Cloud AI subscriptions add up, and not every business wants its data leaving the office. Jan is a free, open-source ChatGPT alternative that runs 100% offline on your own computer — with a polished interface, downloadable models, and even a local API server for your other apps. This tutorial shows you how to install Jan, run your first offline model, and use it across your small business.
What It Is
Jan is an open-source desktop application for running large language models locally on Windows, macOS, and Linux. It looks and feels like a modern chat app, but the models run on your hardware: download a model once from the in-app Hub (sourced from Hugging Face), then chat with no internet connection and no per-message fees. As checked on October 18, 2026, it has 44,986 stars on GitHub (github.com/janhq/jan, redirecting from menloresearch/jan) and is licensed under Apache 2.0.
Features from the README: local AI models (Llama, Gemma, Qwen, GPT-oss, and others from Hugging Face), cloud integration (optionally connect OpenAI, Anthropic, Mistral, Groq, MiniMax, and more when you want flagship quality), custom assistants with tailored instructions, an OpenAI-compatible local API server at localhost:1337 so other applications can use your local models, Model Context Protocol (MCP) integration for agentic capabilities, and a privacy-first design where everything can run locally.
Who It’s For / SMB Use Case
Jan fits small businesses that want everyday AI help without subscriptions or data leaving the building. Practical uses include:
- Private drafting: Write emails, proposals, and job posts on your own machine so client and staff details never go to a cloud provider.
- Offline productivity: Keep an AI assistant available on laptops in the field, on planes, or anywhere with unreliable internet.
- Cost control: Replace per-seat chatbot subscriptions for routine writing and summarization tasks with a free local app.
- Developer building block: Point internal tools at Jan’s local API server (
localhost:1337) to add AI features without cloud API bills.
Jan is not a replacement for the most powerful cloud models on hard reasoning tasks — a laptop-sized model won’t match a flagship cloud model. It’s also a desktop app, not a team server: each person runs their own copy. For shared, multi-user setups, look at server-based tools instead.
Installation
The easiest path is downloading the installer for your OS — no command line needed. The README lists these downloads (also on jan.ai and GitHub Releases):
- Windows:
jan.exe(also available on the Microsoft Store) - macOS:
jan.dmg - Linux:
jan.deb,jan.AppImage, or Flathub
System requirements (from the README)
For a decent experience: macOS 13.6 or later (8 GB RAM for 3B models, 16 GB for 7B, 32 GB for 13B); Windows 10 or later with GPU support (NVIDIA/AMD/Intel Arc); Linux — most distributions work, with GPU acceleration available. Bigger models need more RAM and ideally a dedicated GPU; start with a small 3B model on modest hardware. If anything goes wrong, the README points to the troubleshooting docs on jan.ai and the community Discord.
Build from source (optional)
Developers can build it with git clone https://github.com/janhq/jan, then cd jan and make dev (requires Node.js ≥ 20, Yarn ≥ 4.5.3, Make, and Rust for Tauri). Most businesses should just use the installers above.
How to Use Jan
Step 1: Download your first model
Open Jan and go to the Hub. Browse the model catalog (Llama, Gemma, Qwen, and others from Hugging Face) and download one that fits your hardware — start with a small 3B model if you’re unsure. Downloads can be several gigabytes, so use a decent connection the first time; after that, everything works offline.
Step 2: Start chatting
Select the downloaded model and start a new thread. Ask it to draft, summarize, or explain — just like any chat app. Because the model runs locally, responses are private and free, though speed depends on your hardware. Try the sample prompts in the app to get a feel for what a small model handles well.
Step 3: Create a custom assistant
Jan lets you create assistants with custom instructions — for example, a “Proposal Writer” that always uses your company’s tone and structure, or a “Meeting Notes Summarizer” that outputs action items in your format. Define the system prompt once, then reuse the assistant for consistent results across the team (share the prompt text with colleagues so everyone uses the same one).
Step 4: Connect cloud models (optional)
When a task needs more brainpower, add an API key for OpenAI, Anthropic, Mistral, or Groq in Jan’s settings and switch models per conversation. This gives you a single app for both private local work and heavy-duty cloud work — use local for sensitive drafts, cloud for complex analysis.
Step 5: Use the local API server
Enable Jan’s local API server to expose an OpenAI-compatible API at localhost:1337. Other applications on the same machine can then call your local models with standard OpenAI-format requests — handy for developers adding AI to internal tools without cloud costs. The README’s API reference on jan.ai documents the endpoints.
Pricing / Limitations
Jan is free and open source under Apache 2.0 — no subscriptions, no seat fees, no usage meters in the app. Your costs are the computer it runs on and, optionally, cloud API fees if you connect providers like OpenAI or Anthropic for harder tasks.
Limitations: local model quality trails flagship cloud models, especially for complex reasoning, coding, and long documents — match the model to the task. Large models need serious RAM and preferably a GPU; a basic office laptop should stick to small models. Jan is per-device software, not a centralized team platform, so there’s no shared admin console or usage analytics. And while offline operation is great for privacy, you miss out on the always-updating knowledge of cloud models.
Bottom Line for SMB
Jan is the easiest way to put a private AI assistant on every company laptop: download, pick a model, and start working — no accounts, no fees, no data leaving the building. Use it for everyday drafting and summarizing, and reach for cloud models inside the same app when you need more power. If you want to run models from the terminal or behind other tools too, read our Ollama tutorial, and for another offline option, our GPT4All tutorial.
Frequently Asked Questions
Does Jan really work without the internet?
Yes, after the one-time model download. The app, the model, and your chats all run locally — you can disconnect entirely and keep working. Cloud-connected features (like optional OpenAI/Anthropic models) naturally need internet.
How much RAM do I need?
Per the README: 8 GB for 3B-parameter models, 16 GB for 7B models, and 32 GB for 13B models on macOS, with similar guidance for Windows and Linux. When in doubt, start with the smallest model and upgrade if answers feel weak.
Can my whole team share one Jan setup?
Not centrally — Jan is a desktop app, so each person installs their own copy. For a shared, multi-user AI assistant over your documents, consider a server-based tool; for consistent results across the team, share the same assistant system prompts and model choices with everyone.
