SplitLLM

Registration is currently disabled on this server. Ask the operator to enable it, or use an existing account.

Create account → ← Back to login

Profile & memory

This text is injected so the model can match your tone and remember interests (servers, school, playfulness, etc.). With “auto-extend” on, the server runs a short merge right after an answer (a few seconds apart at most). That answer was still built with the old profile — new facts from chat or from Save apply to your next message. You’ll get a small on-screen note when auto-merge actually saved to disk.

Log in to use profile memory.

Personality

Define your own custom personality instructions for replies. Click outside this panel to close.

Knowledge Modules

Available knowledge modules:

Admin

Only users listed in admins.txt on the server can use this panel. Use your SplitLLM admin username and password.

New registration key (copy now — full key is only shown once):

Registration keys

Users

Features

Global limits

3 concurrent
8192 tokens
On
Off

Admin debugger

Learn Mode

Teach the AI new topics by submitting learning tasks. The system will crawl multiple sources and automatically create knowledge modules. You must be signed in as a server admin; the browser sends a secure HttpOnly session cookie (same as after login).

Learning Task

5

Learning Task Queue

SSH Endpoints

Add Endpoint

Public key — add this to ~/.ssh/authorized_keys on the server:

Change Password

Settings

Account

Profile

These open their own panels; changes apply to your next message.

Knowledge

Visible only to server admins. Actions use your signed-in session (no extra password).

Connections

Admin

Registration keys, user list, toggles, and global limits.

Repository
Projects
Chat history

Tied to your signed-in account on this server. With Save on, threads auto-update after each reply (and can be deleted below). Expire after about 7 days from the last update. Turning Save off removes all stored history for your account.

Hello

Automation
SSH auto Run commands without a confirm step
Extended thinking Deeper reasoning for complex problems
Reasoning mode Show model thinking, code review & analysis
Quick Routing [Beta] Skip LLM routing, use only keyword matching (faster)
Response effort

Low: faster, shorter. Medium: balanced. High: more detail & reasoning.

Max output tokens
2048

Higher = longer answers, slower, more VRAM/RAM pressure (especially with multiple users).

Admin: Can go up to 64k tokens

Web search results
10

1–5: fast, minimal info. 10: Recommended. 15–20: comprehensive, slower.

Model profile
0%
0 / 10 files
    Drag & Drop
    or
    Download ZIP
    Instructions
    0 / 4000
    File
    
                        
                    

    Quick Start

    New Project

    Rename Project