How it works
Your apps call OpenDoor once.
Keep your OpenAI SDK. Change the base URL. OpenDoor handles access, routing, fallbacks, and metering across serverless and dedicated paths.
Designed for developers
Create a workspace & API key
Sign up, top up $20 to unlock $5 of open-weight credit, and mint a key with model allowlists, RPM/TPM, and spend caps.
Policy runs before the provider
Auth, rate limits, spend-tier TPM unlocks, service_tier (standard / priority), and org governance all gate the request.
Route, cache affinity, observe
Smart routing picks a healthy provider. Prompt-cache affinity sticks successful paths. Tokens and cost land in usage + billing.
Scale beyond chat
Need a GPU? Deploy dedicated. Need a fine-tune? Train → call ft: at base price. Need A/B? Use router:slug.
Example
curl -X POST https://opendoor-gcp.web.app/v1/chat/completions \
-H "Authorization: Bearer $OPENDOOR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "llama-3.1-8b-instruct",
"messages": [{"role":"user","content":"Hello"}]
}'Same shape as OpenAI. Point clients at opendoor-gcp.web.app. Model id comes from the live catalog.
Try the playgroundA request, end to end
Four gates. Then the provider.
01
Auth
Bearer key is hashed and loaded. Disabled or over-cap keys never leave the gateway.
02
Policy
Allowlists, RPM/TPM, spend tiers, and org policies decide if the call proceeds.
03
Route
A healthy provider (or dedicated deployment / router) is chosen. Fallbacks fire on failure.
04
Meter
Tokens, latency, and cost are written. Cached input is billed at the published rate.
Then use it in the product
Playground
Pick a live catalog model and send a real chat completion.
API keys
Mint a scoped key, then drop it into any OpenAI SDK.
Deployments
Stand up dedicated GPUs or local Ollama when serverless is not enough.
Docs
Chat, embeddings, batches, vision, and service tiers — on this site.
SDK & CLI
One key for the terminal and TypeScript client, including OpenBot.
Ready to send the first request?
Create a workspace, mint a key, and keep your existing OpenAI client.