llms.py
The self-hosted, OpenAI-compatible AI gateway for text, image & audio - a lightweight CLI, server and OSS Open WebUI alternative for Local and Cloud LLMs
August 24, 2026 - v4 Released! →
One tool, every modality
Generate Anything
From a single self-hosted gateway - chat with 530+ models, create images, and synthesize speech across 24 providers
A fast, private, ChatGPT-like web UI over every local & cloud LLM - with tools, MCP, skills, agents and 200+ system prompts
Create images via Gemini, OpenAI, OpenRouter, Chutes, Z.ai & Nvidia - right inside the chat and gallery workflow
High-quality text-to-speech with Gemini & OpenRouter models, plus voice-to-text input via the microphone
Create → Publish → Showcase
Your Creations, in the Public Gallery
Anything you generate in llms.py can be published with one click - landing in the live showcase at ai.llmspy.org, browsable by everyone with categories, tags, ratings and infinite scroll.
- Browse generated images & audio by category and tag
- Play full Projects built by AI - games and apps, ready to run
- Free & anonymous publisher account - no email required
Playable & self-contained
Projects Built by AI
Complete, self-contained apps & games - generated with llms.py and published for anyone to open and play
Quick Install
pip install llms-py💬 Simple Chat
Ask questions directly from the command line
Run Server
llms --serve 8000Beautiful Themes
Personalize your experience with a selection ofhandcrafted themes
💬 ChatGPT-like Interface
Modern, fast, and privacy-focused web UI for all your local and remote LLMs
What's New
Latest release focused on extensibility, expanded provider support, and enhanced user experience
Powered by models.dev integration with automatic daily updates
Add features, providers, and customize the UI with flexible plugins
RAG workflows with document stores, categories, and contextual chat
Username/Password authentication with Admin UI and CLI user management
Specialized AI agents with custom prompts, tools, themes, and Planner→Coder workflows
Secure workspace isolation that restricts AI agent filesystem access to designated project folders
Design pixel-identical PDFs using Typst templates, real-time live preview, schema-driven forms, and AI assistance
First-class Python function calling for LLM interactions with your environment
Built-in support for provider hosted Anthropic & OpenRouter server tools like web search
Connect to Model Context Protocol servers for extended tool capabilities
Extend AI capabilities with specialized knowledge, workflows, and tools
AI-powered browser automation with live preview, script editor, and element inspector
Customize the look and feel with built-in themes or create your own
Beautiful UIs to evaluate math and execute Python, JS, TS & C# code
Browse and manage all your generated images and audio in one place
Voice-to-text transcription via microphone button or ALT+D keyboard shortcut
Built-in support for Gemini, OpenAI, OpenRouter, Chutes, Z.ai and Nvidia
TTS support for Gemini 2.5 Flash/Pro Preview models
Beautiful LaTeX math typesetting for equations and formulas
Model Selector
Smart search, advanced filtering, sorting, and favorites over 530 models from 24 providers
Learn more →
Credentials Auth
Built-in Username/Password authentication with a Sign In page, Admin Web UI and CLI for user managing accounts, roles, and account locking
Learn more →
Agent Profiles & Projects
Configure specialized AI agents with custom prompts, tools, and themes - and scope their filesystem access to secure project workspaces
PDF Studio
Design pixel-identical PDFs using Typst templates, real-time live preview, schema-driven forms, typed code generation, and AI assistance
Turn your content into a trusted AI Assistant
Ingest files, repositories, and websites into managed Gemini File Stores. Curate the exact knowledge each answer can use, verify every citation, then publish a beautiful support Assistant anywhere with one script tag.
Curate every source
Upload files and ZIPs, sync folders, or crawl websites into inspectable Markdown before indexing.
Retrieve precisely
Scope Gemini by category, type, status, locale, product, version, tags, or a single document.
Publish with confidence
Ship a branded, citation-backed Website Assistant and review real customer conversations.
01 · Ingest & refine
A clean knowledge pipeline, not a black box
Preview folder changes before committing, save repeatable imports, and monitor every upload. The web crawler stages pages as Markdown so you can inspect and transform extracted content before Gemini ever sees it.
- Files, ZIP archives, folders, and websites
- Diff previews and recurring import.json rules
- Resumable background uploads with live progress
02 · Ask with evidence
The right answer from exactly the right documents
Browse by category, compose metadata filters, and carry that precise scope into a grounded Gemini chat. Every response can surface the retrieved evidence and link readers back to the original source.
Categories appear as paths and additional filters remain inspectable, so users always know which knowledge shaped the answer.
03 · Design & publish
Your own support Assistant, beautifully on-brand
Choose a behavior template, system prompt, Gemini model, document scope, opening behavior, and suggested questions. Then style every surface and launcher color before publishing the self-contained Shadow DOM widget with one script tag.
Ready to ground Gemini in your own knowledge?
Install the extension, create a File Store, and ask your first citation-backed question.
Server Tools
Built-in support for provider-hosted OpenRouter & Anthropic server tools like web search, web fetch & code execution
Learn more →
Agent Browser
An integrated workspace for building and running automated browser scripts with AI assistance, live previews, and an interactive element inspector
Learn more →
Image & Audio Generation
Seamless media generation through UI and CLI
Runtime Provider Management
Enable or disable providers on the fly without configuration changes
Learn more →
Ready to Get Started?
Install llms.py and start chatting with 530+ AI models in minutes