New chat
Each API key is stored only on this device, never on the server, so the same key stays with the provider you entered it under. Add as many OpenAI-compatible endpoints as you like — switching between them never overwrites another.
Star models you use with the ☆ in the Main model list; they appear in the selector under the chat, each labelled with its provider. Fallbacks are tried in order when the main model fails, so a fallback may come from a different provider — picking one switches to it.
Your model's identity — injected at message 1 of every chat. This is the \"constitution\" that shapes all responses.
Injected at message 1 of every chat (after soul). Task-level instructions that apply to all conversations.
Facts about you — injected at message 1 of every chat (unless a chat opts out). Add, edit, delete facts below. Reorder by drag if needed.
Save reusable prompt templates. Use variables like {{role}} that prompt for values when applied. One-click to set as global or use in current chat.
Message 1 always injects. The number above is the second trigger.
Applies to the open chat only. Doesn't change your account settings.
When checked, this chat starts fresh without global memory, soul, or global system prompt.
Documents without a text layer (images, scans, Word, Excel) are sent to the parser model. Leave it on Same as main model and your main model reads them. Set a cheaper vision model to control scan costs.
PDFs with a text layer are extracted locally (free). Images, scans, Word and Excel go to the parser model.
Cleanup runs when you sign in. Monthly cost totals are kept even after chats are deleted.
Your key stays on this device. It is never sent to the server, and the server never relays your traffic. Set a spending limit on the key in your OpenRouter dashboard — it's the only guard available.