Web UI

llms.py includes a modern, fast, and privacy-focused web interface that provides a ChatGPT-like experience for accessing all your local and remote LLMs.

Launch the server and Web UI with:

Open Workspace Explorer from the top-right button to inspect project files, or switch its tab to Git to review and commit changes. The left navigation's Decision Studio keeps recipes and typed decisions separate from ordinary chat. Settings includes ChatGPT subscription sign-in.

Overview

The UI is built with simplicity and privacy in mind:

• No Build Tools: Uses native ES modules
• No npm Dependencies: Pure JavaScript
• Offline First: Works entirely offline
• Privacy Focused: All data stored locally in SQLite
• Fast & Lightweight: Minimal footprint

Key Features

🎨 Modern Interface

Access all your LLMs through a clean, intuitive interface:

🌙 Dark Mode

Built-in dark mode support that respects your system preference or can be toggled manually:

🔍 Search History

Quickly find past conversations with built-in search:

Model Selector UI

With over 530 models from 24 providers now available, discovering and selecting the right model required a complete overhaul. Open it from the model chip at the top of the chat prompt. The Model Selector is a full-featured dialog offering:

• 🔍 Smart Search & Discovery - Instantly search across model names, IDs, and providers
• 🎯 Advanced Filtering - Filter by name, providers & input and output modalities
• 📊 Flexible Sorting - Sort by Knowledge Cutoff, Release Date, Last Updated & Context
• ⭐ Favorites System - Star model card to add/remove to favorites quick list
• 💎 Rich Model Cards - In depth model overview at a glance

📚 System Prompt Library

Access 200+ professional system prompts for various use cases:

Categories include:

• Technical assistance
• Creative writing
• Data analysis
• Code review
• Translation
• And many more...

See System Prompts docs for more details.

⚙️ Advanced Settings

Fine-tune AI requests with advanced options:

Available parameters:

• Temperature (0-2): Control response randomness
• Max Completion Tokens: Limit response length
• Seed: For deterministic outputs
• Top P (0-1): Nucleus sampling
• Frequency Penalty (-2.0 to 2.0): Reduce repetition
• Presence Penalty (-2.0 to 2.0): Encourage new topics
• Stop Sequences: Control where generation stops
• Reasoning Effort: For reasoning models
• Top Logprobs (0-20): Token probability analysis

🔌 Enable/Disable Providers

Manage which providers are active in real-time:

• Toggle providers on/off
• Providers are tried in order
• Automatic failover if one fails
• See which models are available per provider

Configuration

UI Configuration

Customize the UI via ~/.llms/ui.json:

{
  "prompts": [...],
  "defaultModel": "grok-4-fast",
  "theme": "auto"
}

Server Configuration

Start the server with custom options:

# Basic server
llms --serve 8000

# With verbose logging
llms --serve 8000 --verbose

# Custom root directory for UI files
llms --serve 8000 --root /path/to/ui

Storage

Data is stored on the server in SQLite databases under ~/.llms/user/<user>/ (~/.llms/user/default/ when you're not signed in):

• Conversations: all chat threads and messages, grouped by project
• Activity Logs: request history for analytics
• Projects: projects/projects.json and each project's folder

Your browser also keeps some UI preferences (such as theme and layout) and your unsent chat drafts, so those are per browser.

Privacy

The UI is designed with privacy as a priority:

• ✅ No external requests (except to configured LLM providers)
• ✅ No tracking or analytics
• ✅ No ads
• ✅ No sign-ups required
• ✅ All data stored locally
• ✅ Works entirely offline (Use local LLMs, Browse History)
• ✅ Open source

Next Steps