03:12 am — shift in progress
Litwindow gives your AI agent workforce an office: a private Paperclip control plane with always-on OpenClaw workers, pre-wired on your own isolated VM at you.litwindow.ai. They keep working when your laptop doesn't.
7-day trial · your own VM · your own API keys · leave anytime
07:30 am
You already run agents. The problem is where they live: a laptop that sleeps, a tmux session that dies, keys scattered across machines, and no idea what happened overnight.
Everything is inspectable and nothing is held hostage. That's the deal.
Paperclip and your OpenClaw workers come already connected — heartbeats, invites, budgets. The wiring you'd otherwise do by hand is done before your first login.
API keys are written to your VM and discarded. Our database stores a fingerprint, nothing more. We can't leak what we don't hold.
Set a monthly cap per agent. When it's hit, the agent stops — Paperclip's native enforcement, surfaced in your dashboard.
Live log streaming for every worker, straight from your VM to your browser. Your logs never transit our servers.
Restart policies, health checks, and a watchdog agent. If something stays down, you get an email or Telegram ping — not a lost day.
One click exports your whole office — database dump, workspaces, compose files — as a tarball you can run anywhere.
One form, one boot, and your agents clock in.
Pick a subdomain, choose EU or US, paste your API keys, set a budget. Import your existing OpenClaw workspace if you have one.
~1 minA dedicated VM spins up from our golden image: Paperclip, your workers, TLS, backups — all wired.
~90 secWorkers send their first heartbeat, pick up their roles, and get to work. You watch from the dashboard — or close the laptop. That's the point.
immediatelyWe charge for the office, not the work. Model usage goes on your own API keys at provider rates — we never mark up tokens.
Why bring your own keys? Because it keeps our incentives clean: we make the same money whether your agents think hard or sit idle, so we'll never nudge you toward burning tokens — and your model bill stays at provider list price.
On your VM, in an env file readable only by root, delivered once at provision time (or when you rotate them). Our platform database stores a SHA-256 fingerprint so the dashboard can show which keys are configured — never the keys themselves.
Yes — it's the compliant way to run agents on a server. Since April 2026, Anthropic blocks third-party tools from riding Claude Pro/Max subscription logins; API keys under their Commercial Terms carry no such restriction. Your OpenClaw workers run on your keys, exactly as the providers intend.
OpenClaw workers are first-class today; Claude Code and Hermes workers are next on the roadmap. Paperclip itself is agent-agnostic — anything that can send a heartbeat can be attached, including your own scripts.
It stops. That's Paperclip's built-in enforcement — the agent halts until the month rolls over or you raise the cap. You'll see spend approaching the cap in your dashboard first.
Always. One click produces a tarball with your Postgres dump, agent workspaces, and the compose files to run it all yourself. Nightly encrypted backups run regardless.
In the EU — Nuremberg or Helsinki, on Hetzner. Each customer gets a dedicated VM — no shared containers, no noisy neighbors. US region is on the roadmap as a higher tier; latency rarely matters for agents that talk to APIs, not to you.
Nothing from us. Your model usage is billed by your providers (Anthropic, OpenAI, OpenRouter) on your own keys at their list prices. Most solo setups spend $20–100 a month on tokens; your caps keep that where you want it.
tonight
Boot your office in three minutes. Close the laptop. See what got done at breakfast.
Start your office — $29/mo