Paperclip hosting guide

How to host Paperclip 24/7, and the cost line most guides miss.

Most Paperclip hosting guides price a single VPS and stop. That is the control plane only. Paperclip does not run agents, it manages them, so a working deployment has a second compute bill underneath it that nobody puts on the invoice. This guide walks the whole stack honestly, including the part we do not sell you.

Direct answer

What does it cost to host Paperclip?

Four lines, not one. The control plane is a Node server that has to stay online. Production needs a Postgres database rather than the embedded one used for local development. Storage is local disk or an object store. Then there is the line most guides skip: the agents Paperclip manages need compute of their own, supplied by a sandbox-provider plugin or by whatever machine already runs your agent runtime. Only the first line appears in a typical VPS price comparison.

Control plane
A Node server that stays online
Database
Bring your own Postgres in production
Storage
Local disk or an object store
Usually missed
Compute for the agents themselves

The four cost lines of a Paperclip deployment

Sourced from Paperclip's own deployment and adapter documentation. Prices vary by provider, so this compares what each line is and who supplies it, not a monthly figure.

LineWhat it isWho supplies it
Control planeA Node server running the Paperclip app and scheduler.You: a VPS, Docker host, or platform of your choice.
DatabasePostgres in production; the embedded one is for local use.You: managed Postgres or self-run.
StorageArtifacts and workspace files, on disk or object storage.You, or an object storage provider.
Agent computeThe environments agent runs actually execute in.A sandbox provider, or the host your runtime already lives on.
Model tokensWhat the agents spend doing the work.Your model provider, on any hosting path.

Line one: the control plane is a server

Paperclip runs as a Node process with a web UI. The documented requirements are Node.js 24.11 or newer with pnpm, and the quickest install is their script. Locally that is fine on a laptop. For anything always-on it needs a host that does not sleep, which puts it in the same small always-on VPS class we price on the OpenClaw hosting page, or in Docker on a server you already run.

The documented install

curl -fsSLO https://paperclip.ing/install.sh
bash install.sh

Sources: Paperclip installation guide

Line two: the database stops being free

Paperclip ships an embedded Postgres so a local install works with no setup. That is a development convenience, and their own deployment reference tells you to bring your own database for production. That means a managed Postgres, or a database you run and back up yourself. Either way it is a second thing that has to be online, and the first thing you will wish you had backed up.

Sources: Paperclip database reference

Line three: storage and secrets

Artifacts and workspace files land on local file storage by default, or in provider-backed object storage if you configure it. Secrets are a separate decision: the agents need provider API keys, and Paperclip supports secret scopes and folders as well as an external vault. None of this is expensive, but all of it is setup, and it is the part that decides whether a compromised agent can read every key you own.

Sources: Paperclip storage and secrets reference

Line four: the agents need somewhere to run

This is the line that makes single-VPS comparisons wrong. Paperclip orchestrates agents; it does not execute them. Work goes to a runtime through an adapter, and when the run needs its own environment, Paperclip provisions it through a sandbox-provider plugin. Their documented providers today include Cloudflare, Daytona, exe.dev, E2B, Modal, Novita, and Kubernetes. Each of those has its own bill, usually metered by the second or by the container. So the honest question is not what a Paperclip VPS costs; it is what your agents cost to run, plus a small server to govern them.

Sources: Paperclip sandbox providers

Line five, unbilled: your time

Self-hosting a control plane is the same shape of work as self-hosting anything else. Provisioning, locking down access, wiring the database, configuring storage and secrets, keeping it patched, and noticing when the scheduler stops. Their docs cover Docker, a VPS or Fly deployment, and Tailscale for private access, which tells you the honest scope: this is an operations project, not an install. If your working hour has a price, put it on the sheet, because it is usually larger than the server.

Sources: Paperclip deployment modes

Where Qoren fits, and where it does not

To be direct: Qoren does not host Paperclip. We run the layer underneath it. If what you actually want is agents that stay online without you operating a server, Qoren deploys OpenClaw, Hermes, or Codex agents into dedicated managed environments, keeps them running, isolates their secrets, monitors their health, and caps their spend. If on top of that you want an org chart, goals, and board approvals across a whole team of agents, Paperclip is a good control plane and you should host it. Those are two different purchases, and anyone telling you one product is both is selling you something.

Related guides

Frequently asked questions

The software is MIT licensed and free to self-host. Running it is not free: you pay for the server that keeps it online, a database in production, storage, the compute the agents run on, and the model tokens they spend.

Run OpenClaw or Hermes without managing infrastructure.

Deploy a managed agent environment, configure the runtime, and keep the agent online without Docker, VPS setup, or server maintenance.

Get started