Private AI on the machines you already own.
Borg turns your idle laptops, desktops and servers into your own AI data center. Your data never leaves your network, and there's no per-token bill.
Your idle machines will be assimilated. They won't mind.
AI is expensive, and it wants your data.
You pay per token.
Every new workflow grows a bill that's hard to forecast.
Your data leaves.
Every request goes to an outside provider. Your most sensitive work often can't go at all, so it stays manual.
Your hardware sits idle.
The machines you already paid for do nothing most of the day.
Sam Altman says the memory in your PC is "being used 5% of the time." We agree. We just disagree about who should get the other 95%.
How it works
base_url = "http://borg.internal/v1"- 01
Install a small background app on the machines you choose. It steps back when someone needs their machine.
- 02
Borg sits inside your network and pools their spare capacity, including the GPU built into every modern laptop, and runs open models on it.
- 03
Existing workflows just point to a new internal address.
It also runs on your idle servers and cloud machines.
Why Borg
Private
- Everything runs inside your network.
- No AI provider sees your data, keeps it or trains on it.
- You choose the models, who can use them and what gets logged.
Cost-efficient
- It runs on hardware you already paid for.
- No per-token bill.
- No new hardware and no data-center project. Capacity grows as you add machines.
What are your idle machines worth?
11.4 billion
tokens of AI work a month, from idle hours
about 3 tokens read for every 1 written
That's $34,000 to $275,000 of AI work a year.
| Option | Cost per year |
|---|---|
| Open-model APIs | |
| Fireworks AI | $34,344 |
| Frontier APIs | |
| Google Gemini 3.1 Flash-Lite | $38,616 |
| OpenAI GPT-6.1 Terra | $137,328 |
| Anthropic Claude Sonnet 5.5 | $274,680 |
On every one of these, your data leaves your network.
Estimates. Based on a typical mix of work laptops, our lab measurements and public list prices as of November 2026, at batch rates where the provider offers them.