Metrics
Add custom model
Omni can make mistakes. Check important info.
Show intermediate reasoning when available.
Omni can make mistakes. Check important info.
URL of your Cloudflare Worker or compatible proxy.
Select your AI provider to populate available models.
Stored locally in your browser. Never sent anywhere except your endpoint.
OpenAI-compatible endpoint URL.
System prompt prepended to every message. Leave empty for none.
Which service runs the code in agent mode. Both bill per second of sandbox lifetime, so sandboxes are killed as soon as a run ends. Keys are stored per provider, so you can switch without re-pasting.
Enables agent mode, which lets the model run code in a sandboxed Linux VM. Your key, your account — stored in this browser and sent only to your own Worker, which keeps none of it.
How many tool calls one reply may chain before the model is forced to answer. Each step re-sends the whole conversation, so higher costs more.
Higher is more creative; lower is more focused.
Cap on tokens generated per response.
Show intermediate reasoning when available.
Show intermediate reasoning when available.
Auto lets the model or a built-in heuristic decide when to search.
Tavily searches this month. Resets monthly.