AI products don't need a big cloud bill

AI products are often assumed to need big cloud bills and complex infrastructure.

For internal tools, that cost and complexity rarely pay off. The AI models already run in the cloud; what the server has to do is run the applications and the agents' work, reliably.

We wanted two always-on AI applications, with background agents, on modest hardware that stays private and cheap.

Lean, reliable, and private

This is how our two internal AI applications run every day. They were built for our own companies, BinaryBits and TheDataHQ. We describe what the setup does, not how it is secured or where it runs.

🔄
Always-on services
Each app and its agents run as background services that restart themselves.
🚦
One agent at a time
Agents run one after another, so the server never overloads.
💾
Backups
Nightly, kept for a limited time.
🧹
Clean-up
By-products are cleaned automatically, so disk use stays flat.
🔒
Sign-in for every page
And private access only.
📱
Alerts by Telegram
Problems reach the owner's phone at once.
⏸️
A shared AI subscription
When a usage limit is reached, the work waits and carries on afterwards instead of failing.
🧱
Separate tools
Each app's AI agents keep their own memory and never touch the other's data.

Two apps, one small server

AI agentAutomated stepPersonData or serviceMarketing appand its agentsLead Deskand its agentsOne small private serverBackupsevery nightClean-upby-products onlyAlertsto the owner's phoneruns onruns onnightlynightlyproblems
A conceptual view only. It shows what the setup does, not how it is built or secured.
1
Services run continuously
Each app and its agents run as background services that restart themselves.
2
Agents work in turn
One agent at a time, so the server never overloads. At a usage limit, work pauses and carries on after the reset.
3
Backups and clean-up run every night
Backups are kept for a limited time, and only by-products are cleaned, so disk use stays flat.
4
Alerts go to the phone
Anything that needs attention reaches the owner on Telegram at once.

One subscription, many agents

A sensible use of one AI subscription for many agents: they take turns, and pause and resume at the usage limit. The agents run on Anthropic's Claude.

WhoWhat it does
Background services (automated)Run each app and its agents continuously, and restart themselves.
Agents (AI)Run one after another, sharing one subscription, and pause and resume at the usage limit.
Backups and clean-up (automated)Nightly backups kept for a limited time; clean-up of by-products only.
The owner (person)Gets alerts on Telegram when something needs attention.
Nothing lost at limits

A run that hits the usage limit pauses and carries on after the reset.

No overload

One agent at a time keeps modest hardware comfortable.

Stable disk use

Storage rules clean only by-products, never the core data.

Separation

Each app's agents keep their own memory and never touch the other's data.

Technologies used

Standard Linux services, our own Python code, relational SQL databases, and Telegram for alerts.

Linux services Python Relational SQL databases Telegram alerts

Challenges and how we solved them

❌ Challenge
Two apps sharing one AI subscription.
✓ What we did
A run that hits the usage limit pauses and carries on after the reset.
Result: Nothing is lost.
❌ Challenge
A long morning run cut off by an outside time limit.
✓ What we did
Aligned the limits.
Result: Runs finish.
❌ Challenge
Disk use creeping up.
✓ What we did
Storage rules that clean only by-products.
Result: Stable disk use.

Always on, unattended

24/7
Both apps run continuously
Unattended
Nightly agents and the email system
1
Small private server

Both apps run continuously, and the nightly agents and the email system run unattended. Business results: not applicable for this setup.

Last updated: November 2026

What this means for you

The same discipline keeps AI tools reliable and affordable, on your hardware or your cloud.

  • Run internal AI tools without big cloud bills. Modest hardware, with the AI models running in the cloud.
  • Self-hosted or private deployments. Your tools and data stay on infrastructure you control.
  • The same discipline on your cloud. AWS, Google Cloud or Azure, with managed databases.

For client systems we build on PostgreSQL, MySQL, MongoDB or Redis, and cloud data warehouses where needed, chosen for the project's scale and team. See our backend development services.

Frequently asked questions

Is a small server enough for AI agents?

For internal tools, often yes. The AI models run in the cloud; the server runs the applications and the agents' work.

What if the AI provider limit is reached?

Work pauses and carries on after the reset; nothing fails.

Is it secure?

Access is private and every page needs sign-in. We don't publish further details.