Description
Vibe Kanban is a web-based Kanban board for orchestrating multiple AI coding agents, including Claude Code, Gemini CLI, Codex, Amp, GitHub Copilot, Cursor, and OpenCode, from the team behind the original Bloop code search engine. Moving a task card through To Do, In Progress, Review, and Done automatically spins up an isolated git worktree, starts the linked agent, and captures diffs and logs in the interface, so multiple agents can work on separate tasks in parallel without interfering with each other or the main branch. Teams can assign different agents to different task types, for example Claude Code for design-heavy features and Codex for tests and refactoring, and compare results side by side. As of this writing, Vibe Kanban’s own release notes state the product is sunsetting as a company-run service, with the project continuing as open source and community maintained going forward; the core codebase remains fully functional to self-host under its permissive license. It suits developers running multiple AI coding agents who want one shared board to plan, review, and orchestrate that work instead of juggling separate terminal sessions.
Key Features
- Multi-agent orchestration — run Claude Code, Codex, Gemini CLI, Copilot, Cursor, and others from one board.
- Isolated git worktrees — every task runs in its own worktree, preventing agents from interfering with each other.
- Kanban task tracking — move cards through To Do, In Progress, Review, and Done to control agent execution.
- Parallel and sequential execution — run agents on separate tasks simultaneously or chain them in sequence.
- PR-based review — captures diffs and logs, with an option to open a GitHub PR directly from a task.
- MCP integration — create and track tasks programmatically from Claude Desktop and other MCP clients.
- Per-task model profiles — assign a lighter or heavier model to a task depending on its complexity.
How It Works
Vibe Kanban launches locally with a single command, npx vibe-kanban, after a user has already authenticated with the coding agents they plan to use, such as Claude Code, Gemini CLI, or Codex. Once running, a browser-based Kanban board lets a developer create task cards describing what needs to be done and assign each one to a specific agent. Moving a card into In Progress spins up an isolated git worktree for that task and starts the assigned agent working inside it, so several agents can run at once without stepping on each other’s changes or the main branch. As agents work, their diffs and logs stream into the task card for review; approving a task can open a GitHub PR directly. Teams can also configure per-task model profiles, sending a simple fix to a fast, cheap model and a complex feature to a more capable one, and centralize each agent’s MCP configuration in one place rather than per project.
Technical Architecture & Overview
- Core Engine: Not one model; Vibe Kanban orchestrates existing coding agents (Claude Code, Codex, Gemini CLI, Amp, Copilot, and others) rather than running its own model.
- Deployment: Runs locally via
npx vibe-kanbanwith no install required, or self-hosted via Docker for production or team use. - API Surface: MCP server support lets external MCP clients, including Claude Desktop, create and update tasks programmatically.
- Known Limits: Per Vibe Kanban’s own official release notes, the product is sunsetting as a company-operated service, with some routes converting to an export-only page; the open-source codebase remains available for self-hosting under a permissive license.
Pros & Cons
| Pros | Cons |
|---|---|
| Free, open source, and works with most major coding agents rather than locking users to one | Vibe Kanban’s own release notes confirm it is sunsetting as a commercially operated service |
| Isolated git worktrees let multiple agents work in parallel without conflicts | Ongoing development shifts to community maintenance rather than a dedicated company team |
| Launches instantly with one command, no Docker or config files required for local use | Long-term roadmap and support are less certain now that the original team has announced sunsetting |
| MCP support lets other tools, including Claude Desktop, create and manage tasks directly | Requires separately authenticating with each underlying coding agent it orchestrates |
| Self-hosting keeps source code on a user’s own machine rather than a third-party cloud | Some product routes are being converted to export-only pages as part of the sunset process |
Pricing
| Plan | Price |
|---|---|
| Free | $0/month (1 user, core features, community support) |
| Pro | $30/user/month (2-49 users, 99.5% SLA, Discord support) |
| Enterprise | Custom (50+ users, SSO/SAML, 99.9% SLA, dedicated Slack channel) |
Platform Availability
Web (local or self-hosted) | macOS | Windows | Linux | VS Code
Best For
Developers running multiple coding agents | Teams doing agent-assisted development | Self-hosting enthusiasts
Frequently Asked Questions
Is Vibe Kanban still being developed?
Per Vibe Kanban’s own official release notes, the product is sunsetting as a company-operated service. The project will continue as open source and community maintained rather than under active commercial development.
Which coding agents does Vibe Kanban support?
Vibe Kanban orchestrates Claude Code, Gemini CLI, Codex, Amp, GitHub Copilot, Cursor, OpenCode, and other agents, letting a task be assigned to whichever agent fits it best.
How do I install Vibe Kanban?
Running npx vibe-kanban in a terminal downloads and launches it locally with no global install or Docker required; a self-hosting guide covers Docker deployment for teams.
Is Vibe Kanban free to use?
Yes, for individual users. The Free plan covers 1 user with core features at $0/month; Pro is $30 per user per month for teams of 2-49, per Vibe Kanban’s official pricing page.
Is Vibe Kanban related to Bloop?
Yes. Vibe Kanban comes from the same team that built the original Bloop AI code search engine; the company’s focus shifted from code search to orchestrating AI coding agents before Vibe Kanban itself began sunsetting as a commercial product.
Similar Tools in This Directory
Also in AI Developer Tools: KushoAI, self-healing API test generation | Eidolon AI, deploy agents as Kubernetes microservices | Sentry, AI-powered error monitoring and autofix
