KanBots: A Kanban System for Parallel AI Agents

KanBots is an open-source Kanban board that allows developers to run multiple AI agents in parallel across separate git worktrees. It features an Autopilot mode where specialized AI personas collaborate on tasks and a decision-driven UI that keeps humans in control of the process. The tool is local-first for maximum privacy but offers a cloud tier for teams needing real-time collaboration and synchronization.

Key Points

• Parallel Agent Execution: Enables multiple AI agents to work simultaneously on different tasks, each within its own git worktree.
• Autopilot and Personas: Uses specialized AI roles to automatically split complex features into manageable subtasks and manage the development lifecycle.
• Human-in-the-Loop UI: Agents pause and request user input for key decisions, providing a transparent and reviewable stream of work.
• Local-First Privacy: The open-source version keeps all code and data on the user's machine with no telemetry or external server requirements.
• Cost and Integration Management: Provides live cost analytics for LLM usage and integrates with GitHub, Sentry, and MCP-compatible tools like Cursor.

Sentiment

The overall sentiment is mixed and somewhat skeptical. Hacker News is interested in the idea of a local, open-source board for coordinating agents, and some commenters are already using similar workflows with success. However, the most substantial parts of the conversation push back on the article's implied emphasis on parallelism, arguing that trustworthy human review, testing, bounded work, and product judgment remain the unresolved constraints.

In Agreement

• A Kanban board is a natural control surface for agent work because cards already represent bounded tasks, status, context, and handoff points.
• Isolated worktrees and local-first storage are important foundations for safely running multiple agents without surrendering project data or polluting the main checkout.
• Agent orchestration can be useful for repetitive, well-scoped, or low-risk work where the desired outcome is clear and regeneration is cheap.
• People already running custom Jira, ClickUp, shell-script, or Vibe Kanban workflows see value in a dedicated open-source desktop tool that makes parallel sessions easier to observe.
• The approach could become more compelling if it exposes plans, logs, diffs, tests, and review queues in one place instead of forcing users to manage many terminals or chats.

Opposed

• The hard problem is not launching more agents, but reviewing and validating the work they produce without overwhelming the human operator.
• Unreviewed AI-generated code can look plausible while hiding subtle defects, especially in production systems, data analysis, and domain-heavy tasks.
• The term Kanban feels mismatched to some readers because Kanban traditionally limits work in progress, while the product pitch emphasizes parallel throughput.
• Several commenters view the product and landing page as too similar to other AI-generated SaaS-style tools and want stronger design, clearer open-source positioning, and fewer broken or confusing site details.
• The space already has many adjacent tools and custom workflows, so KanBots needs stronger differentiation around IDE integration, worktree environments, conflict handling, dependency management, and review ergonomics.