Deployment Scenarios

These pages are supported deployment scenarios you can start from when bringing up the product.

Use them when you need:

  1. a known-good starting manifest
  2. a deployment pattern that maps directly to rendered gateway config
  3. a concrete example to compare against when your own manifest fails

Supported Scenarios

  • Cost-Control — one provider, one project, one gateway credential, and per-model quota enforcement for a small team.
  • Policy-Governor — Anthropic and OpenAI behind one organization, with two projects that have different native-model allowlists and quotas.
  • Meter-Fallback — DeepSeek as the primary provider (USD) with Mimo as the automatic fallback (Credit), separate monthly spend caps per cost unit, and no per-model rate limits.
  • Translation-Pass — one Anthropic provider serving both Anthropic SDK and OpenAI SDK clients behind the same hostname; native pass-through on /v1/messages, OpenAI-to-Anthropic translation with response normalization on /v1/chat/completions.

Each scenario includes:

  • the plain-language use case
  • the manifest in YAML and JSON tabs
  • the rendered nginx.conf shape produced from that manifest

These scenarios are intentionally minimal — each one isolates a single deployment pattern to keep the manifest readable. Real deployments commonly combine patterns: multiple projects with different policy envelopes, fallback wiring alongside translation, spend caps layered on top of per-model limits, or additional providers alongside native pass-through routes. Use these as starting points and adapt or merge them to fit your organization’s shape.