API docs

CPEN API URL and model aliases.

Use https://cpenrouter.space/v1 as an OpenAI SDK base URL. Send cpen/* model aliases to chat completions, then choose a Data Safety boundary per key or request.

Open playground

CPEN base URLhttps://cpenrouter.space/v1

Use this exact URL anywhere a compatible SDK, gateway, agent, or AI coding tool asks for an API base URL. Do not append /chat/completions to the SDK base URL; the SDK adds endpoint paths itself. CPEN is independent and not affiliated with OpenAI.

OpenAI SDK baseURL / base_urlChat completions: /v1/chat/completionsModels: /v1/models

Quickstart

CPEN client setup

Point your OpenAI client at CPEN for chat requests. The gpt-* aliases support tool calling and text-only streaming; streaming is text-only, without tools. Usage billing follows measured input and output tokens, with no separate tool-call fee.

curlPOST /v1/chat/completions

curl https://cpenrouter.space/v1/chat/completions \
  -H "authorization: Bearer cpen_live_..." \
  -H "content-type: application/json" \
  -d '{
    "model": "cpen/gpt-5-4-mini",
    "messages": [{"role": "user", "content": "Summarize this document."}]
  }'

OpenAI SDK (Node)https://cpenrouter.space/v1

const client = new OpenAI({
  baseURL: "https://cpenrouter.space/v1",
  apiKey: process.env.CPEN_API_KEY
});

await client.chat.completions.create({
  model: "cpen/gpt-5-4-mini",
  messages: [{ role: "user", content: "Summarize this document." }]
});

OpenAI SDK (Python)https://cpenrouter.space/v1

from openai import OpenAI

client = OpenAI(
    base_url="https://cpenrouter.space/v1",
    api_key=os.environ["CPEN_API_KEY"],
)

response = client.chat.completions.create(
    model="cpen/gpt-5-4-mini",
    messages=[{"role": "user", "content": "Summarize this document."}],
)

Data Safety Levels

Choose DS0-DS4 per API key or with routing.data_safety_level. DS0 performs no payload inspection. DS1 skips payload inspection; account trace storage is off by default, and when enabled can retain policy metadata and model responses for 30 days. DS2 redacts detected supported patterns; DS3 and DS4 block or deny matching requests. DS0 and DS1 have no Data Safety fee. DS2-DS4 add a fee equal to 1% of the model usage charge after cache reconciliation.

DS0 Skip policy inspection

DS0: no Data Safety payload inspection. CPEN still checks the account, model alias, credits, route health, and rate limits before dispatch.

DS1 Inspection off; trace storage optional

DS1: Data Safety inspection is skipped. Account trace storage is off by default; when enabled, it retains request policy metadata and model responses for 30 days.

DS2 Redact protected fields

DS2: detects supported patterns such as emails, phones, cards, government IDs, financial identifiers, and API secrets; detected matches are redacted before dispatch.

DS3 Block sensitive payloads

DS3: blocks requests matching supported sensitive patterns before dispatch.

DS4 Deny suspected personal data

DS4: denies requests matching supported sensitive patterns before dispatch.

Runnable chat aliases

These are the three stable aliases highlighted in this frontend and accepted by /v1/chat/completions. Use /v1/models for runtime discovery.

Model	AI score
cpen/gpt-5-4-mini	78%
cpen/gpt-5-6-luna	88%
cpen/gpt-5-6-terra	90%

Managing keys from your own tooling

Everything the Keys page does is a REST call you can make yourself, authenticated with a signed-in dashboard session cookie.

# List keys (signed-in dashboard session)
curl -s https://cpenrouter.space/api/account/keys \
  -H "Cookie: cpen_session=..."

