Adding a key
Open Dashboard → Settings → AI Keys. If you have no keys yet you'll see a Connect your own AI key panel with an Add your first key button; once you have one, use Add key in the top-right.
Steps
- Provider — pick OpenAI, Anthropic, Google Gemini, Zhipu GLM, or Moonshot Kimi. A provider you already have a key for is listed as (already added) and can't be selected twice. See Supported providers.
- API key — paste the key from your provider's console. Use the eye icon to check what you pasted. Keys shorter than 8 characters are rejected as a typo.
- Chat model — as soon as the key field loses focus, Promptly asks your provider which models the key can actually use and fills this list from the answer. The hint underneath reads "Models loaded from your key ✓" when that worked. See Choosing a model.
- (Optional) Advanced — set a separate Planner model and Background model, or a Base URL override. Leave these alone unless you have a reason; they default sensibly. See Model roles and Base URL override.
- Click Add & verify.
🖼️ [Image] — The "Add LLM credential" modal with the provider picker, masked key field, and the "Models loaded from your key ✓" hint under the Chat model select.
Verification runs automatically
You don't have to click anything else. The moment the credential is created, Promptly runs a real request against your key and updates the card in place:
- a cheap completion call → Connection ✓
- a forced tool call → Tool calling ✓
The card starts at Unverified and flips to Verified (usually within a few seconds), or to a specific failure status such as Invalid key or Out of credits. The toast "…key added — verifying…" is the create step, not the result — watch the card for the outcome. Full detail in What verification checks.
After the first key
Your first key becomes the Primary. Adding a second provider gives you a Backup #1 that takes over automatically if the primary fails — see Backup keys & failover.
Changing a key later
On the credential card:
- Verify key — re-run the check (after topping up credits, for example).
- Edit models — repick Chat/Planner/Background against a freshly fetched model list.
- Disable / Enable — take a key out of rotation without deleting it. Re-enabling resets it to Unverified, so verify again.
- Delete — removes the key and wipes the stored secret. Chat falls back to the next available credential, or stops working if none remain.
Pasting a new key into an existing credential also resets it to Unverified — the new secret hasn't been checked yet.