Glossary

A 49-character bearer token (sk-kt- plus 43 base62 characters) that authenticates data-plane requests. Shown exactly once at creation; only a SHA-256 hash is stored. Scoped to one organization. See Authentication.

A request to POST /v1/chat/completions, the OpenAI-compatible endpoint for conversational models. Supports streaming via stream: true.

A single optional dial from 1 to 10, used when the smart router's pool mixes cheaper open-weight models with stronger ones. Low lets smaller models win more often; high sends more traffic to your strongest models. See Price & performance.

The prepaid balance an organization holds, denominated in micro-EUR. Each request debits the cost of the tokens it used, estimated when a provider reports no token counts. An exhausted balance returns 402 insufficient_credits. See Rate limits & credits.

A routing method you configure on the smart router that selects among your models on live metrics, such as latency, throughput, or error rate, rather than (or on top of) prompt classification.

The OpenAI-compatible request path at https://api.kontinent.ai: chat completions, responses, embeddings, transcription, speech and model listing. Distinct from the dashboard, where keys and credits are managed.

A vector representation of text, returned by POST /v1/embeddings. Models declare kind: "embedding"; sending one to the chat endpoint is rejected.

The rate limiter's behavior when its backing store has trouble: requests pass rather than being blocked. You are never rate-limited by an infrastructure hiccup on our side.

Automatically retrying a request on the next route after an upstream 429, 5xx, 401/403 or connection error. The next route is a model's standby route, where the catalog has one, or the next model in your routing policy. For streams it happens only before the first byte. See Maximum availability.

What a request leaves behind: token counts, latency, status, requested and served model, provider, API key, your end-user id if you send one, and computed cost. Never the prompt or the completion. See Privacy & retention.

The integer money unit used everywhere in Kontinent: 1 EUR = 1,000,000 µEUR. Prices are quoted per million tokens. Integers avoid floating-point rounding in billing.

A logical model address in provider/model form, e.g. mistral/mistral-medium-3.5. One id can resolve to several upstream routes. Unknown or disabled ids return 404 model_not_found.

The account boundary a key belongs to. It owns the credit balance and the usage attribution for every request made with its keys. Rate limits are set per key.

An upstream 4xx other than 401, 403 and 429, returned unchanged with its original status and the provider's own body. See Handling errors.

The sovereignty tier for a model served from the EU by a provider with a non-EU parent or on third-party cloud infrastructure.

The default route for a model: the provider that serves the request whenever it is healthy.

One upstream provider path behind a logical model id. A model has a primary route and, for many models, a standby route on another provider used for failover.

Requests per minute: the per-key rate limit, enforced over a sliding window of two 1-minute buckets. Exceeding it returns 429 rate_limit_exceeded.

A configurable router that classifies each incoming request and sends it to one of the models you picked for its pool. Kontinent does not pick models you didn't allow. See Price & performance.

A per-model declaration of how strict its data residency is: strict for an EU-owned provider rated sovereign, pragmatic for a provider serving from the EU with a non-EU parent or on third-party cloud infrastructure. Returned on every model from GET /v1/models. See Models & sovereignty.

A fallback route for the same logical model, on a different provider, that takes over when the primary can't serve a request.

Receiving tokens as they are generated over Server-Sent Events, enabled with stream: true. Where the provider reports usage, the final chunk before data: [DONE] carries it. See Streaming.

The sovereignty tier for a model served by an EU-owned provider that Kontinent rates as sovereign.

The design contract that request and response bodies are never persisted or logged anywhere in the pipeline, enforced by an end-to-end test that fails CI if a prompt reaches the logs. See Privacy & retention.