Your first request

A closer look at what GateCtr does with each request.

Request lifecycle

Your app
  → GateCtr API
    → Budget Firewall check
    → Context Optimizer (compresses prompt)
    → Model Router (selects model if route: true)
    → LLM provider (OpenAI, Anthropic, Mistral...)
    → Response + analytics logged
  → Your app

Response shape

GateCtr returns an OpenAI-compatible response with an extra gatectr field:

{
  "id": "chatcmpl-abc123",
  "model": "gpt-4o",
  "choices": [
    {
      "message": { "role": "assistant", "content": "Hello!" },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 12,
    "completion_tokens": 8,
    "total_tokens": 20
  },
  "gatectr": {
    "tokens_saved": 18,
    "original_tokens": 30,
    "model_used": "gpt-4o",
    "optimized": true,
    "cost_usd": 0.00024
  }
}

The gatectr field

Field	Description
tokens_saved	Tokens removed by the Context Optimizer
original_tokens	Token count before optimization
model_used	Actual model used (relevant when routing is enabled)
optimized	Whether the Context Optimizer ran
cost_usd	Estimated cost of this request

Supported models

GateCtr is compatible with any OpenAI-compatible model. Tested providers:

• OpenAI — gpt-4o, gpt-4o-mini, gpt-3.5-turbo

• Anthropic — claude-3-5-sonnet, claude-3-haiku

• Mistral — mistral-large, mistral-small

• Gemini — gemini-1.5-pro, gemini-1.5-flash

Use model: "auto" to let the Model Router decide.