Marginal cost ≈ 0
Download a model once; every inference after that uses your own electricity and GPU — the more you use it, the more you save.
Why Edge AI
Privacy isn't a setting.
It's the architecture.
We don't "promise not to look at your data" — the architecture simply gives your data no path off your computer. Moving AI back onto your own machine changes more than speed.
01 — Data Flow
Where your data goes, in one diagram
Apart from problem reports you choose to send, every outbound connection is listed below, and none of them carries your content. This isn't a policy promise — it's the code-level architecture, continuously verified by automated tests that no cloud AI inference service is ever called.
Your computerSix modules, local models and your data — all AI happens here.
Local models (GPU inference)GPU
Your dataChats · documents · photos · recordings
AI ChatSmart PhotosDocument RecognitionTranslation StudioSpeech StudioImage Generation
Apart from problem reports you send, these are the only outbound connections — none carries your content
Model & update downloadsOne-way · never contains your content
Activate your serialOnce · about a minute
Licence status checkPeriodic · only confirms this computer's licence
Optional component downloadsOnly when you click · never contains your content
Your chats, documents, photos, recordingsNever uploaded
InternetInternetTo this system, the cloud is only a download source — never a destination for your data.
02 — Comparison
Cloud AI vs. Armor EdgeAI
| Item | Typical cloud AI service | Armor EdgeAI |
|---|---|---|
| Where data goes | Uploaded to the provider's servers | Stays on your computer, saved in place |
| Cost | Subscriptions or metered billing | Runs locally — no fees, no quotas |
| Offline | Everything stops | AI keeps working |
| Peak hours | Queues, throttling, slowdowns | Your GPU serves only you |
| Confidential files | Depends on the provider's terms | Never leave the machine — no third party to trust |
| Usage statistics | Telemetry collection is the norm | Zero telemetry, not a single record |
03 — Principles
Three underlying reasons
Latency set by hardware
No network round-trips, no queues; response speed depends on your machine, not someone else's datacenter load.
Privacy as architecture
Data never leaves the machine, so you don't need to study processing terms to feel safe — there is no third party on the path.