Build with MegaRider.

Connect to the API, select an available model, and start a conversation.

Quickstart

Request an integration key.
Set your base URL to https://api.megarider.pro/v1.
Call GET /v1/models with your bearer key and choose an exact model ID.
Send a chat completion. Set stream: true for incremental output.

curl https://api.megarider.pro/v1/chat/completions \
  -H "Authorization: Bearer $MEGARIDER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "YOUR_MODEL_ID",
    "messages": [{"role": "user", "content": "Hello, world."}],
    "stream": true
  }'

Keep API keys on your server. Do not put them in public browser code or source control.

API reference

Endpoint	Purpose
`GET /v1/models`	Available model IDs. Bearer authentication required.
`POST /v1/chat/completions`	Text conversations, optional tool definitions, standard or streamed responses.
`GET /v1/provider/models`	Partner integration catalog with capability declarations and proposed rates. Bearer authentication required.
`GET /health`	Gateway liveness; does not establish inference availability.
`GET /status`	Gateway status and the most recent dependency check.

Download the OpenAPI specification.

Streaming and reasoning

Streams use text/event-stream. Read each data: event until [DONE]. Usage is requested from the inference service and forwarded when supplied. Heartbeat comments can appear between data events after the upstream accepts a streaming request.

Reasoning, when available, is exposed separately in reasoning_content. An initial upstream <think> envelope is moved out of the answer content. Tool definitions are passed to the selected model; your application executes any returned tool calls.

An error event or a connection that ends without [DONE] is an incomplete response. Do not treat it as successful output.

Limits and errors

Limits are assigned per API key: requests and tokens per minute, concurrent requests, output per request, lifetime token quota, spend ceiling and expiration. The initial integration has 60 requests per minute; the gateway allows up to 4 active completions overall, with lower per-model concurrency. The payload limit is 1 MiB; output is limited to 8,192 tokens, with a default of 1,024. The request deadline is 120 seconds. These are configured limits, not throughput guarantees.

Before dispatch, we reserve a conservative input estimate plus the requested maximum output against your quota. Usage reported by inference replaces the reservation. If usage is unknown after interruption, reserved capacity remains held pending reconciliation. A quota-related 429 may require a smaller request or a limit change, rather than repeated retries. Spend ceilings use configured customer rates and are not a prepaid balance.

Text input and one completion per request are supported. Images, audio, batch jobs, log probabilities and guaranteed structured JSON outputs are not currently offered.

Status	Action
400 / 413 / 415	Correct the request, payload size or content type.
401	Check your bearer key.
429	For temporary rate or capacity limits, respect Retry-After with exponential backoff and jitter. For exhausted quotas, reduce the requested maximum or contact support.
502 / 503 / 504	Inference failed, a dependency is unavailable or the deadline expired. Retrying may incur additional usage.

Save the X-Request-Id response header when requesting support. We do not automatically retry generation requests.

Service and integrations

MegaRider operates the API gateway and delivery layer. Model inference currently uses external infrastructure through OpenBroker and the Gonka network. Compute location, third-party retention and availability can vary. We do not claim end-to-end zero data retention, fixed data residency or ownership of the underlying GPU fleet.

Partner evaluation access is available by arrangement. Marketplace acceptance, commercial rates, billing terms and capacity require a separate agreement. API compatibility does not imply a partnership with OpenAI or a listing on OpenRouter.

Support

hello@megarider.pro. Include the request ID, time in UTC and error code. Do not include secrets or sensitive prompt content.