Skip to content

Logesh-waran2003/agent-hub

Repository files navigation

AI Agents Hub

A web-based workspace for running and coordinating multiple AI coding agents (Claude Code, kiro-cli, or any CLI tool) in parallel — with live terminal streaming, cross-agent context injection, and a real-time Kanban board.


Why

When you're running 3+ AI agents in parallel, the problems are:

  • Switching between terminal windows to check progress
  • Copy-pasting output from one agent into another's context
  • No visibility into what each agent is doing at a glance

Hub solves all three.


Key Features

@mention Context Bridge

Type @agent-name in any session's input — Hub automatically injects that agent's last terminal output as context before sending. No copy-paste, no window switching.

You: @researcher summarize what you found and write the tests

Hub intercepts the message, appends the last output of @researcher's terminal, and sends the enriched prompt to the current agent.

Live Terminal Streaming

Real PTY sessions (via node-pty) streamed to the browser over WebSockets. Full xterm.js rendering with resize support. Not a tmux wrapper — actual terminal emulation in the browser.

Real-time Kanban Board

Agents write a PROGRESS.json file as they work. Hub watches it and updates the board live — no refresh needed. See which tasks are done, in progress, and upcoming across all agents.

Tabs + Sessions

Organize agents into tabs. Each tab can have multiple terminal panes in a resizable grid. Sessions persist across page reloads (SQLite-backed). Auto-restart on crash.

Task Queue

Create tasks, assign them to sessions, track status — separate from the agent-driven Kanban.


Getting Started

git clone https://github.com/Logesh-waran2003/agent-hub
cd hub
npm install
npm run dev

Open http://localhost:3000.

Keyboard shortcuts:

  • N — new session
  • K — toggle Kanban board
  • T — toggle light/dark theme

Kanban Board (PROGRESS.json)

Tell your agents to write progress to a JSON file and Hub will display it live.

Set the path via env:

PROGRESS_FILES=/path/to/project/PROGRESS.json npm run dev

Multiple files (comma-separated):

PROGRESS_FILES=/project-a/PROGRESS.json,/project-b/PROGRESS.json npm run dev

Required shape:

{
  "phase": "Phase 2 — Backend",
  "current_task": "Implementing auth middleware",
  "status": "active",
  "updated_at": "2026-03-07T12:00:00Z",
  "columns": {
    "done": [{ "id": "1", "title": "Database schema" }],
    "in_progress": [{ "id": "2", "title": "Auth middleware", "subtasks": [
      { "id": "2a", "title": "JWT validation", "status": "done" },
      { "id": "2b", "title": "Session store", "status": "in_progress" }
    ]}],
    "todo": [{ "id": "3", "title": "API endpoints" }]
  }
}

Stack

  • Frontend: Next.js 16, xterm.js, socket.io-client, Zustand, Tailwind v4, Framer Motion
  • Backend: Express + socket.io, node-pty, better-sqlite3
  • Runtime: Node.js 20+

License

MIT

About

Web workspace for running multiple AI coding agents in parallel — live terminals, @mention context injection, real-time Kanban board

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors