Reading a reply: Served by, cost and notices
Under every reply, Knecty tells you what happened.
“Served by”
The label names the model that actually wrote the reply:
- “Served by: {model}” when you picked the model yourself.
- “Auto → {model} · {reason}” when Auto picked it.
If something else answered instead of the model you asked for, a grey note under the reply says so. See The “Served by” guarantee.
What the reply cost
While a reply is being written you see a running “cost so far ≈ $…”. When it finishes, it becomes the reply’s exact cost. Tap it for the breakdown:
- Tokens in — what the model read: your message plus the conversation so far (and any attached files). Some models show part of this as Cached in, which is cheaper.
- Tokens out — what the model wrote. For thinking models this includes its thinking.
- Search embedding, Document summary or Title — only when the reply used them.
- Vendor cost, Platform fee (25%), and the Total.
Longer chats cost more per message, because every message re-sends the conversation so far. The estimate under the message box (“~$a–$b est. next message”) shows this before you send. See Long chats and compaction for how to bring it down.
Notices under a reply
A grey line under a reply — and often a banner above the message box — explains any change. The common ones:
| Notice | What it means |
|---|---|
| “You’ve used 80% of this week’s premium — switched to {model}…” | Auto-switch moved you to a lower-priced model. Turn it off in Settings → Models. |
| “You’re out of premium for now — continuing on {model}.” | This week’s premium credit is used up. Top up to use premium models again. |
| “Your trial’s premium allowance is used up — continuing on {model}.” | Start your plan to get your weekly credit. |
| “{model} is unavailable right now; serving {model}.” | The provider was down, so a working model answered. |
| “In your region we route to {model}…” | DeepSeek isn’t used in the EU and UK. |
| “Web search failed — answered without sources.” | The search didn’t run; its fee was refunded. |
| “{model} answered as {model} (the provider served a different model)…” | The provider swapped models — the reply says so. |
Banners with a weekly reason also show when your credit resets.
Stopping a reply
Press Stop while a reply is being written. You keep what was written so far and pay for that part only. The reply is marked Stopped.