OpenAI compatibility

Quota is a credit wallet for AI usage: one API key, every major model (OpenAI / Anthropic / Google), and built-in billing so you can charge end-users for their usage instead of eating the bill yourself. The wire format is OpenAI-shaped so the easiest way in is to point your existing OpenAI SDK at Quota — same endpoint path, same request body, same streaming envelope, same error structure, plus a quota block on every response. The list of parameters Quota actually forwards to the upstream provider is below.

Point your OpenAI SDK at Quota

Two lines change. Everything else — model names that start with gpt-, the chat.completions.create call, streaming, tool calling — stays exactly the same.

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.usequota.ai/v1", // <- only change
  apiKey: process.env.QUOTA_API_KEY,      // <- only change
});

const response = await client.chat.completions.create({
  model: "gpt-4o-mini",
  messages: [{ role: "user", content: "Hello" }],
});

console.log(response.choices[0].message.content);
console.log(response.quota); // { credits_used, balance_after, ... }

Provider-prefixed models

Through the same endpoint you can also call Anthropic and Google models by prefixing the model name. No extra SDKs or API keys — Quota translates the request into each provider's native shape and translates the response back.

// Same OpenAI SDK, same call, just a prefixed model.
const claude = await client.chat.completions.create({
  model: "anthropic/claude-sonnet-4.6",
  messages: [{ role: "user", content: "Summarize this" }],
});

Supported parameters

Quota forwards a defined set of OpenAI chat-completions parameters to the upstream provider. Anything outside that set is silently dropped today — the request still succeeds and the response still comes back, but the parameter has no effect. We'd rather you know up front than debug it in production.

Response shape

The same envelope OpenAI returns, plus a quota block with the actual cost, post-call balance, and billing mode. Always check it in production — streaming responses can charge slightly more than the pre-reservation once token usage is finalised.

{
  "id": "chatcmpl-9f3a2b1c",
  "object": "chat.completion",
  "created": 1746823412,
  "model": "gpt-4o-mini",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Hello. How can I help?"
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 12,
    "completion_tokens": 8,
    "total_tokens": 20
  },
  "quota": {
    "credits_used": 18500,
    "balance_before": 8500000,
    "balance_after":  8481500,
    "billing_mode":   "developer",
    "reservation_id": "rsv_5f3a..."
  }
}

Streaming

Streaming uses Server-Sent Events in the OpenAI shape: data: {...}\n\n chunks terminated by data: [DONE]. Existing OpenAI SDK iterators work without changes. The only Quota-specific addition is that the final chunk carries the quota billing block alongside usage. See the Chat completions reference for the full chunk schema and a streaming example.

Limitations and known differences

• Parameter coverage is the table above. Anything not listed as forwarded is dropped silently today. We don't reject the request, so a stray top_p or response_format won't blow up — but it also won't do anything.
• Single completion per request. n isn't forwarded. If you need k samples, fan out k requests.
• No native JSON mode yet. response_format isn't forwarded. Prompt for JSON and parse on your side.
• Reasoning models drop sampling. For o-series and GPT-5+ models, Quota strips temperature and parallel_tool_calls before forwarding — same as if you called OpenAI directly with a reasoning model.
• Per-user attribution uses the bearer, not the user field. The OpenAI user parameter is dropped. To bill a specific end user's wallet, pass their OAuth access token (quota_token_…) as the Authorization bearer instead of your API key — see user billing via OAuth.
• Endpoints surface. Quota implements /v1/chat/completions, /v1/audio/speech (TTS), and /v1/audio/transcriptions (STT) — all OpenAI-shaped, all drop-in for the OpenAI SDK with a swapped baseURL. See Audio for the per-route reference. The /v1/responses, /v1/embeddings, /v1/images/*, and /v1/files surfaces are not implemented.
• Quota-only routes. WS /v1/audio/voice-conversion (voice-to-voice) has no OpenAI SDK shape — it's a WebSocket with a Quota-defined auth frame. Use the @usequota/core / @usequota/nextjs helpers, or speak the protocol from Audio › V2V.

What's next