# Create key
curl -s -X POST https://cpenrouter.space/api/account/keys \
  -H "Content-Type: application/json" \
  -H "Cookie: cpen_session=..." \
  -d '{"label":"ci-bot","rpmLimit":120}'

# Update the per-key request-rate cap
curl -s -X PATCH https://cpenrouter.space/api/account/keys/{keyId} \
  -H "Content-Type: application/json" \
  -H "Cookie: cpen_session=..." \
  -d '{"rpmLimit":1200}'

# Revoke key
curl -s -X DELETE https://cpenrouter.space/api/account/keys/{keyId} \
  -H "Cookie: cpen_session=..."

Authentication

OAuth signup creates an account usage ledger. Paid API calls use cpen_live API keys issued from the account dashboard.

Authorization: Bearer cpen_live_...

Compatibility subset

v1 supports JSON chat with content arrays, image_url/input_image parts, and OpenAI function/tool calling. Text-only streaming is available on the gpt-* aliases; tool calling uses standard buffered responses. Usage billing follows measured input and output tokens, with no separate tool-call fee.

JSON: model, messages, images, tools, tool_choice, assistant tool_calls, tool messages
Streaming: text-only chat on gpt-* aliases with stream and optional stream_options.include_usage
Tool calls: buffered responses (stream + tools returns 400 unsupported_streaming_request)
Billing: measured input and output tokens; no separate tool-call fee

Set Data Safety level

For API requests, choose DS0-DS4 with routing.data_safety_level. Omit it to use the API-key default.

{
  "model": "cpen/gpt-5-4-mini",
  "messages": [{"role": "user", "content": "hello"}],
  "routing": { "data_safety_level": 2 }
}

AI agents and coding tools

For agents that accept a CPEN API endpoint, set the base URL to /v1 and use a cpen_live key. Mini and Luna support tool calling with OpenAI-compatible SSE streaming; streaming is text-only, without tools.

Base URL: https://cpenrouter.space/v1
Model: cpen/gpt-5-4-mini
Minimal body:
{
  "model": "cpen/gpt-5-4-mini",
  "messages": [{"role": "user", "content": "hello"}]
}
Discovery: GET /v1/models

CPEN HTTP endpoints

All JSON APIs live under https://cpenrouter.space/v1 unless noted. Authenticated routes require a cpen_live API key.

Method	Path	Purpose
GET	`/v1/models`	CPEN model list of runnable public model aliases.
POST	`/v1/chat/completions`	OpenAI Chat Completions with buffered responses and documented streaming capabilities.
POST	`/v1/responses`	Non-streaming Responses subset with text and function-call output.
POST	`/v1/extractions`	Structured extraction endpoint.
GET	`/health`	Liveness probe for load balancers.

Response shape

Successful buffered chat completions use the documented JSON shape. Mini and Luna streaming requests use OpenAI-compatible SSE chunks, finish with data: [DONE] only after usage settlement, and can include final usage with stream_options.include_usage. Terra is buffered-only.

{
  "id": "chatcmpl_...",
  "object": "chat.completion",
  "model": "cpen/gpt-5-4-mini",
  "request_id": "req_...",
  "choices": [{
    "index": 0,
    "message": { "role": "assistant", "content": "..." },
    "finish_reason": "stop"
  }],
  "usage": { "prompt_tokens": 42, "completion_tokens": 18, "total_tokens": 60 },
  "cpen_status": "completed",
  "data_safety": { "level": 2, "action": "redacted", "redaction_count": 1 }
}

Common errors

Status	Meaning and action
400	unsupported_field or unsupported_model: correct the request before retrying.
401	Missing or invalid API key: replace the credential before retrying.
402	insufficient_credits: add usage credit before retrying.
422	validation_error: correct the field identified by param/details before retrying.
429	rate_limited: retry with backoff; the API does not currently return Retry-After.
503/502	upstream_unavailable: retry with bounded backoff and retain request_id for support.

{
  "error": {
    "code": "unsupported_model",
    "type": "unsupported_model",
    "message": "Use one of the public cpen/* model aliases.",
    "param": "model",
    "details": {
      "allowed_models": [
        "cpen/gpt-5-4-mini",
        "cpen/gpt-5-6-luna",
        "cpen/gpt-5-6-terra"
      ]
    }
  },
  "request_id": "req_..."
}