AI Providers Explained

An AI provider is your organization's account with an AI service: where it is, and the key that pays for it. The AI Provider Registry (Settings → Artificial Intelligence → AI Provider Registry) keeps every provider you connect, every model each one offers, and exactly what each model can do. Everything in Kamo that uses AI, from the AI Assistant to your AI team members, finds its models here.

Providers, models and capabilities

• A provider is the account: its type, its address and its key. Keys are stored encrypted, and once you save one Kamo never shows it in full again.
• A model is one of the provider's AI models. Kamo lists them for you by asking the provider, and you can add one by hand.
• A capability is something a model can do: generate text, read images, call tools, return structured output, create embeddings, transcribe speech, speak, translate, operate a computer and more. Each capability can carry details, such as the languages a translation model supports or the size of an embedding.

Where the facts come from

Each capability shows its source. Reported by the provider means the service said so itself; Kamo's model catalogue means Kamo knows that model; Guessed from the name means Kamo guessed from the model's name; Found by a check means a verification measured it; Set by an administrator means someone in your organization set it; Set by Kamo marks Kamo's own services. A guess is never trusted blindly: it stays unverified until it is checked.

Verification

Verify runs a small real request against the model: a short message, a tiny image, a tool call, one embedding, a click on a test screen. The capability becomes Verified when it works, and Check failed, with the reason, when it does not.

AI team members only use verified capabilities, so an unproven model is never handed work it cannot do. People using the AI Assistant can use anything that is switched on. See AI Team Members Explained.

The registry's Coverage view lists every capability Kamo knows and whether your organization can use it: Verified, Available, not verified, or Not available. It is the quickest way to see what your AI members can and cannot do yet.

Prices

Each model can carry its Pricing: what it costs per 1,000 input or output tokens, per second of audio, per image, per page and so on. Kamo fills these in for models in its catalogue, and you can set your own. They are used only to report spend and to apply AI members' budgets; Kamo charges nothing for model usage. Your provider bills you directly.

Provided by Kamo

Kamo runs some AI services of its own inside its cluster, such as embeddings, reranking, translation, text recognition (OCR), speech and a local language model. Your data never leaves Kamo, so they are safe for PHI, and they need no key of your own. They are switched on for every organization, and you can switch any of them off under Provided by Kamo.

Which model is used

When a request does not name a model, Kamo picks one that can do the work, preferring the models you list for that kind of work, then your own providers before Kamo's, then healthy providers and your order in this list. That is what makes more than one provider resilient rather than just redundant. You set the preferred models, and AI members' budgets, under Routing & Budgets; see Routing & Budgets Explained.

Health

Kamo checks every provider regularly. A provider whose key stops working is marked Key rejected and is skipped until the key is fixed; nothing is switched off behind your back.