Old hardware from bankrupt European businesses becomes offline AI infrastructure for schools, clinics, and communities that have never had access to AI. Free. Offline. Theirs.
See how it works
When Bobby Burger (46 restaurants) and Disco:VR (VR entertainment) went bankrupt, the founder was left with dozens of POS terminals, UberEats delivery tablets, and gaming laptops destined for e-waste.
Instead of scrapping them, he installed open-source AI software on every single one. They became a distributed inference fleet — small models running on cheap hardware, collaborating to produce answers that rival GPT-4.
Now we're sending them where they're needed most.
Small open-weight model (3-8B params) + semantic knowledge base. Answers questions without internet. Cache-Enhanced Augmented Generation gives +15pp accuracy over raw model — matching frontier quality on known domains.
Whisper speech recognition in 99 languages. Speak your question, hear the answer. No keyboard, no reading, no English required. A farmer asks about crop disease in Hausa — gets a spoken answer with a picture.
Pre-loaded domain expertise: WHO healthcare guidelines, K-8 education curricula, FAO agriculture data. Communities build their own packs in local languages. The device knows what matters here.
Devices sync knowledge over LoRa radio — 5-15km range, no cellular needed. Health alerts cross national borders. Local crop prices propagate through the mesh. An alternative internet, owned by no one.
Curated database of dangerous misinformation — medical myths, harmful advice. Every response is checked before delivery. Confidence scoring warns when the AI is guessing. Community feedback trains the filter.
When internet is available, idle compute serves API requests. A school's AI workstation teaches students by day, earns money as an inference node by night. The hardware pays for itself.
"Companies in San Francisco deploy GPT-5 for $200/month while a billion people have no internet at all. Every month this gap widens, AI becomes a tool of inequality rather than progress."
The AI Gap — 2026
Decommissioned hardware from bankrupt businesses, corporate IT refreshes, European recyclers. POS terminals, office PCs, tablets, laptops — anything with 4GB+ RAM.
One-command install: USB stick boots, auto-detects hardware, selects the right model, loads knowledge packs, configures voice interface. Five minutes, zero expertise required.
Packaged AI workstations shipped to partner schools, clinics, and NGOs. Each device arrives ready — plug in power, start asking questions. Works from minute one, online or off.
Every query builds the local cache. Communities create their own knowledge packs. Multiple devices form mesh networks. The system gets smarter, more local, more useful — without anyone's permission.
Multiple devices in one location automatically form a local ensemble — collaborating to produce better answers than any single device alone.