# Rill OpenAI API endpoint: https://rill.aiinflow.com

> A transparent reverse proxy for the OpenAI API. Use https://rill.aiinflow.com/v1 as the OpenAI base URL with your own OpenAI API key. Requests and responses are byte-for-byte the OpenAI API; nothing else changes.

## Use

- Base URL: https://rill.aiinflow.com/v1
- Auth: `Authorization: Bearer <your OpenAI API key>` (required on every /v1 request; the key is billed, never stored)
- Environment: `OPENAI_BASE_URL=https://rill.aiinflow.com/v1` plus the usual `OPENAI_API_KEY`. Official OpenAI SDKs and most tools need no code change.
- Python: `OpenAI(base_url="https://rill.aiinflow.com/v1")`
- Node.js: `new OpenAI({ baseURL: "https://rill.aiinflow.com/v1" })`
- Streaming: `"stream": true` works; Server-Sent Events are forwarded as they are produced. With curl use `-N`.

## Endpoints

Every HTTP path under /v1/ is forwarded unchanged (method, query, headers, body). Verified:

- POST https://rill.aiinflow.com/v1/chat/completions (plain and streaming)
- POST https://rill.aiinflow.com/v1/responses (plain and streaming); GET /v1/responses/{id}
- POST https://rill.aiinflow.com/v1/embeddings
- POST https://rill.aiinflow.com/v1/decisions (e.g. model gpt-6-luna; predicate / choice / score questions)
- GET  https://rill.aiinflow.com/v1/models

Also forwarded as-is: images, audio (speech, transcriptions), files, uploads, batches, moderations, vector_stores, assistants, fine_tuning, legacy completions.

Not supported: WebSocket / Realtime (`/v1/realtime`) returns 501.

## Examples

```bash
curl -N https://rill.aiinflow.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"gpt-4.1-mini","input":"Say hello","stream":true}'

curl https://rill.aiinflow.com/v1/chat/completions \
  -H "Authorization: Bearer $OPENAI_API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"gpt-4.1-mini","messages":[{"role":"user","content":"Hello"}]}'

curl https://rill.aiinflow.com/v1/decisions \
  -H "Authorization: Bearer $OPENAI_API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"gpt-6-luna","input":"The mug arrived cracked, I want my money back","questions":[{"type":"predicate","name":"refund_requested","instructions":"Does the customer ask for a refund?"}]}'
```

## Errors

- Errors in OpenAI's JSON shape (`{"error":{...}}` with OpenAI wording) come from OpenAI unchanged, including status, `x-request-id` and rate-limit headers. Retry on 429 / 5xx per `retry-after`.
- From this proxy:
  - 401 `Supply your OpenAI API key as Authorization: Bearer <key>.`: no Bearer key was sent.
  - 404 `Use /v1/* for API requests.`: the path is not under /v1/ (check the base URL ends in /v1, without a second /v1).
  - 405 method not allowed; 501 WebSocket not supported; 502 upstream unreachable.
- 403 `unsupported_country_region_territory` comes from OpenAI: the caller is not in a supported region. The proxy cannot change this.

## Other paths on this host

- GET https://rill.aiinflow.com/ : this guide as a web page
- GET https://rill.aiinflow.com/llms.txt : this file
- GET https://rill.aiinflow.com/health : proxy liveness (`{"status":"ok","service":"rill"}`), not OpenAI status

## Notes

- No response cache; no logging of keys, prompts or completions by Rill. The host's infrastructure may keep its own logs.
- Source and self-hosting (Cloudflare Workers Free or a single binary): https://github.com/yzfly/rill