Your control plane,
already wired.

Paperclip hands out the work, records what it costs, and stops an agent at its budget cap. litwindow runs a private instance of it on your own VM, with your workers registered before you log in.

you.litwindow.ai

Rack your office — $29/moread the service record

PINNED v2026.824.1ITS OWN POSTGRESHARD BUDGET CAPSYOU ARE THE OWNER

The figures in this panel are illustrative sample data, not litwindow's measured performance or any customer's real usage.

PAPERCLIP OPENCLAW HERMES CODEX
06:41:02 paperclip: task #318 assigned to openclaw-106:41:09 openclaw-1: heartbeat ok (1.4s)06:42:55 paperclip: budget check: $18.40 / $50.00 ok06:43:12 hermes: heartbeat ok (0.9s)06:47:30 paperclip: task #318 closed · 12 files changed
WORKERS REGISTERED3
MONTHLY BUDGET · HARD CAP$18.40 / $50.00

DEMONSTRATION READOUT — YOUR OFFICE, YOUR NUMBERS

What Paperclip is

Paperclip is the open-source control plane that assigns tasks to your agents, tracks what they spend, and enforces a hard monthly budget cap per agent; litwindow runs a private instance of it on your own VM, already wired to your workers.

It holds your organization and the people and agents in it. Workers register with it and send heartbeats, so it knows which of them are alive; tasks are assigned through it, and what each one spends is recorded against the cap you set. Daily work happens in Paperclip's own interface — the litwindow dashboard is for the infrastructure underneath it: health, logs, keys, SSH, plan, export, delete.

Paperclip is not litwindow's software. It is an open-source project, and what we sell is the machine it runs on and the wiring around it: a tested pin, its own database, a certificate, workers already joined, backups, and a way out. The instance is yours — including the owner account and everything in the database.

You could run it yourself

Paperclip is open source, so nothing stops you. If you have already tried, you know what the work actually is — and that most of it is not the control plane.

Self-hosted, by you

  • Self-hosted, by you: You provision the machine, the database, and the certificate
  • Self-hosted, by you: Every worker is joined to the org by hand
  • Self-hosted, by you: Upgrades mean reading the migration notes before you pull
  • Self-hosted, by you: Backups are a cron job you have never restored from
  • Self-hosted, by you: When a container stops at 3am, nothing notices until you do

In your litwindow rack

  • On litwindow: Paperclip v2026.824.1, baked into a tested golden image
  • On litwindow: Its own Postgres 16 in the same compose stack, on your VM
  • On litwindow: Workers registered and heartbeating before your first login
  • On litwindow: Nightly offsite backups, and one click gives you the whole thing
  • On litwindow: Restart policies and health checks watch the stack, not you

Pre-wired before you log in

These are not instructions. They are what happens between checkout and your first assigned task, and most of it happens without you.

  1. Your office bootsAbout five minutes. The VM comes up from the golden image, Caddy takes a certificate for your subdomain, and Paperclip runs its own database migrations against a fresh Postgres 16 before it serves a single request.
  2. You claim the owner linkThe dashboard shows a one-time owner claim link, valid for seven days. Open it, create your account, and that link makes you the Paperclip owner — not litwindow. It is removed from our systems the moment it is used.
  3. Set the budget capPick a monthly ceiling per agent. Paperclip enforces it natively: at the cap the agent stops until the month rolls over or you raise it. Your dashboard shows the spend climbing toward it first.
  4. Connect a brainPush an Anthropic, OpenAI, or OpenRouter key from the dashboard, or sign in to a sanctioned subscription — the Codex worker runs as Paperclip's codex_local adapter and accepts either an OpenAI key or a ChatGPT login via device auth.
  5. Your workers are already registeredOpenClaw and Hermes are attached to the org and sending heartbeats before your first login. Nothing about the join is left for you to do by hand.

Unit specification

The pins and paths below are the real ones, taken from the image and compose files your office is built from.

Version
Paperclip v2026.824.1 — the paperclipai npm package on its stable channel. The pin lives in image/versions.env; migrations run automatically on startup, and the 0196–0222 range was verified additive at this pin.
Database
Its own Postgres 16, in your tenant compose stack, on your VM. The paperclip_data and paperclip_home volumes are durable, so CLI logins made inside the container survive a container recreation.
Access
Served on port 3100 behind Caddy at https://yourname.litwindow.ai. TLS terminates at Caddy on your machine, not on ours.
Workers
OpenClaw and Hermes arrive attached, and a Codex worker runs as Paperclip's codex_local adapter — the Codex CLI inside the Paperclip container. Beyond those, Paperclip is agent-agnostic: anything that can send a heartbeat can attach, including your own scripts.
Budget enforcement
Paperclip's own, surfaced in your litwindow dashboard. At the cap the agent stops until the month rolls over or you raise the cap.
Custody
The one-time owner claim link is valid for seven days and is removed from litwindow's systems the moment it is used. Your ANTHROPIC_API_KEY, OPENAI_API_KEY and OPENROUTER_API_KEY are written to the VM's compose .env at first boot and never stored by the platform — our database holds a truncated SHA-256 fingerprint, nothing more.
Backups and exit
Nightly restic backups offsite, plus a one-click export: the Postgres dump, your agent workspaces, and the compose files to run the same stack yourself.
Daily use
In Paperclip's own UI. The litwindow dashboard exists for the infrastructure around it — health, logs, keys, SSH, plan, export, delete.

Service record

Is Paperclip litwindow software?

No. Paperclip is an open-source project; litwindow hosts a private instance of it for you on your own VM, pinned at a version we have tested and wired to your workers before you log in. We did not write it and we do not resell it as ours.

Who owns my Paperclip instance?

You do. After provisioning, your litwindow dashboard shows a one-time owner claim link, valid for seven days. Opening it makes you the Paperclip owner, and the link is removed from litwindow's systems the moment it is used.

What happens when an agent hits its budget cap?

It stops. That is Paperclip's own native enforcement, not a litwindow wrapper around it: the agent halts until the month rolls over or you raise the cap. The litwindow dashboard surfaces the spend so you see it approaching the cap first.

Which agents can attach to my Paperclip?

Paperclip is agent-agnostic: anything that can send a heartbeat can register with it, including your own scripts. Your office ships with OpenClaw and Hermes attached as always-on workers, plus a Codex worker running as Paperclip's codex_local adapter inside the container.

Which version of Paperclip do you run?

Paperclip v2026.824.1, the paperclipai npm package on its stable channel. The pin lives in image/versions.env and is baked into the golden image, and image/build.sh is the only entrypoint that bakes one — so the version on your machine is a version that was tested against our integration first.

Where does Paperclip keep my data, and can I take it with me?

In its own Postgres 16 on your VM, alongside durable paperclip_data and paperclip_home volumes. Nightly restic backups run offsite, and one click produces an export with the Postgres dump, your agent workspaces, and the compose files needed to run the same stack yourself.

The rest of the rack

Every office ships with all of these mounted and wired to each other. You do not pick one.

Boot the control plane.

Five minutes to a running Paperclip you own, with the workers already reporting to it.

Rack your office — $29/mo