AI models
AI models and providers
Choose who analyses your camera frames: OpenAI, TypeSafe AI or a local model on your own computer.
Three ways to analyse your cameras
Every check sends a short series of frames from your Station to the AI model you choose, together with your Events. The model answers whether each Event is happening, and your workspace turns the answer into Messages and alerts.
You choose the model per Station, and you can override it for a single camera. Changing it never changes your Events, schedules or alerts.
OpenAI (default)
Works on every Station, including the Android app. Your Station sends frames straight to OpenAI with your own API key, and OpenAI bills that usage to your account. It needs no extra hardware and is the quickest way to start.
TypeSafe AI (experimental)
An alternative cloud provider with its own API key and pricing. TypeSafe AI image analysis is experimental and may change or become unavailable.
Local model
The model runs on the computer with your Station, so camera frames are analysed there and never go to a cloud AI provider. There is no token bill; the cost is your hardware and electricity. It needs a desktop Station (Windows, macOS or Linux) with enough memory, ideally with an NVIDIA GPU or Apple Silicon.
Two models are supported: Qwen 3.5 9B, which we recommend, and Gemma 4 12B.
How to choose
- Start with OpenAI if you want the simplest setup or run Station on a phone or a small device.
- Choose a local model if frames must stay on your premises, or if you watch many hours and prefer hardware to a usage bill.
- Mix them: keep OpenAI as the Station default and switch only the cameras that need it to a local model, or the other way round.
- Camera Chat, where you ask a camera a question, always uses your OpenAI key, whichever model watches for Events.
Explore guides
Questions & answers
Can I use different models for different cameras?
Yes. Each Station has a default provider, and every camera can override it in its AI provider settings.
Does a local model work offline?
No. Analysis happens on your computer, but Station still needs the internet to receive your Events, deliver alerts and keep your subscription active.
Which model is the most accurate?
It depends on your cameras and Events. In our small October 2026 test, Qwen 3.5 produced fewer false alerts than Gemma 4. Try your own Events on your own cameras before relying on any model.
$AGSAUR · Solana · x402
The Agents in this guide are metered in $AGSAUR.
Every observation, question and webhook an Agent runs is settled in $AGSAUR over x402 on Solana — no cards, no subscription, no invoice. Station operators earn the same token for keeping the network awake. Launching soon.
Launching on Solana — CA drops with the token