Docs

A request, end to end

What happens when your app calls the OpenDoor gateway: key check, policy, provider, and usage write.

Every production call hits the gateway (Bun + Hono), not the Next.js dashboard. Default local URL: http://localhost:3001.

mermaid
flowchart LR
  App["Your app / playground"] --> GW["Gateway :3001"]
  GW --> Auth["API key + org"]
  Auth --> Policy["Policies & spend"]
  Policy --> Route["Provider or deployment"]
  Route --> Log["Request + usage row"]

1. You send an OpenAI-shaped body

POST /v1/chat/completions with Authorization: Bearer opd_… (or the prefix your org issued). The playground does the same thing: it stores a real key named Playground in localStorage and streams from ${NEXT_PUBLIC_GATEWAY_URL}/v1/chat/completions.

Other live routes on the same host:

POST /v1/completions
POST /v1/embeddings
POST /v1/rerank (and /v1/reranking)
POST /v1/batches plus GET /v1/batches/:id
GET /v1/models
GET /health and GET /status

2. The gateway authenticates the key

Auth middleware hashes the bearer token and looks up api_keys for your organization. Revoked keys and missing keys fail here. Restricted keys only see the model ids you ticked when you created them.

3. Policy and spend run before the model

Governance policies, rate limits (RPM/TPM), service tier (standard vs priority), and credit / plan spend are checked before a provider is called. A violation can 403; a spend cap can 402; load-shed on standard can 503.

4. A provider (or your GPU) answers

The model field is a catalog id (qwen2.5-7b-instruct, a vendor id, or custom:<deployment-uuid>). The gateway picks the matching provider adapter (Ollama, Together, DeepSeek, OpenAI, Anthropic, Azure Foundry, a Cloud Run URL, …) and forwards the body.

If you attached LoRAs on a dedicated deployment, that runtime is what answers — not a mock.

5. Usage is written

Prompt/completion tokens, latency, cost, and status land in requests (and roll up for the Usage page). Overview sparklines are the last 30 days of your org’s rows, not sample series.

Where to watch it

Surface	What you see
Dashboard → Overview	30-day request / token / spend / latency from `requests`
Dashboard → Usage	Daily series and per-model breakdown
Dashboard → Audit logs	Key create/revoke and admin actions
`GET /status` (and `/status` on the site)	Gateway, Postgres, Redis, configured providers

Next: Dashboard or Chat completions.

Run locally

Dashboard