Frequently asked.

Open models are inexpensive but operationally unreliable: providers silently quantize, throttle, and go down. Keln is the reliability layer. It continuously does quality verification, predictive routing, and output normalization, so you get consistent quality, low tail latency, and zero-downtime failover behind one price and one API.

A single provider is a single point of failure: when they go down, your product goes down, and when they serve a quantized model, your output quality drops. Catching that yourself means retry code, benchmarks, and quality monitoring. Keln removes the single point: a fleet of providers behind one endpoint, verified continuously, rerouted automatically, recovered mid-stream when a node fails.

Most gateways act like a proxy: they forward your request to a provider and hand back what comes out.

Keln actively manages what's behind the endpoint. Every provider is verified continuously, outputs are normalized to one schema so tool calls and reasoning fields behave the same everywhere, and routing is optimized for speed with a learned first-token deadline, a hedged backup route, and mid-stream failover. Enterprise plans add a contractual SLO on latency and uptime. Keln is a reliability layer for open models, not just a router.

Every request is scored against live health and performance signals for each provider: TTFT, throughput, error rate, and quality checks. Keln only routes to capacity that clears your targets, an expected latency under 5 s to first token and at least 50 tok/s, and it predicts slowdowns before they land. If a provider stalls mid-stream, Keln fails over to healthy capacity, so you always end up on the fastest verified provider.

Minutes. Point your existing OpenAI-compatible client at Keln’s endpoint, swap in your API key, and you’re live, no SDK migration and no infrastructure changes.

Yes. Keln is OpenAI-compatible, so any OpenAI SDK or tool works out of the box, just change the base URL and key. Keln is also 100% compatible with the OpenRouter and Vercel AI SDK interfaces.

Keln sells inference to developers and businesses, and routes those requests to independent GPU fleets. You connect your nodes, Keln fingerprints and benchmarks them, and admitted nodes join the routing pool. Traffic is white-labeled in both directions, and Keln handles billing, support, and the customer relationship.

Yes, that is the point. Sign up, e-sign the provider agreement, and run one command on each box. The agent registers your GPUs and the models they can serve. Set a utilization cap so Keln only ever fills the headroom you offer, and your own workloads always come first.

You are paid per token at a fixed, published rate for each model, so you never have to undercut anyone to win traffic. Earnings accrue in real time and you withdraw on your own schedule.

No. Traffic is fully white-labeled in both directions. You never see whose requests you serve, and their end users never see you. Billing and support stay with Keln.

Automatically. Once you e-sign the provider agreement you can connect a node immediately. Most providers are serving paid traffic within an hour.

Yes. Set a utilization cap, reserve specific GPUs, schedule maintenance windows, or pause entirely. Keln only ever fills the headroom you offer.

Only what's needed to serve the request, in transit. Nothing is persisted on your side: zero data retention is enforced platform-wide.

Keln keeps a small margin on each model it serves. You see your rate per token up front, before you connect. Details in the Provider docs ↗

Earnings accrue in real time. Withdraw via ACH, instant-to-debit, or wire, settled through Stripe on the schedule you choose.