ai-agent-board

A visual dashboard for managing autonomous AI coding agents.

Kanban board • Real-time chat • Diff review • One-click PRs • MCP integration

What is agent-board?

agent-board is a local web dashboard that lets you create tasks, assign them to AI coding agents, and monitor everything from a Kanban-style interface — real-time execution logs, live chat with the agent, code diffs, and automatic PR creation.

It ships as a single native binary (~8 MB) — no runtime dependencies, instant startup.

It auto-detects the coding CLIs you already have installed:

| Agent | Command | Notes | |-------|---------|-------| | Claude Code | claude | Requires Anthropic API key or CLI login | | Codex | codex | Requires OpenAI API key or CLI login | | Gemini | gemini | Requires Google API key or CLI login | | GitHub Copilot | copilot | Requires GitHub Copilot subscription | | MiniMax | (API) | MiniMax-M2.5, M2.5-highspeed, M2.1, M2.1-highspeed, M2 | | OpenRouter | (API) | Use any model via OpenRouter API key |

No CLI installed? Use MiniMax or OpenRouter to access models through the API without any CLI setup.

Quick Start

npx ai-agent-board

That's it. The dashboard opens automatically in your browser. No configuration files, no Docker, no database setup.

Requirements: Node.js >= 18. At least one supported AI coding CLI installed, or a MiniMax/OpenRouter API key.

CLI Options

npx ai-agent-board [options]

  --port <number>   Server port (default: 51767)
  --no-open         Don't open browser automatically
  --clear-cache     Delete cached binary and re-download
  --version, -v     Show version
  --help, -h        Show help

Features

Kanban Board

Manage tasks across stages: To Do → In Progress → In Review → Done. Each task card shows its current status, assigned agent, and progress at a glance.

Task Workflow

Describe what you need, and the agent handles the rest:

 You describe a task        Agent works in an         You review the          Approve & create
 in natural language   ───▶  isolated worktree   ───▶  code changes    ───▶   a Pull Request

1. Create task — describe what you need in plain language
2. Agent codes — the AI agent works in an isolated git worktree, with full context of your repo's stack and conventions
3. Chat in real-time — send feedback, redirect the approach, or approve the agent's plan mid-execution
4. Review & merge — inspect the diff, then create a PR directly from the dashboard

Real-time Chat

Chat with the agent while it works. Send feedback, ask questions, or redirect its approach — all streamed live via SSE. The agent can present an implementation plan for your approval before writing any code.

Diff Viewer

Review all code changes before they go anywhere. Inline diff viewer shows exactly what the agent modified, added, or removed.

Git Integration

• GitHub — create PRs, review comments, request changes, and track merge status
• GitLab — connect via Personal Access Token
• Merge conflict resolution — open VS Code directly at the worktree to resolve conflicts

MCP Integration

agent-board exposes a Model Context Protocol (MCP) server, allowing AI assistants like Claude Code to manage tasks programmatically:

• Create, list, update, and delete tasks and repositories
• Start agents, send feedback, approve or request changes
• Review diffs and manage PRs — all through MCP tool calls

Repository Management

• Scan your filesystem to discover local git repositories
• Auto-detect tech stack and conventions per repo
• Track learned patterns across tasks for smarter specs

Credential Management

API keys and tokens are stored locally with AES-256-GCM encryption. Connect your providers from the Settings page — no .env files needed.

Dark Mode

Full light and dark theme support. Automatically follows your system preference.

How It Works

agent-board runs a native server binary that:

1. Detects which AI coding CLIs you have installed (Claude Code, Codex, Gemini, Copilot) or uses your MiniMax/OpenRouter API key
2. Assigns a coding agent to work on your task in an isolated git worktree
3. Streams real-time output, chat messages, and tool activity to the browser via SSE
4. Isolates code changes in git worktrees — your main branch is never at risk
5. Creates Pull Requests via the GitHub/GitLab API when you approve changes

All data is stored in a local SQLite database. Nothing leaves your machine except the API calls your AI coding CLIs already make and the git provider API calls for PR creation.

