AI Gateway

​

What is the AI Gateway?

• One key, multiple providers: Route requests to OpenAI, Anthropic, Google, Mistral AI, and EU cloud backends (eu/...) without managing separate vendor keys.
• Drop-in OpenAI SDK compatibility: Point standard OpenAI SDKs, coding agents, and internal apps at https://connect.omnifact.ai/v1/gateway.
• Unified billing: Consumes your organization’s shared monthly AI usage credits on Billing & Usage.

​

Getting Started

1. Enable API Access: Go to Team Settings > Developer Tools and switch on Enable API Access to generate your organization’s API key (no separate Gateway toggle is required).
2. Enable Models: Ensure the models you want to call are enabled in Team Settings > Models.
3. Authenticate Requests: Pass your API key in Authorization: Bearer YOUR_API_KEY (standard for OpenAI SDKs) or X-API-Key: YOUR_API_KEY.

​

Available Models

​

GET /v1/gateway/models

curl https://connect.omnifact.ai/v1/gateway/models \
  -H "Authorization: Bearer YOUR_API_KEY"

{
  "object": "list",
  "data": [
    {
      "id": "eu/gpt-6.1-terra",
      "object": "model",
      "created": 1756684800,
      "owned_by": "azure-openai"
    },
    {
      "id": "eu/claude-sonnet-5-5",
      "object": "model",
      "created": 1756684800,
      "owned_by": "vertex-anthropic"
    },
    {
      "id": "mistral-large-4",
      "object": "model",
      "created": 1756684800,
      "owned_by": "mistralai"
    }
  ]
}

​

EU-Hosted Model IDs (eu/)

• /v1/gateway/models returns both direct model IDs (for example, gpt-6.1-terra or claude-sonnet-5-5 when US providers are enabled) and EU-prefixed IDs (for example, eu/gpt-6.1-terra on Microsoft Azure EU or eu/claude-sonnet-5-5 on Google Cloud EU).
• Prefixing the model ID with eu/ guarantees routing to the EU-hosted backend; unprefixed US provider IDs route only to direct US providers.

​

Chat Completions

​

POST /v1/gateway/chat/completions

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://connect.omnifact.ai/v1/gateway',
  apiKey: process.env.OMNIFACT_API_KEY,
});

async function main() {
  const response = await client.chat.completions.create({
    model: 'eu/gpt-6.1-terra',
    messages: [{ role: 'user', content: 'Summarize three benefits of zero-trust networking.' }],
    stream: true,
  });

  for await (const chunk of response) {
    process.stdout.write(chunk.choices[0]?.delta?.content || '');
  }
}

main();

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://connect.omnifact.ai/v1/gateway",
    api_key=os.environ["OMNIFACT_API_KEY"],
)

response = client.chat.completions.create(
    model="eu/claude-sonnet-5-5",
    messages=[
        {"role": "user", "content": "Summarize three benefits of zero-trust networking."}
    ],
    stream=True,
)

for chunk in response:
    print(chunk.choices[0].delta.content or "", end="")

curl https://connect.omnifact.ai/v1/gateway/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "eu/gpt-6.1-terra",
    "messages": [
      {
        "role": "user",
        "content": "Summarize three benefits of zero-trust networking."
      }
    ]
  }'

​

Tool Calling (Function Calling)

1. Send tool definitions: Pass up to 128 function definitions in tools (type: "function") alongside your user message, with optional tool_choice ("auto", "none", "required", or a specific function).
2. Execute tool calls: When finish_reason is "tool_calls", inspect message.tool_calls (or streamed delta.tool_calls).
3. Return tool results: Append the full assistant message (both content and tool_calls) followed by one { "role": "tool", "tool_call_id": "...", "content": "..." } message per tool call.

​

Request and Response Constraints

• Messages: The final message must have role user or tool. system and developer roles are permitted only at the start of messages.
• Content: String, array of { type: "text", text: "..." } parts (up to 256 parts, flattened up to 200,000 combined characters per message), or null on assistant messages with tool_calls.
• Size & String Limits:
  • Total JSON request body: up to 1 MB
  • Messages per request: up to 1,024 (200,000 characters per message)
  • Tools per request: up to 128 (tools[].function.description and response_format.json_schema.description each support up to 200,000 characters for rich agent prompts)
  • model, tool_call_id, tools[].function.name, and response_format.json_schema.name: up to 128 characters
  • stop: Non-empty string or array of up to 4 strings (up to 256 characters each)
• Parameters:
  • temperature: 0 to 2 (scaled to each provider’s native range; unsupported models return 400 unsupported_value).
  • top_p: 0 to 1.
  • max_completion_tokens (or max_tokens): max_completion_tokens takes precedence when both are sent.
  • reasoning_effort: none, minimal, low, medium, high, or xhigh (supported on OpenAI / Azure OpenAI reasoning models; returns 400 unsupported_value elsewhere).
  • response_format: text, json_object, or json_schema (json_schema is supported on OpenAI and OpenAI EU models; returns 400 unsupported_value on Anthropic and Gemini).
  • stream & stream_options.include_usage: SSE streaming and final-chunk token usage reporting.
• Strict Parameter Validation: Unknown top-level fields (presence_penalty, frequency_penalty, seed, n, logprobs, or image/audio content parts) return 400 Bad Request with param rather than being silently ignored.

​

Usage & Billing

​

Error Codes

• 400 Bad Request: Invalid message sequence, unknown parameter, context exceeded (context_length_exceeded), or unsupported model/provider parameter (unsupported_value).
• 401 Unauthorized: Missing or invalid API key (invalid_api_key).
• 402 Payment Required: Inactive subscription (subscription_required) or exhausted monthly credits (insufficient_quota).
• 403 Forbidden: API key is not associated with an organization (organization_required).
• 404 Not Found: Requested model does not exist or is disabled (model_not_found).
• 429 Too Many Requests: Upstream provider rate limit reached (rate_limit_exceeded).
• 500 Internal Server Error: Upstream provider terminated generation with an error (completion_failed).
• 503 Service Unavailable: Upstream provider or model is temporarily unavailable (service_temporarily_unavailable or model_temporarily_unavailable).

​

FAQ