AI providers

Sidebar → Setup → AI providers

Your keys, your bill. Nothing generates until you add one.

How this works

Every generation your customers run uses your API key and is charged to your account with that provider. The platform never generates on its own key and never falls back to one.

That means you keep the margin, you choose the models, and you control the spending. It also means an expired or over-limit key stops your business, so it is worth setting up properly once.

Adding a key

1. Setup → AI providers → Add a key
2. Choose the provider
3. Paste the key
4. Save, then press Test

A green result means the provider accepted it. A red one shows the provider's own error message, which is usually specific enough to fix immediately — wrong key, no billing set up, no access to that model.

Which provider does what

If you are starting out, add OpenAI. One key covers every module, and you can add others later once you know what your customers actually use.

OpenAI-compatible endpoints

Choosing this asks for a base URL as well as a key. Use it for a provider not listed above, a self-hosted model, or a gateway of your own — anything that accepts OpenAI-shaped requests.

Choosing models

Set a provider and model per tool, in Tools and prompts. Leave them blank and the platform's recommendation is used.

Mixing providers is normal and often the right answer. A sensible setup:

• Long-form articles → Anthropic
• Quick summaries and chat → Groq, for speed
• Images → OpenAI
• Voice → ElevenLabs

Your customers see none of this. They see a tool that works.

Controlling what you spend

The credit system is your first line of defence: a customer cannot exceed their balance, and charge happens before the provider is ever called.

Add two more:

Set a spending cap with your provider. Every one of them offers a monthly limit. Set it. It is the only thing that protects you from a bug, a bad prompt or someone gaming a tool.

Price your credits above your cost. Look at what a run actually costs you at your provider, then set the credit cost for that tool with margin in it. Review it against Your customers → Content activity after a month.

Image quality

Images are the one place where a single run can cost real money, and the price climbs steeply with quality. So every image is currently generated at the lowest quality tier, whichever provider makes it:

This is not something you can change yet. Choosing quality per tool — the same way you already choose provider and model — is planned, and this tier is what it will default to. Until then, nobody discovers what images cost from a surprise bill in their first week.

The exception is the older dall-e-2 and dall-e-3 models: they do not accept these tiers at all, so if you set one as a tool's model, OpenAI's own default quality applies.

When a key stops working

You are emailed when a provider starts rejecting your key. Tools that depend on it will fail for your customers until it is fixed — and every failed run automatically refunds the credits, so nobody is charged for something they did not get.

Common causes, in order of likelihood:

1. Billing lapsed at the provider
2. The key was rotated or revoked
3. A spending cap was hit
4. The model was retired or you lost access to it

Fix the key, press Test, and everything resumes. Nothing needs to be switched back on.

Keeping keys safe

• Keys are encrypted before they are stored, and are never shown back to you in full after saving.
• No other organization on the platform can see or use your keys — ever, including by accident. The system loads your keys for your requests and blanks every provider you have no key for.
• Use a key created for this platform alone, not one you also use elsewhere. Then rotating it costs you nothing.
• Rotate on a schedule you can live with, and press Test after every rotation.

Next: Plans and pricing