API Reference Overview

Use the GonkaGate API at https://api.gonkagate.com/v1 with Authorization: Bearer gp-..., send generation requests to POST /v1/chat/completions, and refresh model IDs from GET /v1/models. Requests use the OpenAI-compatible chat format and return either JSON or SSE. Need the first working request? Start with Quickstart.

API basics

• Base URL: https://api.gonkagate.com/v1
• Auth: Authorization: Bearer gp-...
• Main generation endpoint: POST /v1/chat/completions
• Model list endpoint: GET /v1/models
• Non-streaming responses: JSON with choices and usage
• Streaming responses: add stream: true and parse Server-Sent Events (SSE)

Request skeleton

Use this shape for any basic POST /v1/chat/completions call. Replace model with a fresh ID from GET /v1/models.

export GONKAGATE_API_KEY="gp-your-api-key"

curl https://api.gonkagate.com/v1/chat/completions \
  -H "Authorization: Bearer $GONKAGATE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "<model-id-from-get-v1-models>",
    "messages": [
      {
        "role": "user",
        "content": "Reply with the word pong."
      }
    ]
  }'

The minimum request is base URL, Bearer auth, a current model ID, and OpenAI-compatible messages.

Request rules

• Send JSON to POST /v1/chat/completions.
• Include Authorization: Bearer gp-... and Content-Type: application/json.
• Set model to a fresh ID from GET /v1/models.
• Send messages in the OpenAI-compatible chat format.
• Add stream: true only when your client already parses SSE.

Response and failure basics

• Non-streaming responses are JSON. Read the model output from choices[0].message.content.
• Streaming responses arrive as SSE data: events and end with [DONE].
• Classify failures by HTTP status + error.code, not status alone.
• Keep x-request-id when a request fails or behaves unexpectedly.

Common mistakes

• Using a stale model ID instead of refreshing it from GET /v1/models.
• Debugging auth, model selection, and request body shape at the same time. Check them in that order.
• Treating every 429 as retryable throttling. insufficient_quota is a billing state, not a backoff case.
• Treating this overview as the full schema. Exact request and response fields live on the endpoint reference pages.

See also

• Create a chat completion for the full POST /v1/chat/completions request schema, JSON vs SSE responses, error payloads, and x-idempotency-key.
• GET /v1/models for current machine-readable model IDs.
• Streaming Responses for stream: true, SSE parsing, and end-of-stream handling.
• API Error Handling for retry, stop, and escalation rules.
• Authentication and API Keys if the Bearer header or key state is still in doubt.