FAQ and glossary
Short answers and the terms Merki uses.
FAQ
Why is inference cheaper here? The proprietary Merki serving stack. Hosted prices sit 25–35% below the cheapest published list price for the same model. See About and Pricing.
Do I need an SDK? No. Point an OpenAI or Anthropic SDK at Merki. See OpenAPI and SDKs.
What does a cache hit save? Half the input cost on matched tokens, plus latency. See Caching.
What happens if my key leaks? It is revoked automatically, on git commit or public exposure, typically within 24 hours of detection. Rotate it. See API keys.
How do I get Roleplay or Cybersecurity? Request the tier; complete the check; lower tiers stay available. See Tier upgrade.
Are unused credits refunded? Within 14 days of purchase where required by law or at Merki's discretion; drawn-down credits are never refunded. See the Credit and refund policy.
Glossary
| Term | Meaning |
|---|---|
| BYOK | Bring your own key: route through your provider key. See BYOK. |
| TTFT | Time to first token: latency to the first streamed token, per catalog row. |
| Tok/s | Output tokens per second, measured by Merki per catalog row. |
| ZDR | Zero data retention: no prompt/completion retention except cache entries. |
| Quant | Quantization served, e.g. FP8, BF16, INT4, GGUF. |
| Challenge | DNS TXT or .merki file proving domain control, max 7 days. |
| Bonus credits | 5% on single top-ups of 100 USD or more, lifetime cap 300 USD per account, spent first. |
Next step
Start building: Quickstart.