Task Lifecycle

| Status | Description | Board Column | |--------|-------------|:------------:| | draft | Task created, not yet started | To Do | | coding | Agent working on implementation | In Progress | | plan_review | Agent proposed a plan, waiting for your approval | In Progress | | review | Changes ready for review / PR created | In Review | | merge_conflicts | Merge conflicts detected, needs resolution | In Review | | changes_requested | You requested changes on the PR | In Review | | done | PR merged, task complete | Done | | canceled | Canceled by user or PR closed | — | | failed | Error during execution | — |

Architecture

┌──────────────────────────────────────────────────────────┐
│                     ai-agent-board                       │
│                                                          │
│  ┌──────────────┐  ┌───────────────┐  ┌───────────────┐ │
│  │  Dashboard    │  │  Server (Rust)│  │    Shared     │ │
│  │  React 19     │◄─┤  Axum + Tokio │  │  Zod Schemas  │ │
│  │  Vite 7       │  │  rusqlite     │  │  TypeScript   │ │
│  │  TanStack     │  │  SSE + Chat   │  │  Types        │ │
│  │  Tailwind 4   │  │  MCP Server   │  │               │ │
│  └──────────────┘  └───────┬───────┘  └───────────────┘ │
│                            │                             │
│               ┌────────────┴────────────┐                │
│               │    Agent Orchestrator   │                │
│               │   Task ──▶ AI Agent     │                │
│               └────────────┬────────────┘                │
│                            │                             │
│         ┌──────────┬───────┼────────┬──────────┬─────────┐│
│         ▼          ▼       ▼        ▼          ▼         ▼│
│     claude      codex   gemini   copilot   MiniMax  OpenRouter│
└──────────────────────────────────────────────────────────┘

Tech Stack

| Layer | Technologies | |-------|-------------| | Frontend | React 19, Vite 7, TypeScript, Tailwind CSS 4, shadcn/ui, TanStack Router & Query, Zustand | | Backend | Rust, Axum, Tokio, rusqlite (bundled SQLite), SSE, MCP | | Shared | Zod schemas, TypeScript types, npm workspaces | | CLI | Lightweight npx wrapper, platform-specific native binary distribution via Cloudflare R2 |

Key Design Decisions

• Rust backend — single native binary (~8 MB), instant startup, low memory usage, no runtime dependencies
• SSE over WebSockets — unidirectional streaming is sufficient for logs and chat; simpler, works with any proxy, and browsers auto-reconnect
• Bundled SQLite — zero-config embedded database compiled into the binary, no external server needed
• Git worktrees — agent works in isolated worktrees so your main branch is never at risk
• npx distribution — single command to run, native binary auto-downloaded and cached per platform
• MCP server — enables AI assistants to create tasks, manage repos, and orchestrate agents programmatically

Development

# Clone and install
git clone https://github.com/ezeoli88/ai-agent-board.git
cd ai-agent-board
npm install

# Run dashboard in development mode
npm run dev:dashboard

# Run Rust server in development mode
npm run dev:server-rs

# Build frontend
npm run build:dashboard

# Build Rust server (release)
cd packages/server-rs && cargo build --release

The dashboard dev server runs on port 3003 and the backend on port 51767. Vite proxies /api requests to the server automatically.

Environment Variables

Most configuration is handled through the Settings page in the dashboard. Environment variables are optional and mainly useful for development:

PORT=51767                          # Server port (default: 51767)
DATABASE_PATH=./data/agent-board.db # SQLite database path
LOG_LEVEL=info                      # debug | info | warn | error

AI keys and git tokens are managed through the dashboard UI with encrypted local storage — no .env file needed for normal usage.

Contributing

Contributions are welcome. Please open an issue first to discuss what you'd like to change.

1. Fork the repository
2. Create your branch (git checkout -b feature/my-feature)
3. Commit your changes (git commit -m 'Add my feature')
4. Push to the branch (git push origin feature/my-feature)
5. Open a Pull Request

License

MIT © Ezequiel Olivera