Your own server: why MONO isn't SaaS

The multi-tenant pattern: cheap, but dangerous

Every modern SaaS is built the same way: a single app instance, a single database, and a user_id column that separates millions of customers. Every query says "give me data where user_id = X".

This is economically optimal for the vendor: one server serves 50,000 users, infra costs are marginal, and you scale vertically. It's why Notion, Slack and ChatGPT charge $20/mo and stay profitable.

But security is shared by contract, not physics. A bug in the user_id filter — documented examples every year — exposes everyone. An admin-panel employee can see your data as easily as any other customer's. A court order against the vendor covers their entire system.

What "one server per person" means in MONO

When you sign up for MONO, this happens in ~90 seconds:

1. A VPS is provisioned on Hetzner (Finland) exclusively for you.
2. The agent binary (Go), SurrealDB, and a Cloudflare tunnel are installed at {your-handle}.mono.vip.
3. Encryption keys are generated locally, sealed with a passphrase only you know.
4. WhatsApp is connected using your dedicated number.

From that point on: your messages, memories, passwords, documents and calendar never leave your VPS. There's no central database. No user_id column. Just your server, and only your server.

The control plane (Supabase) exists, but only stores the minimum needed for billing and message routing: your email, your plan, your VPS IP. No sensitive data touches the control plane.

What this gives you (and costs you)

Benefits:

• Blast radius of 1. If a VPS is compromised, one person is exposed — not the whole base.
• Real portability. We can give you a full dump of your VPS. It's your data.
• Scoped legal process. A subpoena against MONO only touches you.
• Auditability. One user per machine means every access correlates with your session.
• Consistent performance. No noisy neighbor. The VPS resources are yours.

Costs:

• More expensive infra. A cpx11 VPS costs ~€5/mo. That's why MONO is $20, not $5.
• More ops complexity. Deploying to 10,000 servers is harder than deploying to 1. We solve it with auto-update + signed binaries.
• No cross-user magic. No "global model learning" from your data. The model learns from you, in your VPS, locally.

What about the AI? Isn't it in the cloud?

Yes — for now. MONO calls Anthropic, OpenAI and other models via API. Those calls go over TLS, with contractually-agreed no-retention policy. But your MONO's state (history, memories, files) never lives on those providers. Only the minimum context for the current request is sent.

If you want zero cloud, you can bring your own API key (BYOK) and route to a local model — we're working on self-hosted Llama 3.3 support on the same VPS. By mid-2026 you'll be able to run MONO 100% offline if that's what matters to you.

Why doesn't everyone do this?

Because it sells worse. "Private server" sounds enterprise, not consumer. "$20/mo" competes with services that charge $10 and put everything in Postgres. The advantage only shows up when there's a breach — and by then, it's too late.

But as AI assistants accumulate more context on you — your spending, your passwords, your conversations with your partner, your health — the cost of a breach grows too. In 5 years, a multi-tenant personal assistant will look as anachronistic as a database without backups.