Choosing a model
Promptly doesn't ship a fixed menu of models. When you paste a key, it asks your provider which models that key can use and builds the picker from the answer — so a model released yesterday is selectable today, and a model you don't have access to never appears.
Reading the model list
Under the Chat model select, one of four hints tells you where the list came from:
| Hint | Meaning |
|---|---|
| Used for live chat replies… | No fetch attempted yet — enter your key first. |
| Checking your key for available models… | The request to your provider is in flight. |
| Models loaded from your key ✓ | The list is live, straight from your provider. |
| Showing defaults — couldn't fetch models (…) | The fetch failed; you're seeing Promptly's built-in defaults for that provider. Shown in amber, with the provider's error in brackets. |
Falling back to defaults is not fatal — those models generally work — but it usually means the key, the base URL, or the provider itself has a problem. Fix that first rather than saving on top of a failed fetch. See Troubleshooting keys.
🖼️ [Image] — The Chat model select showing "Models loaded from your key ✓" with the "Fetch models" link beside it.
Fetching and refreshing
- In the Add modal, the list is fetched automatically when you leave the API key field. A Fetch models link appears next to the hint if you want to retry.
- On an existing credential, Edit models fetches the list on open and shows a header row: Source: Live · fetched 5 minutes ago, or Source: Fallback (defaults) with the error. Refresh forces a new call to the provider.
Results are cached for about an hour, and Refresh is rate-limited to roughly one call every 10 seconds per credential — spamming it returns the last known list rather than hammering your provider.
What gets filtered out
Providers return everything from one listing endpoint — embeddings, image and video generation, speech, moderation, rerankers. Promptly removes those, because none of them can hold a conversation, and keeps only chat-capable model IDs. If filtering leaves nothing at all, that's treated as a failed fetch and you get the defaults instead.
When a model disappears
Providers retire models. If a model you have saved is no longer in the live list:
- The credential card shows a Model unlisted badge under the affected role.
- In Edit models, the value stays selected but is labelled (no longer listed), with No longer listed by provider underneath it.
Nothing breaks the instant a model is delisted, but it will stop working eventually. Open Edit models, pick a current model, and Test & save.
The card's badge only appears after the live list has been fetched at least once — opening Edit models is what surfaces it.