gpulm chat

This chat has no accounts. It runs on your GPU load manager key, and it can do exactly what that key allows — no more, and it will tell you when it is less.

Your key remember on this device

The key is kept in this browser and sent with each request so the server can pass it to gpulm. Nothing about this chat — not a message, not a picture, not the key — is stored on the server.

…

model reasoning can draw with temp max tokens

Settings

These are this chat's settings. The global ones are what every new chat opens with.

System prompt for this chat

Images

The can draw checkbox under the input gives the model the tool. These are the settings it draws with.

width height steps cfg

negative prompt

0 for steps or cfg means "use the model's own default from gpulm".

Move this history to another device

Everything is in this browser and nowhere else, so carrying it means carrying a file. It is encrypted here, with a passphrase that never leaves this page — whoever carries the file cannot read it.

Passphrase

replace instead of merge

Models on the GPU

Reasoning markers

How each model opens and closes a reasoning block. An edited block is handed back to the model in exactly these tokens.

New version is out