Skip to content

Vibe Kanban - Technical Overview

What is Vibe Kanban?

Vibe Kanban is an open-source orchestration platform for AI coding agents, developed by BloopAI. It provides a web-based Kanban board interface that enables developers to manage, monitor, and execute multiple AI coding agents in parallel. The tool addresses the paradigm shift where AI agents increasingly write code while human engineers focus on planning, reviewing, and orchestrating tasks.

Key Value Proposition: Get 10X more out of Claude Code, Gemini CLI, Codex, and other coding agents by running them in parallel, isolated environments with centralized management.

High-Level Architecture

How It Works

Task Execution Flow

Git Worktree Isolation Model

Key Concepts

Task States

Tasks flow through the Kanban board with three primary states:

StateDescription
TodoQueued tasks waiting for execution
In ProgressActive tasks being worked on by agents
DoneCompleted tasks ready for merge/review

Task Attempts

Each task can have multiple "attempts" - separate executions with different:

  • AI agents (Claude, Gemini, Codex, etc.)
  • Agent variants/profiles (different prompts, models)
  • Base branches

Agent Profiles

Configurable agent variants allow customization of:

  • Planning behavior
  • Model selection
  • Sandbox permissions
  • Custom instructions

Technology Stack

Repository Structure

vibe-kanban/
├── crates/           # Rust backend modules
├── frontend/         # React web application
├── remote-frontend/  # Remote access interface
├── npx-cli/          # NPM distribution CLI
├── shared/           # Common code
└── Cargo.toml        # Rust workspace config

Supported AI Coding Agents

MCP (Model Context Protocol) Integration

Vibe Kanban exposes a local MCP server, enabling bidirectional integration:

MCP Configuration

json
{
  "mcpServers": {
    "vibe_kanban": {
      "command": "npx",
      "args": ["-y", "vibe-kanban@latest", "--mcp"]
    }
  }
}

MCP Tools Reference

ToolDescriptionRequired Parameters
list_projectsFetch all projectsNone
list_tasksList tasks in projectproject_id
create_taskCreate new taskproject_id, title
get_taskGet task detailstask_id
update_taskModify tasktask_id
delete_taskRemove tasktask_id
start_task_attemptStart agent on tasktask_id, executor, base_branch

Installation & Usage

Quick Start

bash
# Prerequisites: Node.js 18+
npx vibe-kanban

Development Setup

bash
# Clone repository
git clone https://github.com/BloopAI/vibe-kanban

# Prerequisites
# - Rust (latest stable)
# - Node.js 18+
# - pnpm 8+

# Start development server
pnpm run dev

Docker Deployment

bash
docker run -p 8080:8080 bloopai/vibe-kanban

Key Features

1. Parallel Agent Execution

Run multiple AI agents simultaneously without conflicts:

2. Code Review Interface

  • Line-by-line diff visualization
  • Comment and feedback system
  • Direct feedback to agents
  • Approve/reject changes

3. Real-time Monitoring

  • Live execution logs
  • Progress streaming
  • Pause/resume/stop controls
  • Step-by-step visibility

4. GitHub Integration

  • Automated PR creation
  • Branch management
  • Rebase conflict resolution
  • Direct merge support

Ecosystem Integration

Key Facts (2025-2026)

  • GitHub Stars: 14.3k+ stars
  • Contributors: 40+ contributors
  • Releases: 150+ releases (v0.0.146 as of Jan 2026)
  • License: Apache 2.0 (open source)
  • Languages: ~51% Rust, ~46% TypeScript
  • Architecture: Local-first, privacy-focused
  • Cost: Free and open source (you only pay for AI agent APIs)

Use Cases

1. Feature Development at Scale

Run multiple agents in parallel working on different features:

  • Agent 1: Implements backend API
  • Agent 2: Creates frontend components
  • Agent 3: Writes tests

2. Bug Fixing Sprints

Assign multiple bugs to different agents simultaneously:

  • Each bug gets isolated worktree
  • No conflicts between fixes
  • Merge when each is verified

3. Code Refactoring

Distribute refactoring tasks across agents:

  • One agent per module/component
  • Isolated changes prevent cascading conflicts
  • Review and merge incrementally

4. Research & Exploration

Use agents to explore solutions in parallel:

  • Different approaches to same problem
  • Compare results before choosing
  • Delete unused worktrees

Security Considerations

AspectImplementation
Local-FirstRuns entirely on your machine
No Code UploadCode never sent to external servers
Git IsolationEach agent in separate worktree
MCP Local OnlyMCP server accessible only locally
Permission ControlAgent sandbox permissions configurable

Comparison with Direct Agent Usage

FeatureDirect AgentWith Vibe Kanban
Parallel executionManual worktree setupAutomatic isolation
Task trackingMemory/notesVisual Kanban board
Code reviewTerminal diffVisual diff interface
Agent switchingRestart sessionOne-click switch
MCP configurationPer-agent setupCentralized config
Progress monitoringTerminal watchingReal-time dashboard

Resources

Technical research and documentation