Paperclip AI Hosting & Self-Hosting Guide: VPS Requirements, Tailscale, Updates & Uninstall

Paperclip AI is self-hosted software, and how you host it decides whether your agent company survives a reboot, a bad update or a stranger finding your dashboard. This guide covers where to run it, how big a server you need, how to expose the dashboard without exposing it to the internet, and how to update, back up and uninstall cleanly. It assumes you have read the Paperclip AI guide and have a working install from our installation walkthrough. Details reflect September 2026 (release v2026.916.1).

Local, VPS or managed host?

There are three realistic places to run Paperclip, and most teams move through them in order.

Option	Best for	Trade-offs
Your laptop or desktop	Learning the product, single-person experiments, coding agents that need your local repos	Heartbeats stop when the machine sleeps. Agents that should run at 3 a.m. will not. No remote access without extra setup.
A VPS you manage	Anything that needs to run unattended: routines, monitoring agents, content pipelines	You own updates, backups and security. Lowest monthly cost for what you get.
A managed host or one-click template	Teams that want a server without doing server work	Convenient, but you are trusting the host's image and update cadence. Hostinger, Contabo, xCloud and ServerAvatar all sell Paperclip templates now. A first-party hosted cloud is on the Paperclip roadmap but was not generally available as of mid-2026.

Our recommendation for a business: prototype locally for a week, then move to a VPS before you let agents touch anything that matters. The whole point of the platform is agents that wake up on a schedule, and a schedule that depends on your laptop lid is not a schedule.

Server sizing

Paperclip itself is light. The Node server, the React dashboard and a Postgres instance idle at well under a gigabyte of RAM. What eats resources is the agents: each active adapter may spawn a process, clone a git worktree, hold files in memory and write logs.

Agent team	Reasonable VPS	Notes
1 to 5 agents, mostly API or HTTP adapters	2 vCPU, 4 GB RAM, 40 GB SSD	The practical minimum. Do not go lower for anything you rely on.
5 to 15 agents, some coding agents with worktrees	4 vCPU, 8 GB RAM, 80 GB SSD	Worktrees for busy repos add up fast on disk.
15 or more agents, or several companies on one instance	8 vCPU, 16 GB RAM, 160 GB SSD	Consider a separate managed Postgres at this size.

Two things scale worse than CPU: disk, because of worktrees and logs, and outbound network limits, because every heartbeat is at least one model API call. Watch disk first.

Why Postgres persistence matters

Everything that makes Paperclip useful lives in the database: the org chart, every ticket, every comment, budget counters, the immutable audit log and the tool-call traces. The embedded Postgres that a local install creates is fine for a laptop. On a server, run a real Postgres service with its data on a persistent volume, and make sure that volume is included in backups. If you use Docker, name the volume explicitly and never rely on an anonymous one that docker compose down -v can wipe.

For larger deployments, a managed Postgres from your cloud provider removes one more thing to maintain and gives you point-in-time recovery for free.

Exposing the dashboard safely

This is the section that matters most. The Paperclip dashboard can approve hires, change budgets, read secrets and direct agents that have real credentials. Never expose port 3100 directly to the public internet. A self-hosted instance with no authentication in front of it is an open control panel for whatever your agents can reach.

Option A: Tailscale (recommended for most teams)

Install Tailscale on the server and on each machine that needs the dashboard. Bind Paperclip to the server's Tailscale IP or to localhost and reach it over the tailnet. Nothing is listening on a public interface, there is no certificate to manage, and access is controlled by who is in your tailnet. Tailscale's own serve and funnel features can add HTTPS if you want a clean URL.

Option B: Reverse proxy with authentication

If you need public access, put Caddy or nginx in front of localhost:3100, terminate TLS there, and require authentication at the proxy. Basic auth is the floor. An OAuth proxy such as oauth2-proxy tied to your Google Workspace or Microsoft tenant is the standard. Pair it with a firewall rule that blocks 3100 from everything except the proxy.

