Models & Routing

Model Selection

When you specify a model in your API request (e.g., gpt-4o-mini), the gateway finds all operators serving that model and routes your request to the best available one.

Routing Strategy

By default, requests are routed based on a composite score of operator latency, uptime, and price. You can override this with the X-Tangle-Operator header to pin a specific operator.

Fallbacks

If the selected operator fails or times out, the gateway automatically retries with the next-best operator. This happens transparently — your application sees a single successful response.

Operator Selection Headers

# Pin to a specific operator
X-Tangle-Operator: tangle-core

# Prefer operators in a region
X-Tangle-Region: eu-west

# Set a max price (per million tokens)
X-Tangle-Max-Price: 5.00