An OpenAI-compatible gateway for your agents

Quickstart

Base URL for all OpenAI-compatible calls:

https://promptopoly.com/api/v1/proxy/v1

Create a proxy key in the cockpit at /app → AI Proxy. The key starts with pk_ and is shown once.

Your first request (OpenAI-compatible):

curl -s https://promptopoly.com/api/v1/proxy/v1/chat/completions \
  -H "Authorization: Bearer pk_REPLACE_WITH_YOUR_PROXY_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o-mini","messages":[{"role":"user","content":"Sag Hallo."}]}'

Authentication

Every request carries the proxy key. Three equivalent ways — the OpenAI-conformant Bearer header is recommended:

• Authorization: Bearer pk_… — recommended, works with every OpenAI SDK out of the box.
• X-Proxy-Key: pk_… — alternative header.
• x-api-key: pk_… — fallback for the Anthropic SDK / Claude Code (which sends the key as x-api-key).

Endpoint reference

Error format & status codes

Errors are returned as an OpenAI-compatible object — typed exceptions in the OpenAI SDK work:

{
  "error": {
    "message": "Budget exceeded for today.",
    "type": "insufficient_quota",
    "code": "budget_exceeded",
    "param": null
  },
  "limit_info": { "budget": "daily", "used_cents": 1234, "limit_cents": 1000 }
}

An unknown model returns model_not_found with an available:-list of allowed models in error.message.

Integration examples

OpenAI SDK — Python

from openai import OpenAI

client = OpenAI(
    api_key="pk_REPLACE_WITH_YOUR_PROXY_KEY",
    base_url="https://promptopoly.com/api/v1/proxy/v1",
)

resp = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Sag Hallo in einem Satz."}],
    stream=False,
)
print(resp.choices[0].message.content)

OpenAI SDK — Node.js (ESM)

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "pk_REPLACE_WITH_YOUR_PROXY_KEY",
  baseURL: "https://promptopoly.com/api/v1/proxy/v1",
});

const resp = await client.chat.completions.create({
  model: "gpt-4o-mini",
  messages: [{ role: "user", content: "Say hello in one sentence." }],
});
console.log(resp.choices[0].message.content);

LangChain

from langchain_openai import ChatOpenAI
from langchain_core.messages import HumanMessage

llm = ChatOpenAI(
    model="gpt-4o-mini",
    api_key="pk_REPLACE_WITH_YOUR_PROXY_KEY",
    base_url="https://promptopoly.com/api/v1/proxy/v1",
)

answer = llm.invoke([HumanMessage(content="Fasse Promptopoly in einem Satz zusammen.")])
print(answer.content)

Claude Code / Anthropic SDK — native /v1/messages

Claude Code uses the same base URL via the environment variable ANTHROPIC_BASE_URL:

export ANTHROPIC_BASE_URL="https://promptopoly.com/api/v1/proxy"
export ANTHROPIC_API_KEY="pk_REPLACE_WITH_YOUR_PROXY_KEY"

The Anthropic SDK calls the native endpoint directly:

from anthropic import Anthropic

client = Anthropic(
    api_key="pk_REPLACE_WITH_YOUR_PROXY_KEY",
    base_url="https://promptopoly.com/api/v1/proxy",
)

msg = client.messages.create(
    model="claude-3-5-haiku-latest",
    max_tokens=256,
    messages=[{"role": "user", "content": "Sag Hallo in einem Satz."}],
)
print(msg.content[0].text)

Generic agent framework

Works with any OpenAI-compatible agent framework: set the provider to "OpenAI-compatible", then set the base URL, key and model.

{
  "provider": "openai-compatible",
  "model": "gpt-4o-mini",
  "api_key": "pk_REPLACE_WITH_YOUR_PROXY_KEY",
  "base_url": "https://promptopoly.com/api/v1/proxy/v1",
  "options": { "stream": true, "temperature": 0.7 }
}

cURL — streaming + non-streaming

# Non-streaming
curl -s https://promptopoly.com/api/v1/proxy/v1/chat/completions \
  -H "Authorization: Bearer pk_REPLACE_WITH_YOUR_PROXY_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o-mini","messages":[{"role":"user","content":"Sag Hallo."}]}'

# Streaming (SSE): text/event-stream, endet auf data: [DONE]
curl -N https://promptopoly.com/api/v1/proxy/v1/chat/completions \
  -H "Authorization: Bearer pk_REPLACE_WITH_YOUR_PROXY_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o-mini","stream":true,"messages":[{"role":"user","content":"Zähle bis drei."}]}'

Operations

Hard budgets & wallet

Daily, weekly or monthly budgets are reserved race-safely before the call leaves the gateway. Prepaid wallet debit in cents — no call without sufficient balance.

Model aliases

Stable alias names (e.g. gpt-4o → gpt-4o-mini) decouple your code from concrete provider models. Aliases whose target does not exist are not created; resolution is fail-open.

Failover

If a provider fails (timeout/5xx) before the first byte is streamed, the gateway routes to the next model in the failover chain. The native Anthropic endpoint has no provider failover due to its format.

Audit & GDPR

Every call is recorded in a tamper-proof hash chain. Response payloads are automatically removed after 7 days (retention). Made in Germany, GDPR-compliant.

Rate limits

On top of budgets, a rate limit protects the gateway. Exceeding it returns 429 (rate_limit_error).

Limits

• n > 1 (multiple completions per request) is not supported.
• Native /v1/messages is limited to the Anthropic provider (no failover, no cross-alias model rewrite).
• If provider usage is missing from a stream, the gateway estimates tokens (approx. 4 chars/token) and marks usage_estimated=1.