# Caddyfile - minimal example
paperclip.example.com {
    basicauth {
        admin $2a$14$REPLACE_WITH_BCRYPT_HASH
    }
    reverse_proxy 127.0.0.1:3100
}

Option C: SSH tunnel

For a single administrator, ssh -L 3100:localhost:3100 user@server works and costs nothing. It does not scale to a team and it is easy to forget it is the only thing standing between the dashboard and the world.

Secrets and API keys

Paperclip has an encrypted secrets store for provider keys and any credentials agents need. Use it. Do not paste keys into agent job descriptions or ticket bodies, because those are readable by every agent with access to the ticket and they end up in the audit log. Scope keys per agent where a provider supports it, and set per-agent budgets so a runaway agent cannot drain a shared key. If you rotate a key, rotate it in the secrets page and restart affected agents rather than editing environment files by hand.

Backups

A working backup has three parts:

1. The database. A nightly pg_dump to object storage, or your managed provider's automated snapshots. This is the one you cannot skip.
2. The data directory. Uploaded files, workspace metadata and local storage. Its location depends on your install path. Include it in the same job.
3. Configuration. Your compose file, proxy config and the environment file, minus the secrets, in version control.

Test a restore once before you trust it. Spin up a second server, restore the dump and data directory, start Paperclip and confirm the org chart and tickets are there.

Updating Paperclip

Paperclip ships releases frequently, and versions are date-stamped (v2026.916.1 landed on September 21, 2026). Frequent releases mean frequent migrations, so treat updates as a small maintenance window rather than a background task.

1. Read the release notes on the GitHub releases page. Look for migration warnings and adapter changes.
2. Take a database backup.
3. Pause agents from the dashboard so no ticket is mid-checkout during the restart.
4. Update: re-run the install script for script installs, run the CLI update command for managed installs, or pull the new image tag and docker compose up -d for Docker.
5. Start the server, check the activity log for migration output, then resume agents.

If an update breaks something, the fix is to restore the backup and pin the previous release, which is exactly why step two exists.

Uninstalling

Uninstalling is mostly about not leaving keys and data behind.

• Pause and terminate agents in the dashboard so nothing is scheduled to wake up.
• Stop the service or run docker compose down.
• Remove the install directory and the data directory. For Docker, delete the named volumes deliberately with docker volume rm after taking a final backup if you want one.
• Revoke any provider API keys you created specifically for Paperclip.
• Remove the reverse proxy or Tailscale config and close firewall rules.

What hosting costs

The software is free under the MIT license. Hosting is the smallest line item in a Paperclip budget:

Setup	Typical monthly hosting
Laptop	$0
2 vCPU / 4 GB VPS	$6 to $15
4 vCPU / 8 GB VPS	$20 to $40
8 vCPU / 16 GB VPS plus managed Postgres	$70 to $150

Model usage will dwarf these numbers for any active team. Our Paperclip AI pricing breakdown works through the full monthly picture, and our integrations guide covers the Slack and Telegram alerts you will want once the instance is running unattended.

Frequently asked questions

What are the minimum server requirements for Paperclip AI?

For anything you rely on, 2 vCPU, 4 GB RAM and 40 GB of SSD, running Node.js 24.11 or newer with PostgreSQL. Coding agents that create git worktrees need more disk.

Is it safe to expose the Paperclip dashboard to the internet?

Not directly. Reach it over Tailscale or put a reverse proxy with authentication in front of it. The dashboard controls agents that hold real credentials, so treat it like an admin panel.

Can I run Paperclip on Hostinger or another shared host?

Yes, on a VPS. Several hosts sell one-click Paperclip templates. Shared web hosting without root access will not work because you need a persistent Node process and Postgres.

How do I update Paperclip AI without losing data?

Back up the database, pause agents, apply the update, then resume. Releases are frequent and sometimes include migrations, so read the notes first.

How do I completely uninstall Paperclip?

Terminate agents, stop the service, remove the install and data directories or Docker volumes, revoke the API keys you created for it, and remove any proxy or Tailscale configuration.