ChatGPT
AI conversational assistant for answering questions, writing, and coding help
Jan
Run ChatGPT-style AI models locally with complete privacy
Desktop application for running open-source LLMs locally or connecting to cloud providers. Full offline capability with OpenAI-compatible API at localhost:1337.
Overview
Privacy-First AI Desktop Application
Jan is a cross-platform desktop application that lets you download and run large language models entirely on your own hardware. Built for users who want ChatGPT-like capabilities without sending data to external servers, Jan supports popular models from HuggingFace including Llama, Gemma, Qwen, and others.
Flexible Deployment Options
While Jan excels at local inference, it also integrates with cloud providers like OpenAI, Anthropic, Mistral, and Groq when you need them. The application exposes an OpenAI-compatible API server on localhost:1337, allowing other tools to leverage your local models. Model Context Protocol (MCP) integration enables agentic workflows and extended capabilities.
Built for Control and Customization
Create specialized AI assistants tailored to specific tasks, switch between local and cloud models seamlessly, and maintain complete control over your AI infrastructure. Built with Tauri and TypeScript, Jan runs on Windows 10+, macOS 13.6+, and most Linux distributions, with GPU acceleration support for NVIDIA, AMD, and Intel Arc hardware. System requirements scale with model size—8GB RAM handles 3B parameter models while 32GB enables 13B+ models.
Highlights
Download and run LLMs from HuggingFace completely offline
OpenAI-compatible API server on localhost:1337 for integration
Hybrid deployment: local models or cloud provider connections
Model Context Protocol support for agentic capabilities
Pros
- Complete data privacy with 100% offline operation
- Cross-platform support for Windows, macOS, and Linux
- OpenAI-compatible API enables drop-in replacement workflows
- Flexible cloud integration when local resources are insufficient
Considerations
- Requires significant RAM for larger models (32GB for 13B parameters)
- GPU acceleration limited to specific hardware vendors
- Local inference speed depends entirely on your hardware
- Steeper learning curve than cloud-only solutions
Managed products teams compare with
When teams consider Jan, these hosted platforms usually appear on the same shortlist.
Looking for a hosted option? These are the services engineering teams benchmark against before choosing open source.
Fit guide
Great for
- Privacy-conscious users handling sensitive data locally
- Developers needing OpenAI-compatible API without external dependencies
- Organizations with strict data residency requirements
- Users with capable hardware wanting cost-effective LLM access
Not ideal when
- Users with limited RAM or older hardware configurations
- Teams requiring enterprise-grade model management features
- Users expecting cloud-level inference speed on modest hardware
- Organizations needing centralized multi-user deployment
How teams use it
Offline Document Analysis
Process confidential documents with local LLMs, ensuring sensitive information never leaves your network
Development Environment Integration
Point existing tools to localhost:1337 for AI features without modifying code or paying API fees
Custom AI Assistants
Build specialized assistants for legal review, code generation, or domain-specific tasks with tailored prompts
Hybrid Cloud-Local Workflows
Use local models for routine tasks and switch to cloud providers for complex queries requiring larger models
Tech snapshot
TypeScript82%
Rust13%
Python2%
JavaScript1%
Shell1%
PowerShell1%
Frequently asked questions
What models can I run with Jan?
Jan supports LLMs from HuggingFace including Llama, Gemma, Qwen, and GPT-based open-source models. You can also connect to OpenAI, Anthropic Claude, Mistral, and Groq cloud models.
How much RAM do I need?
8GB RAM handles 3B parameter models, 16GB works for 7B models, and 32GB is recommended for 13B+ parameter models. Requirements scale with model size.
Does Jan work completely offline?
Yes. When using local models, Jan runs entirely offline with no internet connection required. Cloud integrations are optional.
What is the OpenAI-compatible API?
Jan runs a local server at localhost:1337 that mimics OpenAI's API format, letting you point existing applications to your local models without code changes.
Which operating systems are supported?
Windows 10+, macOS 13.6+, and most Linux distributions. GPU acceleration is available for NVIDIA, AMD, and Intel Arc graphics cards.
Project at a glance
Active- Stars
- 40,117
- Watchers
- 40,117
- Forks
- 2,491
Repo age2 years old
Last commityesterday
Self-hostingSupported
Primary languageTypeScript
Last synced yesterday