Switch from another gateway
Move an app from OpenRouter, LiteLLM, Vercel AI Gateway, Portkey or another LLM gateway to the Ertas AI Gateway, with a prompt for your coding agent.
On this page
If your app already calls models through an OpenAI-compatible gateway, moving an API call to Ertas changes three things: the base URL, the key and the model ID.
| Change | From | To |
|---|---|---|
| Base URL | Your current gateway's URL | https://gateway.ertas.ai/v1 |
| Key | Your current gateway key | ERTAS_API_KEY |
| Model | Your current model name | ertas-ai/medertas-27b or ertas-ai/medertas-31b |
What to check before you switch
| Coming from | Check |
|---|---|
| OpenRouter | Map each model name to an Ertas model ID rather than assuming the same names exist |
| LiteLLM | Model aliases, fallbacks and proxy-only settings in your config |
| Vercel AI Gateway | Use the AI SDK's OpenAI-compatible provider for Ertas instead of gateway-specific methods |
| Portkey | Remove x-portkey-* headers and virtual keys from the Ertas client only |
| Helicone | Remove Helicone proxy URLs and headers from the Ertas client only |
| Cloudflare AI Gateway | Provider-specific paths, gateway headers and caching settings |
| Kong AI Gateway | Routes, plugins and policies that need an equivalent before traffic moves |
| Others | Gateway-specific features your app depends on |
Keep the old route in place until the new one works on real traffic, and move one feature at a time.
Migration prompt
Help me move one model call in this project from <current gateway> to the Ertas AI Gateway.
Find where the OpenAI-compatible client is created for that call. Use https://gateway.ertas.ai/v1 as the base URL and ERTAS_API_KEY from my environment as the key. Use the model id ertas-ai/medertas-27b.
Check for gateway-specific headers, model aliases, fallbacks and caching settings, and tell me which ones have no equivalent. Preserve streaming, tool calling, retries and app behaviour. Show a minimal diff and a rollback plan, never copy credentials into source or chat, and ask before sending a paid test request. Keep the old route available until I confirm the new one works.