09 / CLIENTS

Use it from your client

deCloud is an OpenAI-protocol-compatible gateway — any client that lets you set a custom base URL can use it. Here's how for the common ones.

Before you start

Create an API key in the console (it starts with dcld-sk-) and make sure the account holds some MA, or calls come back 402. If a key is bound to specific models, calling an unbound model returns 403 model_not_allowed; a key that has spent its quota returns 403 quota_exhausted — raise the limit or reset usage in the console to restore it. Every client uses the same base URL:

https://macdecloud.com/v1

Open the console →

About the protocol

The gateway natively serves OpenAI Chat Completions at POST /v1/chat/completions, the model catalog at GET /v1/models, Anthropic Messages for Claude Code at POST /v1/messages, and OpenAI Responses for Codex at POST /v1/responses. Configure the endpoint for the protocol your client actually speaks.

opencode

Works directly

Terminal coding agent, speaks the OpenAI-compatible protocol.

Write the config

Either the global ~/.config/opencode/opencode.json or an opencode.json in your project root. They merge, project wins.

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "decloud": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "deCloud",
      "options": {
        "baseURL": "https://macdecloud.com/v1",
        "apiKey": "{env:DECLOUD_API_KEY}"
      },
      "models": {
        "moonshotai/kimi-k3": {
          "name": "Kimi K3",
          "limit": { "context": 1048576, "output": 65536 }
        }
      }
    }
  }
}

02
Three things people get wrong
npm must be @ai-sdk/openai-compatible — @ai-sdk/openai targets /v1/responses, which deCloud does not serve. baseURL needs the /v1 but not /chat/completions; the SDK appends that itself. Without limit.context, opencode doesn't know how much fits and will either truncate early or overrun.

Set the key and go

Keep the key out of the config file — a project-level opencode.json usually ends up in git.