Vynaris Frequently Asked Questions

Q: How does Vynaris ensure quality when routing to cheaper models?

A: Vynaris starts each request with a frontier model and only routes down to a cheaper model when exact-version eval evidence certifies that the cheaper model clears the quality gate for that task. Escalation occurs automatically if needed.

Q: What is a routing receipt?

A: Every response includes headers showing the model that served the request, its cost, and the baseline cost you would have paid directly. This data is also available in an exportable ledger for auditing.

Q: Can I use Vynaris with my existing OpenAI or Anthropic SDK?

A: Yes, Vynaris is fully compatible. Just change the base URL and set model to "auto"—no other code changes are needed.

Q: What are hosted uncensored models?

A: Vynaris offers privately hosted, reduced-refusal models (e.g., Qwen, DeepSeek, GLM variants) with 128K context and no prompt/output retention. These are suitable for lawful security testing, red teaming, and model evaluation.

Q: How does pricing work?

A: You can use prepaid credit with a 3% routing fee (1% after $500 monthly usage) or choose a monthly plan ($20, $50, $150) that provides API credit, which rolls over. Hosted uncensored models are billed per token at published rates.

Q: Can I pay anonymously?

A: Yes, Vynaris accepts Monero (XMR) on the $50 and $150 plans at +20% for complete payment privacy.

Q: Is my data used for training?

A: No, Vynaris does not train on your data. Privacy-restricted providers are used by default, and hosted models have no prompt or output retention.