kansh

A local, read-only dashboard for your Claude Code sessions.

Run many Claude Code sessions at once? kansh auto-detects every one on your machine and shows them on a single screen — live — so you stop switching windows to check what each agent is doing.

• 🗂 One card per session — mission, progress, health, and the exact question it's waiting on.
• 🧭 Timeline view — every session as a swimlane on one shared clock; see cross-project concurrency at a glance.
• 🔀 Workflow map — the MK agentic flow (brainstorm → plan → cook → review → journal) with loop counts and the sub-agents each phase spawned.
• ⏵ Workflow replay — pick any task (one per prompt) and replay it at high speed: phases light up and sub-agents fade in as they happened, with play/pause, speed, and a scrub bar.
• ⚠️ Conflict alerts — a red banner when two live sessions edit the same file, so agents don't step on each other.
• 🔒 Read-only & local — never writes to ~/.claude, binds to 127.0.0.1 only, no telemetry.

_{Cards view — mission, workflow map (phases + sub-agents), health badges, and file activity per session.}

Quick start

git clone https://github.com/phuc-nt/kansh
cd kansh
bun install
bun run build
bun run start          # → http://127.0.0.1:4777

Requires Bun ≥ 1.3 and Claude Code (macOS). Zero config — no hooks, no plugins.

_{Timeline view — every session as a swimlane on one shared clock, with prompt/error/blocked markers, waiting stretches, and an attention ribbon.}

_{Workflow replay — pick a task and watch its phases reveal over time; the current phase pulses while later ones stay dimmed until reached.}

Documentation

• 📖 Installation & Usage — full setup, every feature, configuration, troubleshooting.
• 🏗 System Architecture — how the tailer, parser, liveness, and views fit together.
• 📝 Changelog — what changed in each release.

Contributing

Contributions welcome — see CONTRIBUTING.md. The core promise is that kansh stays local, read-only, and offline; please keep it that way.

License

MIT © phuc-nt