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.
Two apps, one small server
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.
| Who | What 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. |
A run that hits the usage limit pauses and carries on after the reset.
One agent at a time keeps modest hardware comfortable.
Storage rules clean only by-products, never the core data.
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.
Challenges and how we solved them
Always on, unattended
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.