Docs

OpenDoor mirrors Fireworks-style capacity controls on `POST /v1/chat/completions`.

Service tiers & rate unlocks

OpenDoor mirrors Fireworks-style capacity controls on POST /v1/chat/completions.

`service_tier`

Value	Behavior
`standard` (default)	Normal RPM/TPM. May return 503 when `GATEWAY_SHED_STANDARD=1` and in-flight load is high.
`priority`	2× RPM/TPM vs spend unlock; never load-shed.

bash
curl https://YOUR_HOST/v1/chat/completions \
  -H "Authorization: Bearer $OPENDOOR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama-3.1-8b-instruct",
    "service_tier": "priority",
    "messages": [{"role":"user","content":"hello"}]
  }'

Or header: x-opendoor-service-tier: priority.

Spend-tier TPM unlock

Lifetime spend on the API key (spend_used_usd_cents) unlocks higher TPM/RPM floors:

Lifetime spend	TPM floor	RPM floor
$0	100k	60
$10	300k	120
$100	1M	300
$1,000	5M	600
$10,000	20M	1200

Plan multipliers: pro ×3, enterprise ×10. Key-level rate_limit_tpm / rate_limit_rpm still apply as a minimum.

Prompt cache billing

When the upstream reports cached prompt tokens (usage.cached_tokens or prompt_tokens_details.cached_tokens), those tokens bill at cached_input_cost_per_1k (~50% of input). OpenDoor also sets prompt_cache_key and Redis session affinity so repeat prefixes prefer the same provider.

Routing

Dedicated deployments