$ curl https://api.lxg2it.com/v1/chat/completions \
    -H "Authorization: Bearer $KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "model": "standard",
      "prefer": "cheap",
      "messages": [{"role": "user", "content": "Hello"}]
    }'

You pick what matters — the capability tier and the optimisation direction. The router picks the model. When a cheaper option launches or a provider goes down, your requests adapt automatically. No model names to track. No code to change.

You can also pass a specific model name (gpt-4.1, claude-sonnet-4-6) to pin routing and bypass tier selection entirely.

Context-window guard: never routes to a model that can't handle your input. Circuit breakers reroute around provider outages automatically.

Available models →

Fast models via Groq and Cerebras are routed at no cost — no credits, no card required. Sign up and start making requests immediately. Add credits when you need the full range of premium models.

A 4% fee on credit deposits. Requests are billed at actual provider market rates — you pay what the model costs, nothing more. Every response includes X-Model-Router-Model and X-Model-Router-Provider headers so you always know exactly what ran and what it cost.

Set model: "auto" and the router analyses your full conversation context — system prompt, code blocks, message history, tool use, reasoning markers — and picks the right tier automatically. No heuristics on individual messages; it reads the whole picture. Every auto-routed response includes X-Model-Router-Auto-Tier and X-Model-Router-Auto-Score headers so you can see exactly why a tier was chosen. How it works →

Circuit breakers detect provider outages and reroute requests in real time. Context-window guards ensure requests never go to a model that can't handle them. Your code doesn't change — routing adapts automatically.

Block providers you don't want to fund. Set daily spend limits. Enable auto-recharge so you never hit a wall mid-project. Export request traces to any OTLP backend — Axiom, Grafana, Honeycomb, Datadog.

Same key, same endpoint pattern. embed-small, embed-large, and embed-titan aliases route to the best available embedding model. Batch inputs, optional dimension truncation, billed at input tokens only.

Full documentation →