Your models.
One interface.

Connect OpenAI and Anthropic-compatible clients to a single HTTPS gateway. Familiar protocols. Streaming responses. No new workflow.

Connect your client Explore the API

A client API key is required for every API request.

chat-completions.shHTTPS / JSON

curl -N https://model.misakacloud.net/v1/chat/completions \
  -H "Authorization: Bearer $MODEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6-luna",
    "stream": true,
    "messages": [
      {"role": "user", "content": "Hello."}
    ]
  }'

TransportHTTPS / JSON requests

InterfacesOpenAI + Anthropic

StreamingSSE / standard SDKs

01 / CONNECTION

Bring your client.

Set the base URL for your SDK. Use a gateway client key supplied by the service operator, not an upstream provider key.

OpenAI-compatible clients

https://model.misakacloud.net/v1

Includes /v1. Send Authorization: Bearer <key>.

Anthropic SDK clients

https://model.misakacloud.net

The SDK adds /v1/messages. Send x-api-key and anthropic-version.

Keys belong in headers. Set MODEL_API_KEY in your environment before running an example. Keys in URL query parameters are rejected. This page never asks for, stores, or sends your API key.

02 / ENDPOINTS

A deliberately small surface.

Five published routes. Choose a model returned by the authenticated models endpoint; model support and availability depend on the selected provider.

GET/v1/modelsDiscover available model IDs

POST/v1/chat/completionsOpenAI Chat Completions

POST/v1/responsesOpenAI Responses, HTTP / SSE

POST/v1/messagesAnthropic Messages

POST/v1/messages/count_tokensAnthropic token counting

Paths and methods are exact. Alternate aliases, trailing slashes, WebSocket routes, and administrative interfaces are not published here.

03 / REQUEST CONTRACT

Know the boundaries.

A predictable interface for long-running agents. These are gateway limits, not a replacement for each model's context window or provider constraints.

[01]

80 MiB per request.

The hard body ceiling is 83,886,080 bytes, including JSON and base64 attachments. It applies to both fixed-length and chunked uploads. Provider limits can be smaller.

[02]

Stream the response.

Set stream: true for SSE. Use curl -N to see events as they arrive. The request-size ceiling does not cap the response length.

[03]

Send uncompressed JSON.

POST requests require a JSON content type. Compressed bodies are rejected. Keep credentials in authentication headers; never put them in the URL.

400Invalid request or query credentials

Check the JSON body and the model's requirements. Remove authentication parameters such as key, api_key, and token from the URL; use an authentication header instead.

401Missing or invalid client key

Use the gateway key supplied by the service operator. OpenAI clients send a Bearer token; Anthropic clients normally send x-api-key. An upstream provider key is not a substitute.

404Route or method not published

Use an exact path and method from the reference above. Management pages, OAuth flows, upload APIs, realtime interfaces, and unspecified aliases are intentionally unavailable on this hostname.

413Request exceeds the body ceiling

Reduce accumulated history, tool results, or inline attachments. Base64 increases binary size by approximately one-third. The ceiling covers the entire serialized request, not just the newest message.

415Unsupported content type or encoding

Send application/json, optionally with a charset parameter, and an uncompressed request body. Do not use multipart uploads or gzip-compressed JSON on these routes.