xllm

xllm

…

Admin

admin

Each user signs in with their own key; their conversations are kept on the server and are the same on every device. Every change here is written to the server's log with your name.

What xllm can do here

on this server, now

Guardrails

Checks that run on the server beside every reply. They read the model's own attention state, not the text, and add almost no time. An enforced one applies to every request. An advisory one is yours to switch off or keep only as a warning for the messages you send; it is still scored and logged on the server.

Sign in

Paste your key, or open the invite link your admin sent you. (A server with one shared key: paste that key.)

↑↓ move↵ openesc close

New conversation

Enter send · Shift+Enter newline · Esc stop

Appearance

Text size

Notify when a reply finishes while this tab is in the background

Sampling

Temperature0.70

Top-p0.90

Top-k

Max tokens

Max tokens 0 = until the reply is done. A capped reply ends at a sentence, and can be continued.

Think budget

Conversations

Conversations live in this browser. Export to keep or move them; import a file exported elsewhere.

Saved prompts

Type / in the message box to pick one. A prompt can belong to a workspace and show only there.

System prompt

Settings are saved in this browser.

Tools

Server

Live from the server's /health. Counts are totals since the server started and reset on restart; tokens/sec is the lifetime average. The same numbers feed the daily report.