Loading the catalogue…
Loading the catalogue…
Most teams end up in one of three places: too many AI vendors, keys, contracts and invoices; one gateway run by a US company; or a US cloud’s EU region. Here’s how Stav compares to each.
Wiring up OpenAI, Anthropic, Mistral, a hosting provider or two and your own router — and owning all of it.
| DIY multi-vendor | Stav | |
|---|---|---|
| Integrations & keys | One SDK, key and rate-limit budget per provider. | One key for every model, through the OpenAI or Anthropic API you already use. |
| Contracts, DPAs & security reviews | A fresh procurement cycle for every vendor you add. | One contract, one DPA, one review — the whole catalogue. |
| Price | Each provider’s list price, plus the time it takes to run it all. | The providers’ own prices, no markup. A 5.5% fee when you buy credits. |
| Billing | Many invoices, credits, minimums and renewal dates to track. | One prepaid balance for every model; usage and cost in one place. |
| Choosing the right model | You build and maintain routing rules by hand. | With auto, the Smart Router picks the model that gives the best answer for the money, per request. |
| Reliability | You handle every provider outage yourself. | The Smart Router fails over to the next model. For a model you name, you choose the fallbacks. |
| Audit & compliance | Stitch logs together across providers yourself. | One request log with the model, the Inference Provider and the country for every request. |
| Sovereignty | Each vendor’s jurisdiction is your problem to verify. | You set the policy once. A request that would break it is refused, never quietly sent elsewhere. |
A single-API gateway solves the sprawl — but if a US company runs it, it reopens the problem you were trying to avoid.
| A US-hosted gateway | Stav | |
|---|---|---|
| One key, every model | Yes | Yes |
| Who runs the gateway | A US company. The CLOUD Act and FISA 702 reach everything that passes through it. | Stav.ai AS, a Norwegian company, running in Norway. The CLOUD Act does not apply to us. |
| Where open models run | Wherever the gateway prefers — often in the US. |
Azure, AWS Bedrock and Google Vertex all offer models in European data centres. The servers are in Europe; the company that owns them is not.
| A US cloud’s EU region | Stav | |
|---|---|---|
| Contract | Often already in place. | One DPA with a Norwegian company. |
| Who can be compelled | A US parent company. The CLOUD Act applies wherever the servers are. | Stav is Norwegian, and your policy can require European-owned Inference Providers. |
| Where data is processed | An EU region, for the models the cloud offers there. |
One key, every model, one balance — and a policy that decides where your data may go. Moving over takes three lines: the base URL, the key and the model name.
| GPT, Claude and Gemini | Available. | Available too, from their US-owned providers — but only if your policy allows those providers. |
| Proof of where it ran | You take the gateway’s word for it. | Every response names the provider and the country. The request log exports as CSV. |
| Data used for training | Depends on each upstream provider’s terms. | Turn on no-training, and every endpoint that may train on your data is refused. |
| Model choice | That cloud’s own catalogue. Changing clouds is a migration. | Models from every Inference Provider in our catalogue, behind one API. |
| Leaving | Your code is tied to that cloud’s SDK and access controls. | OpenAI- and Anthropic-compatible. You leave by changing the base URL. |