How Auto picks the right model

When you chat with VIGPT, you don't pick the AI model — Auto does. And the models aren't equal: some are cheap and fast but a bit dim, others cost more but are genuinely smart. Auto's whole job is to read your message and hand it to the right one.

Get that choice right and the app just feels smart. Here's how Auto makes it.

The front desk analogy

Think of Auto as the front desk at a clinic. When you walk in, someone sizes up what you came in with and decides — fast — whether you see a nurse, a doctor, or a specialist. Auto does the same with every message: it reads the request and routes it to the right level of care. A good front desk makes a handful of sensible moves, and so does Auto.

How Auto decides

Here are the moves any good front desk would make:

• It reads the evidence. Did you attach files? Is this a long, multi-part request? Does it involve something technical, like code? More signals, better call.
• There's one dial — quality vs. cost — and we've set ours toward quality. Care comes first; cost is the thing we watch, not the thing we chase.
• When it's unsure, it sends you up, not down. Uncertainty should buy you the doctor, not the intern.
• It can take a second look. When a first answer comes back clearly weak — a non-answer, an "I can't help with that" — Auto quietly bumps you up to a senior, once.
• For genuinely hard cases, it thinks before it answers instead of blurting out the first thing.
• And it gets tuned with real use. We watch how answers land — including when you thumbs-down a reply — and use that to recalibrate Auto over time.

The idea in one line

Auto reads the evidence, leans toward quality when it's unsure, thinks harder on the hard ones, double-checks weak answers, and keeps getting tuned by what people actually do.

Everything else is just plumbing in service of those moves.

Smart by default isn't expensive by default

"Smart by default" doesn't mean "expensive always." Trivial messages — "hi," "what time is it" — still go to the cheap model. We only spend up when the request warrants it, and that second look only kicks in when a first answer comes back weak.

So it's quality-first and cost-aware — not burn-money.