Open source Kanban desktop app that runs parallel agents on every card

The creators of an open source desktop Kanban app have launched a new version that enables users to dispatch multiple AI agents to work in parallel on individual cards, supporting both solo and collaborative workflows. This development allows users to automate tasks and decision-making directly within their project boards, without relying on cloud services.

The new Kanban app, available on GitHub under the MIT license, supports running multiple AI agents—such as Claude Code or Codex—on each card, with each agent operating in its own git worktree. Users can drop a folder to generate a board, assign agents to cards, and monitor their progress live. The system is designed for local-first operation, storing all data in a directory next to the project repository, with no cloud dependencies or telemetry.

Key features include autopilot mode, which automatically splits work and manages multiple personas (product, engineer, reviewer, tester) in up to four parallel slots. Users can set cost budgets, track live analytics, and integrate with GitHub issues and draft pull requests. The app provides a full UI for dispatching, reviewing, splitting, and shipping agent work, making AI-driven automation accessible and manageable directly from the desktop environment.

Why It Matters

This development is significant because it introduces a flexible, privacy-conscious tool that integrates AI automation into project management workflows without requiring cloud infrastructure. It enables solo developers and teams to leverage AI agents for complex tasks like splitting issues, reviewing code, or managing backlogs, potentially increasing productivity and consistency. The open source nature allows customization and local operation, appealing to users concerned about data privacy and control.

Background

Traditional Kanban tools focus on visual task management but rarely incorporate AI automation at this level. Existing AI integrations often rely on cloud APIs, raising privacy and cost concerns. The recent release builds on ongoing interest in AI-assisted development workflows, especially as open source tools become more capable. Prior developments include AI-driven code review tools and automation scripts, but this project uniquely combines a Kanban interface with parallel AI agents operating locally, supporting both solo and collaborative environments.

“Our goal was to create a local-first, open source Kanban system that empowers users to run multiple AI agents in parallel, directly on their machines, without cloud dependencies.”

— Project developer

“Supporting multiple personas and parallel agents transforms how teams and individuals can automate complex workflows within a familiar Kanban interface.”

— Open source contributor

What Remains Unclear

Details about scalability, performance, and integration with other tools are still emerging. It is not yet clear how well the system performs with large projects or in highly collaborative environments. The extent of customization for different AI providers and future feature roadmap remains to be seen.

What’s Next

Developers plan to expand features, improve user interface, and enhance integration with existing development workflows. Further updates are expected to include more personas, advanced cost management, and broader provider support. Community feedback will likely shape future iterations.

Key Questions

Can this app be used for team collaboration?

Yes, the app supports team workflows, especially with cloud features like GitHub integration. It is also designed for solo use, with local-first operation that does not require cloud accounts.

What AI providers are supported?

The app supports Claude Code and Codex, with the ability to switch providers per dispatch. Users can also bring their own CLI tools.

Is this software free and open source?

Yes, it is MIT licensed, open source on GitHub, and free to use both for individuals and teams.

How does the parallel agent system work?

Up to four agents can run in parallel, each claiming work in a round-robin fashion, splitting parent issues into subtasks, and working simultaneously on different aspects of the project.

Source: Hacker News