Getting started

MeshBoard turns machines you already own into a working mesh: they do real work — code, research, long-running jobs — using the AI accounts you already pay for, and everything comes back with proof. Setup is one sign-in, one command, about ten minutes.

What you need

• One machine that's usually on — a Mac, a Linux box, a PC, a NAS. More can join later; one is enough to start.
• An AI account the machine can use: a Claude or ChatGPT subscription signed in on that machine, or an API key. MeshBoard adds no per-token cost on top — your accounts, your rates.
• A browser for the other end. (There's a phone app too — it's in beta; a browser is all you need to start.)

1. Sign in

Go to app.meshboard.ai and create your account. You land on your mesh — empty for now. That page is the product: your machines will appear there as stations, and the work moves between them as light.

2. Add your first machine

Open Mesh → Add a machine. You'll see a one-line installer command, scoped to your account — copy it, run it in a terminal on the machine you're adding. It downloads the agent, verifies it, pairs the machine to your account, and starts the agent as a background service. No configuration file, no ports to open.

Install commands are single-use and expire after about thirty minutes — if one sits too long, the installer will tell you plainly and Add a machine mints a fresh one.

Within a minute the machine appears on your floor with a live status. The agent's log lives at ~/.meshboard/agent.log on the machine if you ever want to watch it think.

3. Give it a way to work

The agent works through the AI tools on that machine. If Claude Code or Codex is already signed in there, the mesh can use it immediately — the credentials never leave the machine, and nothing routes through us. That's the design, not a promise: your AI accounts sign in on your devices, and MeshBoard orchestrates locally.

No CLI signed in on that machine? See Connect your AI for every supported path — subscriptions, local models, and your own API keys.

4. Ask for something

Press Ask (or / on a keyboard) and describe the work the way you'd text a colleague — "fix the RSS feed on my site", "summarize what changed in this repo this week". Your mesh picks it up, a light starts moving, and you can watch or walk away. When it's done you don't get a "done ✓" — you get the artifact, the receipts, and the trail, all on the same page you're already looking at.

Where to go next

• Your machines — platforms, roles, and adding more.
• Connect your AI — every way to give the mesh a brain.
• Runs and proof — what "done" means here.
• Troubleshooting — real failure modes and their fixes.