Supported providers
Promptly talks to every provider over the same OpenAI-compatible chat-completions API, which is why the list below mixes vendors freely — and why a key from any one of them is enough to run the whole assistant.
The five providers
| Provider (as shown in the picker) | Default endpoint | Where the key comes from |
|---|---|---|
| OpenAI | https://api.openai.com/v1 | The API keys page in the OpenAI platform console |
| Anthropic | https://api.anthropic.com/v1/ | The API keys section of the Anthropic Console |
| Google Gemini | https://generativelanguage.googleapis.com/v1beta/openai/ | An API key from Google AI Studio |
| Zhipu GLM | https://api.z.ai/api/paas/v4 | The API keys page in the Z.ai / Zhipu console |
| Moonshot Kimi | https://api.moonshot.ai/v1 | The API keys page in the Moonshot platform console |
You never type the endpoint — it's applied for you. Override it only for a proxy or gateway, as described in Base URL override.
🖼️ [Image] — The provider dropdown with one entry marked "(already added)".
One key per provider
A workspace holds at most one credential per provider, so up to five in total. Trying to add a second key for the same provider is refused — the picker greys it out as (already added) and the server rejects it as a duplicate.
If you need to rotate a key, don't add a second one: open the existing credential and paste the new key in, which resets it to Unverified for re-verification.
Multiple providers are how you build redundancy, not extra capacity. See Backup keys & failover.
Your provider account needs credits
An API key proves who you are; it doesn't guarantee you can spend. Before adding a key, make sure the provider account has a payment method or a positive credit balance:
- A key with no credit authenticates but fails on the first real message. Promptly marks it Out of credits.
- A revoked or mistyped key fails immediately as Invalid key.
- Free tiers vary in whether they allow tool calling, which the assistant requires for chat. If tool calling verification fails, the key is only usable for background jobs — see Model roles.
Statuses and what to do about each one are in Key statuses.
Picking a provider
Any of the five will run the assistant. What differs in practice:
- Latency and tool-calling reliability matter more than raw benchmark quality here, because every chat turn plans tool calls before answering. See Multi-step reasoning.
- Cost per answer depends on the model you select inside that provider, not the provider itself — a small model at OpenAI can cost less than a large one at a cheaper vendor.
- Data handling is between you and your provider; the request goes from Promptly's servers to the endpoint above using your key.