Code Agent Kanban

A real-time web dashboard that monitors AI coding sessions across multiple servers (currently supports Claude Code and Codex), displaying them as a kanban board with AI-powered task summaries.

If you run coding agents on multiple machines simultaneously, it's easy to lose track of what each agent is doing. This tool SSHs into your servers, collects sessions, and presents them in a single unified view.

Features

• Multi-server monitoring — SSH into remote servers to collect session data in parallel
• Live kanban board — Sessions organized into Running / Completed / Errors columns
• Multi-provider support — Switch between claude and codex providers from Settings
• AI-powered summaries — Sessions are summarized by local CLI (claude or codex) with task description, progress status, and estimated completion percentage
• Auto-refresh — Dashboard updates automatically; fast-polls on first load to pick up AI summaries quickly
• Session lifecycle tracking — Sessions move from Running to Completed when the Claude Code process exits; resumed sessions move back to Running
• Web-based configuration — Add/edit/remove SSH servers and test connections from the Settings panel, no config files needed
• Summary caching — Summaries are cached on disk; unchanged sessions won't be re-summarized, even across restarts

Prerequisites

• Python 3.11+
• uv package manager
• For claude provider: Claude Code CLI installed
• For codex provider: Codex CLI installed (codex)
• SSH key-based access to remote servers (for multi-server monitoring)

Quick Start

Install with uv (recommended)

uvx claude-kanban

Install with pip

pip install claude-kanban
claude-kanban

Run from source

git clone https://github.com/ShenJiahuan/claude-kanban.git
cd claude-kanban
uv run claude-kanban

Open http://localhost:5555 in your browser.

By default, it scans the local machine for Claude Code sessions. You can switch provider to Codex in Settings.

How It Works

1. Session discovery — Scans provider-specific session logs (~/.claude/... or ~/.codex/...) on each server
2. Conversation parsing — Extracts first task messages and latest progress messages from JSONL logs
3. AI summarization — Sends conversation excerpts (first 3 + last 6 messages) to local CLI (claude -p or codex exec)
4. Lifecycle tracking — Sessions move between Running and Completed based on activity/process state

Configuration

All configuration is done through the web UI (Settings button). Under the hood, settings are persisted to ~/.claude-kanban/config.yaml:

include_local: true
provider: claude  # claude | codex
servers:
  - host: gpu-server-1.example.com
    user: ubuntu
    label: GPU Server 1
  - host: 10.0.0.50
    user: root
    port: 2222
    key: ~/.ssh/id_ed25519
    label: Dev Box

Server options

Remote server requirements

• Python 3 installed
• For claude provider: Claude Code installed (sessions stored in ~/.claude/)
• For codex provider: Codex installed (sessions stored in ~/.codex/)
• SSH key-based authentication configured

Environment variables

API

License

MIT