One OpenAI-compatible gateway in front of every major model. Use one base URL and one Miavo key, then choose the model route slug that fits your provider, region, and compliance needs.
Send your sk-maas-... key as a bearer token. Provider-side credentials stay in the Miavo pool.
POST https://api.miavo.xyz/v1/chat/completions
Authorization: Bearer sk-maas-...
Content-Type: application/jsonSet request_format to anthropic or gemini on POST /v1/chat/completions to send that provider’s native request body and receive its native JSON or SSE response. Use one matching model route from /v1/models. Omit the field for OpenAI Chat Completions.
curl https://api.miavo.xyz/v1/chat/completions \
-H "Authorization: Bearer sk-maas-..." \
-H "Content-Type: application/json" \
-d '{"request_format":"anthropic","model":"YOUR_CLAUDE_ROUTE","max_tokens":256,"messages":[{"role":"user","content":[{"type":"text","text":"Hello"}]}]}'curl https://api.miavo.xyz/v1/chat/completions \
-H "Authorization: Bearer sk-maas-..." \
-H "Content-Type: application/json" \
-d '{"request_format":"gemini","model":"YOUR_GEMINI_ROUTE","contents":[{"role":"user","parts":[{"text":"Hello"}]}],"generationConfig":{"maxOutputTokens":256}}'A coding agent with access to your local terminal can test the API without receiving your secret in chat. Keep the key in a temporary environment variable and let the agent report only redacted results.
Test GET https://api.miavo.xyz/v1/models with my Miavo API key.
Do not ask me to paste the key into chat. Do not write it to source code, command history, logs, screenshots, or reports.
Create a script that reads MAAS_API_KEY from the environment and ask me to enter it through a no-echo terminal prompt.
Print only the HTTP status, content-type, cf-ray, x-maas-request-id, model count, and a redacted error body.
Unset the environment variable when finished.If the AI cannot use your local terminal, run the generated script yourself and send it only the redacted output—not the key.
GET /v1/models returns concrete route slugs. It only includes routes with an active upstream key mapped to that model. Coming soon routes stay visible in the public catalog but are not returned here.
curl https://api.miavo.xyz/v1/models \
-H "Authorization: Bearer sk-maas-..."Docs stay short: model-specific examples, prices, capabilities, and provider routes live on model detail pages. 67 canonical models / 67 provider routes / 67 available
| Model family | Canonical models | Available routes | Modalities |
|---|---|---|---|
| Claude | 5 | 5 | Text |
| OpenAI | 12 | 12 | Image, Text |
| Gemini | 15 | 15 | Image, Text, Video |
| Kimi | 4 | 4 | Text |
| MiniMax | 1 | 1 | Text |
| Qwen | 23 | 23 | Image, Text, Video |
| DeepSeek | 2 | 2 | Text |
| GLM | 4 | 4 | Text |
| Seedance · ByteDance | 1 | 1 | Video |
When a canonical model is available from multiple providers, append a routing suffix to let Miavo rank the eligible routes and fall back automatically.
| Mode | Model suffix | Optimizes for |
|---|---|---|
| Balanced | :balanced | Price, reliability, latency, and throughput |
| Fast | :fast | Output throughput with latency and health safeguards |
| Agent reliable | :agent | Tool support and recent route success rate |
const response = await client.chat.completions.create({
model: 'claude-opus-4-8:fast',
messages: [{ role: 'user', content: 'Summarize this document.' }],
});The response reports the concrete provider route that handled the request. To pin a request to one provider, use that route's unsuffixed model slug instead.
All errors are OpenAI-shaped: HTTP status plus JSON body with error.code, error.message, and error.type.
{
"error": {
"code": "rate_limit_exceeded",
"type": "rate_limit",
"message": "RPM exceeded - 200 requests in 60 seconds."
}
}| Status | Code | When |
|---|---|---|
| 401 | authentication_error | Missing or invalid sk-maas key. |
| 402 | budget_exhausted | Credit balance is below the required request hold. |
| 403 | model_not_allowed | Model is not on this key allow-list. |
| 400 | invalid_request | Body failed schema validation. |
| 429 | rate_limit_exceeded | Per-key RPM or TPM tripped. |
| 502 | upstream_error | Provider returned an error after we forwarded. |
| 503 | no_upstream | No pooled credential is available for this model. |
Each API key carries its own rpmLimit and tpmLimit, both rolling 60-second windows. New keys have no platform RPM or TPM cap. A value of 0 means unlimited. Existing configured limits, account balances, budgets, and upstream provider quotas still apply.
Miavo implements the OpenAI Chat Completions schema. Any SDK that targets OpenAI works by swapping baseURL and apiKey.
- const client = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });
+ const client = new OpenAI({
+ baseURL: 'https://api.miavo.xyz/v1',
+ apiKey: process.env.MACAW_API_KEY,
+ });
- model: 'gpt-4o-mini',
+ model: 'claude-opus-4-8'