AI Providers
Connect a Provider
Run your chats and agents on an AI account you already pay for, for 0 Nirvai credits — connect it, let Nirvai verify it, and choose which models to enable.
Using your own AI provider is like bringing your own coffee beans to a café that brews them for you. Nirvai still does all the work — the chat, the agents, the tools — but the "beans" (the AI model answering) come from an account you already pay for. Every reply made with your own model costs 0 Nirvai credits: your provider bills you directly, usually at a much lower price.
If you don't connect anything, nothing changes — everything keeps running on Nirvai's built-in models, billed in credits as usual.
Nirvai verifies the keyConnection verifiedgpt-5.2| You can… | What that means for you |
|---|---|
| Connect a provider | Add an account from OpenAI, Anthropic, Google Gemini, Azure, Fireworks, DeepInfra, OpenRouter — or any compatible service |
| Use it in any chat | Pick your model from the model menu in the chat box — that conversation stops spending credits |
| Make it your default | Turn on "Use it for new chats" and every new conversation starts on your model automatically |
| Put an agent on it | Set your model as an agent's brain — it answers on your account everywhere it's deployed, channels included |
| See what you saved | The Analytics tab shows your usage and what those replies would have cost in credits |
The Credentials section also lists AI services — those are credentials your agent uses inside a tool (for example, a tool that calls Gemini to label images). AI Providers is about which model powers the conversation itself.
Before you start
- An account with an AI provider (OpenAI, Anthropic, Google Gemini, Azure, Fireworks, DeepInfra, OpenRouter, or a compatible service).
- That account's API key — a private password that lets Nirvai connect to the provider for you. You create it in the provider's own dashboard, usually under "API keys".
Step-by-step
- Open AI Providers from the main menu.
- Click Connect a provider and pick your provider from the list.
- Paste your key. For a Custom provider, also paste the service's address (the provider tells you this — it's usually on the same page as the key).
- Pick the models you want available. Nirvai suggests recommended models with speed and quality stats; you can also add any model your provider offers.
- Save. Nirvai immediately sends a tiny test message ("hi") to check the connection, then shows the result on the provider's card.
My OpenAIsk-p…4kF2ReadyKeys are stored encrypted and are never shown again after you save them — the card only displays the first and last few characters so you can recognize which key it is.
Connection statuses
| Status | What it means | What to do |
|---|---|---|
| Ready | The test message worked | Nothing — start using it |
| Not tested yet | Saved but not verified | Click Test connection |
| Needs attention / Not working | The provider rejected the key (revoked, out of funds, or mistyped) | Open the provider, fix or replace the key |
| Spend not tracked | Works, but the provider doesn't report usage numbers | Fine to use — the Analytics tab just shows less detail |
While a provider is not working, conversations set to its models fall back to Nirvai's models so nothing breaks — see the technical reference for exactly how.
Troubleshooting
| Problem | Likely cause | Fix |
|---|---|---|
| "Not working" right after connecting | Key copied with a space, or from the wrong provider page | Re-copy the key and use Test connection |
| The model I want isn't listed | It's not in the recommended list | Add it manually by its exact model name from your provider's docs |
| Connected fine but a chat still bills credits | The chat is on a Nirvai model | Pick your model in that chat's model menu — see Use your own models |
What's next
Your provider is connected — now use it from the chat, set it as your default, or put an agent on it.