Archived legacy site. These historical setup instructions are not current. See the AgentsDock site or current source guide.

ZenithDock

Your Lab Never Sleeps.
Neither Does Your Agent.

Schedule AI jobs from your phone. Get results streamed back while you're away from the lab.

🖥 macOS 14+ ⬜ iPadOS 17+

Built for Researchers Who Hate Babysitting Terminals

📅

Schedule Jobs

Queue experiments to run every N minutes, hours, or days. Kick off an eval sweep before you leave the lab — results are waiting when you're back.

🎬

Results Stream Back

Rendered videos, plots, and output artifacts land directly in your timeline the moment the agent writes them. No SSH tunnel. No stale screen session.

⚡

Live Run Trace

Every tool call, code execution, and reasoning step streams in real time over WebSocket. Know exactly what your agent is doing — and when it's going off the rails.

💬

Per-Experiment Chats

Separate chats per experiment, organized into folders. Fork a promising run into a new branch. Your agent history is actually navigable for once.

📄

Push Config & Data Anywhere

Drop a config file, checkpoint path, or prompt update from your phone mid-commute. The server picks it up immediately — no laptop required.

🔒

Private by Default

Bearer token auth on every API call, upload, and WebSocket stream. Self-hosted means your model outputs and data never touch a third-party server.

How It Works

You run a lightweight server on the machine with compute. ZenithDock is the remote control.

â‘ 

Server runs on your machine

ZenithBotServer sits on your lab workstation, cloud VM, or wherever Claude/Codex CLI is installed and authenticated.

→
â‘¡

You send jobs from your phone

Open ZenithDock on iPhone, iPad, or Mac. Type a prompt, schedule a loop, or attach a file. The server picks it up instantly.

→
â‘¢

Results stream back live

Agent output, videos, plots, and artifacts appear in your timeline in real time. Review results anywhere, even offline-first with local caching.

Quick Setup

1

Install Claude CLI or Codex CLI on your server machine

The server needs at least one AI backend installed and authenticated:

# Claude (recommended)
npm install -g @anthropic-ai/claude-code
claude login

# Codex
npm install -g @openai/codex
export OPENAI_API_KEY=your-key
2

Clone & Install ZenithBotServer

On the same machine (Linux or macOS):

git clone https://github.com/ZhengyiLuo/ZenithBotServer
cd ZenithBotServer
uv venv && source .venv/bin/activate
uv pip install fastapi "uvicorn[standard]" python-multipart pydantic

No uv? Install it: curl -LsSf https://astral.sh/uv/install.sh | sh

3

Start the Server

uv run python agent_server.py serve --bind 0.0.0.0 --port 7850

Verify it's alive:

curl http://localhost:7850/api/health
# → {"ok": true, ...}
4

Run as a Service (Optional but Recommended)

Keep it running across reboots with systemd (Linux):

cp systemd/zenithbot-agent.service ~/.config/systemd/user/
systemctl --user enable --now zenithbot-agent.service
5

Connect ZenithDock

Open the app → Inspector panel → Security → set your server URL:

http://<your-server-ip>:7850

Click Apply & Reconnect. Green dot = you're live. Create a new chat and start running jobs.

💡 Use Tailscale to reach your server from anywhere without exposing port 7850 to the internet.

Documentation