Last Beacon is a pocket-sized AI survival assistant that works completely offline — no signal, no cloud, no power grid required. When everything fails, it's the one device that doesn't.
🔒 No charge until shipping · 30-day money-back guarantee
Ruggedized orange case · 240×280 IPS display · Built-in speaker & mic · Belt-strap compatible · RPI 5 inside
Power grid failures, natural disasters, geopolitical instability — events that were once rare are becoming routine. When they happen, the tools most people rely on become useless.
During major emergencies, towers are overwhelmed within minutes. 911 becomes unreachable. Your phone becomes a paperweight with a bright screen.
Every AI assistant — ChatGPT, Siri, Alexa — requires internet. The moment connectivity drops, they all go dark simultaneously. Your survival knowledge disappears.
Reading a 300-page field manual while lost and panicking is not a plan. You need a calm voice that answers "what do I do right now?" in plain, direct language.
Last Beacon runs a local AI language model and a searchable database of 118 survival guides entirely on the device. Ask anything by voice or text — it answers in seconds, out loud.
No apps, no setup, no account. Just press, speak, and listen.
Press and hold the physical button on the side of the device. It starts recording immediately.
Ask anything — "how do I treat a deep cut?", "what plants can I eat?", "how do I signal a rescue plane?"
The device transcribes your voice, searches 115k indexed survival passages, and feeds the best matches to the AI model.
The AI generates a concise, step-by-step response and speaks it aloud. No screen required — works in the dark.
Every design decision was made with one question: will this work when everything else has failed?
AI model, voice recognition, text-to-speech, and the entire knowledge base run locally. Zero internet dependency, ever.
Hands-free operation. If you're injured, cold, or in the dark — you can still get an answer without looking at a screen.
Pre-loaded with the US Army Survival Manual, SAS Guide, wilderness first aid references, and 100+ additional texts.
Built-in battery lasts hours of active use. Charges via USB-C. No proprietary cables, no special chargers.
Remembers what you've told it during the session — you don't have to repeat your location or situation on every question.
Powered by Whisper — one of the most accurate multilingual speech models. Speak in your language, it understands.
No black box. Every component is documented, open-source, and field-replaceable.
| Compute | Raspberry Pi 5 (8 GB LPDDR4X)ARM |
| Display & Audio HAT | PiSugar Whisplay — 240×280 IPS LCD, speaker, dual microphone array |
| Battery | PiSugar 3 Plus — 5000 mAh, USB-C charge, real-time UPS protection |
| Storage | 32 GB A2 microSD — pre-loaded with all models & knowledge |
| Enclosure | Custom compact case — belt-clip & pocket compatible |
| Button | Dedicated physical push-to-talk GPIO button |
| Connectivity | None required — fully air-gapped by design |
| Charge port | USB-C (5 V / 3 A) |
| OS | Raspberry Pi OS Bookworm (64-bit) |
| Speech recognition | faster-whisper base model — offline, 60+ languages |
| Language model | Ollama · qwen3:1.7b — runs fully on-device |
| Text-to-speech | Piper TTS · en_US-amy-medium |
| Knowledge retrieval | Qdrant vector store + nomic-embed-text embeddings (RAG) |
| Interface | Flask web UI · physical button · LCD status display |
| Source code | Python 3.12 — open source, MIT licensed |
| Updates | USB drive or local Wi-Fi — no cloud account needed |
| Speech → text | ~0.8 s (10-second utterance, Whisper base) |
| RAG retrieval | ~0.3 s (115 k passages, Qdrant ANN) |
| First token | ~1.2 s (qwen3:1.7b on Pi 5 8 GB) |
| Full response | ~3 s end-to-end (100-word answer) |
| Battery life | ~4 h active Q&A · ~18 h standby |
| Cold boot → ready | ~35 s |
| Temperature range | 0 °C to 50 °C operating |
We're accepting a limited number of early-access pre-orders. Pre-order holders get the device first, at the lowest price it will ever be, with a 30-day money-back guarantee.