Jan

A free, Apache 2.0 desktop app that works like a private ChatGPT running offline. No paid tier exists — the pricing page 404s.

Price
Free
Free tier
Free tier: yes
Updated
Updated Sep 6, 2026
Website
Website

Good for / not for

  • Privacy-focused individuals who want a ChatGPT-shaped desktop app running on their own machine.
  • Anyone who wants MCP tools in the same window: Jan connects to Gmail, Notion and Slack.
  • People who do not want a plan to outgrow — Jan has no paid tier, no caps and no cloud credits.
  • Anyone who needs a clean open licence: Apache 2.0, at 44,366 GitHub stars, last pushed on 6 September 2026.
  • Non-technical users wanting zero setup. A reviewer found it not plug-and-play, and it wants 8GB or more of free RAM.
  • Anyone who needs mobile or web access. Jan is desktop-only, with no mobile app and no web interface.
  • Switching models constantly: one reviewer measured 30 to 60 seconds of lag on each switch.

What it actually does

Jan is a desktop app shaped like ChatGPT that runs on your machine instead of someone else's. You download a model, it answers locally, and nothing has to leave the computer. Jan's site claims more than 6.5 million downloads, and the repository sits at 44,366 GitHub stars with a push on 6 September 2026. The licence is Apache 2.0, confirmed against the LICENSE file itself rather than GitHub's label. Beyond chat, Jan connects MCP tools, so the same window can reach Gmail, Notion or Slack; v0.8.4, on 23 July 2026, moved credentials into a backend-managed store, added an OpenAI-compatible translating gateway and native web search, and patched a handlebars remote-code-execution advisory. What makes Jan unusual here is the pricing, in that there is none. No caps on the free tier, no upgrade path, no cloud credits: jan.ai/pricing returns a 404 because there is nothing to put on it. Your costs are the hardware and your time. The constraints are that it is desktop-only, and that a reviewer found it not plug-and-play for non-technical users — it wants 8GB or more of free RAM, and lagged 30 to 60 seconds when switching models.

What it costs

Download Jan and run a model on your own computer.

Nothing, with no caps and nothing to upgrade to. Jan has no paid tier at all, and jan.ai/pricing returns a 404 because there is nothing to price.

From the vendor's docs

Pricing and free-plan limits

Free tier: Free with no caps at all, and no paid tier to move up to — the pricing page 404s because there is nothing to price.

PlanPriceWhat you get
FreeFree
  • The full desktop app, Apache 2.0 licensed
  • Models run entirely on your own computer
  • MCP tool connections including Gmail, Notion and Slack
  • Native web search and an OpenAI-compatible translating gateway, added in v0.8.4

There are no paid tiers and no caps. jan.ai/pricing returns a 404 because there is nothing to price.

Prices checked on Sep 6, 2026 against the vendor's pricing page. Plans change often — confirm before you pay.

Three honest limitations

  1. A reviewer found it not plug-and-play for non-technical users: it wants 8GB or more of free RAM, and model switching lagged 30 to 60 seconds in their testing.
  2. Desktop only. There is no mobile app and no web interface, so it does not follow you off the machine it is installed on.
  3. GitHub's licence detector labels the repository "Other" rather than Apache-2.0, because Jan's LICENSE file is a short pointer to the Apache text rather than the text itself. Reading the file settles it — it is Apache 2.0 — but anywhere the GitHub label is what gets checked, it will not say so.

Alternatives

  • LM StudioChoose LM Studio instead if you want a graphical local app and can live with proprietary software and a paid cloud layer above the free tier.
  • OllamaChoose Ollama instead if you are happy at a terminal and want an OpenAI-compatible local API rather than a desktop window.
  • Open WebUIChoose Open WebUI instead if several people need to share one instance through a browser rather than each installing a desktop app.

Getting started in 5 minutes

  1. Download Jan from jan.ai. There is no tier to choose, because there is only one and it is free.
  2. Have 8GB or more of free RAM available — that is what a reviewer records Jan as wanting.
  3. Pick one model and stay on it while you evaluate; the same reviewer measured 30 to 60 seconds of lag on each switch.
  4. Connect an MCP tool once the chat works. Gmail, Notion and Slack are the ones Jan names.
  5. Plan around the desktop. There is no mobile app or web interface, so anything you need on a phone needs a different tool.

Jan questions

Is Jan really free?
Yes, with no caps, and there is nothing to upgrade to. Jan has no paid tier at all — jan.ai/pricing returns a 404 because there is nothing to put on it. Your only cost is the machine you run it on.
What licence is Jan under?
Apache 2.0. GitHub shows the repository as "Other" because the LICENSE file is a short pointer to the Apache text rather than the text itself, but reading the file resolves it to Apache License 2.0.
Does Jan work offline?
That is the design: models run on your own computer, and Jan describes itself as running entirely offline. The MCP connections are the exception by definition — reaching Gmail, Notion or Slack means reaching the network.
Jan or LM Studio?
Both are desktop apps for local models; the difference is the licence and the business model. Jan is Apache 2.0 with no paid tier of any kind. LM Studio is proprietary, free for personal and internal business use, and since 2026 sells cloud credits above its free tier.
How big is the project?
Jan's site claims more than 6.5 million downloads, and the repository is at 44,366 GitHub stars, last pushed on 6 September 2026. The GitHub organisation was renamed from janhq to menloresearch — a branding shift, and nothing found suggests more than that.

Sources

Research-compiled, not hands-on tested. Compiled from official docs, the vendor's pricing page and community reviews as of Sep 6, 2026. We haven't tested it hands-on yet.

Ollama

from $20/month

The default way to run models locally, driven from a terminal. Free and unlimited on your own hardware; the 2026 cloud tier is separate.

The graphical answer to Ollama: download and chat with local models by clicking. Free for personal and internal business use, and proprietary.

A self-hosted, multi-user ChatGPT-style web interface for models you supply. Free to run as-is; the 2025 licence protects its branding.