Frequently Asked Questions — MCLY Technology

Answers current as of September 2026

Getting started

Do I need to change my code to switch models?

No. Every model is exposed through the same OpenAI-compatible schema. Change the model field from gpt-4o to claude-3-5-sonnet, for example, and the gateway routes to that provider. Only the model name changes.

Which SDKs can I use?

Any OpenAI-compatible client — the official Python and Node SDKs, .NET, Java, Go, Rust, or plain HTTP. Point the base URL at https://api.mclytechnology.com/v1 and use your MCLY key. See the API docs for working examples.

Do you support streaming and embeddings?

Yes. Server-sent-event streaming works exactly as it does upstream, and the embeddings endpoint covers every embedding model in the catalogue. Tool calling, JSON mode and vision inputs are passed through to any upstream model that supports them.

How do I get an API key?

Keys are issued to accounts, so start by ordering a plan. Once the order is provisioned you will receive a key bound to that account. Sign in at any time to review associated usage and orders.

Billing and payment

Can I pay by bank transfer?

Yes — larger orders are settled by bank transfer or corporate card. Place the order, then email [email protected] with your order number and we will send the details the same business day. Card payments are being brought online; until then, orders route through our team.

How are token credits consumed?

Credits are drawn down per request based on the upstream provider's token accounting, at the rate card attached to your plan. Unused credits stay on the account until the validity window in your plan ends.

Which currency are invoices in?

US dollars. Orders are quoted in USD and taxes are added where applicable on the invoice issued by MCLY TECHNOLOGY INC.

What happens if a request fails?

Errors attributable to the gateway are not metered. If an upstream provider fails, our routing layer retries another healthy region or returns an explicit error rather than charging for a partial result. See the refund policy for details.

Compute

Which GPU instances are available?

RTX 4090, A100 and H100 instances, from single cards to eight-GPU clusters, reserved monthly with root access and persistent volumes. Hourly burst options exist for short workloads — see the compute section on the pricing page.

Can I deploy my own model on a reserved instance?

Yes. Instances ship with root access, so you can run vLLM, TGI, Ollama or your own serving stack. You are responsible for the licence terms of the weights you deploy, and for securing the instance.

Can I get more quota or a dedicated deployment?

Yes. Rate limits, quota increases and isolated deployments are arranged through contact — tell us the models, peak request rate and data-residency constraints and we will size it.

Trust and data

Is my prompt data used for training?

No. Requests are forwarded to the upstream provider you selected and are not retained by us for training or any secondary purpose. See the privacy policy for what metadata we keep and why.

What uptime do you commit to?

99.9% monthly for the gateway API, with service credits if we miss it. The full measurement method, exclusions and credit tiers are in the service level agreement.

Still stuck?

Email [email protected] or open a ticket from the contact page. Engineers, not a call centre, reply within one business day.