HAI Gateway Docs

Unified API (Preview)

Unified API Chat Completions

Call OpenAI, Anthropic, Google, and other providers through a single OpenAI-format endpoint. The gateway detects each model's native schema and performs the conversion transparently.

POSTv1/chat/completions
Preview · Gated Rollout

This endpoint is currently in gated rollout and is available to a limited set of tenants only. If your request returns a permission error, your account has not been added to the allowlist yet. For production, please use the stable native endpoints (/openai, /anthropic, /google). API behavior may change in future releases during the preview without prior notice.

Authorizations

Authorizationstringheaderrequired

Use the following format for authentication: Bearer sk-pat-YourAccessToken

The key value is your HAI Gateway Personal Access Token (starting with sk-pat-), identical to the OpenAI Relay channel. No separate API key from any upstream provider is required.

Base URL

https://api.hai.network/unified-preview/openai

Body Parameters

The request body format is fully compatible with the official OpenAI Chat Completions API. For detailed parameter descriptions, refer to the OpenAI official documentation.

Set model to any configured model ID (retrieve the full list via List Models). The gateway automatically detects its native schema:

  • OpenAI / compatible providers: passthrough, no format conversion
  • Anthropic (claude-*): converted to Anthropic Messages format upstream; response stream is translated back to OpenAI chunks
  • Google (gemini-*): converted to Google generateContent upstream; response stream is translated back to OpenAI chunks

Streaming Request Notes

Consistent with the native OpenAI endpoint, this endpoint always returns usage in streaming responses:

  • If the request body does not contain stream_options, {"include_usage": true} is added automatically
  • If stream_options already exists, include_usage is forced to true

Preview-period Feature Limitations

When calling Anthropic / Google models through the unified endpoint, the following features are unavailable or behave differently. If you need these capabilities, use the native endpoints for the respective providers:

  • Basic tool calling: the format is mappable, but parallel calls and tool_choice semantics differ
  • Structured output: implementations diverge too much across providers; not supported
  • Extended thinking / reasoning: Anthropic thinking blocks and OpenAI reasoning are provider-specific; not supported
  • Prompt caching: Anthropic's explicit cache_control has no OpenAI counterpart; not supported
cURL
cURL:OpenAI
curl https://api.hai.network/unified-preview/openai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-pat-YourAccessToken" \
  -d '{
    "model": "gpt-5.4-2026-03-05",
    "messages": [
      {
        "role": "developer",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Hello!"
      }
    ]
  }'
200
200:Default
{
  "id": "chatcmpl-9XYZ...",
  "object": "chat.completion",
  "created": 1741132800,
  "model": "claude-opus-4-6",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Hello! How can I help you today?"
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 12,
    "completion_tokens": 10,
    "total_tokens": 22
  }
}