Connecting a provider.
Paste your key and we test it against the provider before we store it — a bad key never gets saved. One connection per provider, and reconnecting replaces the key.
Connecting a provider is one screen and one paste. Open AI Models in the sidebar, click Connect on the provider you want, and put your API key in. The important detail: we test the key against the provider before we store anything, so a bad key never gets saved.
Where to get a key
You create the key on your provider's own site — we can't do that part for you. Each provider has a full walkthrough, including the billing step that catches most people out:
- OpenRouter — the easiest start. One key covers text and images, pre-paid, no surprise bills.
- OpenAI — needs a card on file before the key will work.
- Anthropic (Claude) — text only. It cannot generate images.
- Google (Gemini) — has a free tier you can start on.
The Connect modal links straight to both — the provider's key page, and its guide here — so you can open them side by side.
The Connect modal
Paste your API key
Into the key field. It's masked as you type; the eye icon reveals it if you want to check for a stray space, which is the single most common reason a paste fails.
Pick a default text model
The dropdown is searchable and curated — a ★ marks our pick for each provider, and each model carries a one-line note on what it's for. You can also type or paste any model ID your provider supports; an inline “Use…” row appears and saves it as a custom model.
Pick a default image model
Same control, image models only. On Anthropic this field is replaced by a note explaining there isn't one — Claude is text only.
Test & save
The button reads Testing… while we call your provider, then ✓ Verified. The modal closes itself.
Both model fields open empty and nothing is pre-selected. That's deliberate — but if you save without picking a model, that capability has no model to run on, so it silently falls back to ScoutRival's included AI even after you route to the connection. Pick a model. It takes two seconds and it's the difference between your key being used and quietly ignored.
We test the key before we store it
Hitting Test & save makes a real, read-only call to your provider — we ask it to list the models your key can see. Nothing is written to our database until that call comes back clean.
- If it works — the key is encrypted, stored, and the card turns green with a masked version of the key.
- If it fails — nothing is saved, and you're shown the provider's own error message, verbatim, with its HTTP status code. We don't translate it, because the provider's wording is usually the actual answer.
In practice a rejected key means one of three things: the key is wrong or was rotated, there's a stray space in the paste, or you haven't set up billing on the provider's side yet. The per-provider guides list the exact error strings each one returns. The check gives up after 15 seconds.
One connection per provider
You get exactly one connection per provider, not a list. Connecting OpenAI when you already have an OpenAI connection replaces it — the old key is deleted and the new one takes its place.
That's how you rotate a key: click Manage on the card, paste the new key, save. Your routing rows keep pointing at the same connection, so nothing else needs touching.
If a key has expired, reconnect over the top of it rather than disconnecting first. Disconnecting clears your routing rows; reconnecting leaves them exactly as they were.
What we store, and what you see
- The key is encrypted before it touches the database, and it's kept apart from the connection record itself. It's decrypted only server-side, at the moment a generation needs it.
- You never see it again. After saving, the card shows a masked version — the first few characters, then the last four. There is no “reveal key” button, for you or for us.
- Your models and base URL are not secret and are shown plainly on the card as chips.
If you lose the key, you don't recover it from ScoutRival — you create a new one at your provider and reconnect.
Disconnecting
The bin icon on a provider card removes the connection. It asks first, then:
- The encrypted key is deleted.
- Every routing row pointing at it is cleared back to blank — you don't have to un-route it yourself.
- Text and Blog generation carry on using ScoutRival's included AI. Nothing breaks and nothing is lost.
- Image rendering stops. There is no included image model, so with nothing connected you'll get image prompts and no pictures. Why →
Anything you already generated — posts, articles, rendered images — stays exactly where it is.
“Other providers” and the Legacy badge
If you connected xAI, DeepSeek or a custom endpoint in the past, it still appears in a separate Other providers row with a Legacy badge, and it still works. Those aren't offered for new connections any more, so once you remove one you can't add it back. The four providers above are the supported set.