Merki Inference documentation

Public documentation and reference data for Merki Inference Pte. Ltd.

Overview

• About Merki: what we do, in one page.

Product

• Quickstart: first request in five minutes.
• Models: hosted models, and how to read the catalog.
• Endpoints and compatibility: OpenAI and Anthropic compatible endpoints.
• Authentication: bearer keys, rotation, scopes.
• Bring your own key: run Claude, OpenAI, and other providers through Merki.
• BYOK routing: model-name pass-through and provider failures.
• API keys: key lifecycle and automatic revocation.
• Errors: codes, retry rules, and error shape.
• Rate limits: RPM/TPM ceilings, headers, and 429s.
• Streaming: SSE, mid-stream failures, and billing.
• Usage and metering: token counting and reconciliation.
• Caching: the ZDR exception, lifetime, and hit billing.
• Zero data retention: what ZDR means, and the caching exception.
• OpenAPI and SDKs: base URLs, spec, and client setup.
• Status and incidents: status page and SLA remedies.
• Changelog and deprecation: notices and retirement policy.
• Security: responsible disclosure and safe harbor.
• Content safety: refusals and tier differences.
• FAQ and glossary: short answers and terms.

Access

• Access tiers: Regular, Roleplay, Cybersecurity.
• Tier upgrade: how to request a higher tier.
• Age and identity: age assurance for Roleplay, identity for Cybersecurity.
• Cybersecurity verification: DNS TXT and .merki challenges.

Billing

• Pricing: per-token model, packs, bonus, BYOK, enterprise.
• Credits: credits only, unlimited accounts, and enforcement.
• Enterprise: support, SLAs, and agreements.

Legal

• Terms of service
• Privacy policy
• Acceptable use policy
• Data processing agreement
• Subprocessors
• Cookie policy
• Service level agreement
• Credit and refund policy

Reference data

• Model catalog
• Access tier matrix
• Credit packs
• Revocation triggers
• Retention matrix
• Verification methods