscriptions, APIs, compatible services, and self-hosted models.

Use Cases

• Developers can use Code tasks to ask an agent to make a bounded change, run checks, and review the resulting diff.
• Designers and developers can move from a design canvas to implementation within the same task when a feature needs both.
• Writers and researchers can use Work mode to draft Markdown, analyze supplied documents, or organize research materials.
• Teams coordinating agent work can use Rooms when tasks benefit from delegation, isolated changes, and review before integration.
• People automating repeatable work can explore scheduled tasks, Hooks, Loops, MCP, and Skills when manual workflows become repetitive.

Getting Started

For a source-based development setup, the README lists Node.js 22.19+ and npm as requirements:

git clone https://github.com/KunAgent/Kun.git
cd Kun
npm ci
npm run dev

See the README for desktop downloads, provider setup, and further commands. Documentation is available at kun-agent.com/docs.

Alternatives

• OpenMausBot: OpenMausBot presents agents as chat contacts and connects them to computers and third-party apps, while Kun also targets writing, design, research, and automation.
• Orkas: Orkas centers on a Commander directing specialist agents through one chat, while Kun provides a broader workspace with both desktop GUI and terminal TUI.
• holaOS: holaOS focuses on connecting agents, apps, integrations, and shared memory, while Kun emphasizes a shared runtime across desktop and terminal interfaces.
• synara: Synara focuses on local coding-agent sessions with files, terminals, previews, and Git review, while Kun supports broader work beyond software development.

Considerations

• Kun is local-first, but choosing a cloud model sends prompts, attachments, and task context to that provider. Check the provider's data policy before use.
• A compatible model connection is required to use agents; available models and access depend on provider configuration.
• The README states that the project uses the PolyForm Noncommercial License 1.0.0 and requires separate written authorization for commercial use. The repository metadata supplied here reports the license as NOASSERTION, so verify the current license terms before adopting it.
• The README notes that the TUI is bundled with the desktop application rather than distributed as a separate archive from version 0.3.8 onward.