Jan is a free, open-source ChatGPT alternative from Menlo Research that runs local or online models on Windows, macOS and Linux, with an OpenAI-compatible local API server.
Jan is an open-source AI chat application built by Menlo Research. Its homepage describes it as a personal intelligence you can run with any model, using local or online models, with memory that carries context across chats. The page labels it free and open source and shows about 44.9K GitHub stars. The GitHub repository states an Apache 2.0 licence.
The homepage lists Jan Desktop and Jan Agent as products but offers only a Mac download on the page we read. The GitHub README lists installers for Windows, macOS (universal build) and Linux, including .deb and AppImage packages and an Arm64 guide. Jan can also run a local server at localhost:1337 that is OpenAI-compatible, so other applications can send requests to the model on your machine.
No paid plans are listed on the homepage. Jan is presented as free software, so the cost is the hardware needed to run models and any fees you pay if you connect to an online model provider with your own key. The pages we read do not describe a hosted subscription, and they do not name any compliance certifications or data-residency options.
For a buyer choosing among local-model apps, Jan sits between a command-line runner and a managed desktop product: it gives a chat window and settings screen, while keeping the code open for inspection. Model size remains the practical limit, because a model must fit in your computer's memory. If you do not want to manage downloads or hardware, a hosted chatbot will be simpler to use, at the price of sending your prompts to a vendor's servers.
Yes. The Jan homepage describes it as free and open source, and no paid tiers are listed. You still need hardware able to run the model you choose.
The GitHub repository states Apache 2.0. The homepage itself does not name a license, only that Jan is built in public and open source.
The GitHub README lists Windows, macOS and Linux installers. The homepage we read showed only a Mac download, so use the GitHub releases for other systems.
Website: https://www.jan.ai/
Company: Menlo Research
Free Apache 2.0 chat app for running local or online models on Windows, macOS and Linux, with an OpenAI-compatible local server. No paid plan; model size is limited by your computer's memory.
| Pricing model | Open source |
|---|---|
| Free tier | Yes · Free and open source; no paid tiers listed |
| Starting price | Not verified |
| Data residency | Not verified |
| Compliance (as stated by vendor) | Not verified |
| Integrations | Not verified |
| Public API | Yes |
| Self-hostable | Yes |
| Official site | Online (auto-checked 2026-10-12) |
| Can a general AI assistant replace it? | Hosted chatbots are easier and stronger on small hardware. Jan fits users who want local, open-source control; a command-line runner suits scripting better. |
| Last verified | 2026-10-10 by editor |
XX.AI offers a simple and cost-effective solution for unlimited access to GPT-4, Claude 3, and DALL-E 3. With essential shortcuts like Shift + A for instant activation, you can harness the power of AI in any writing or content creation task.
Google for startup ideas
AnythingLLM is a free, MIT-licensed AI assistant that runs on your own computer, with document chat, agent skills, background jobs and an on-device meeting assistant. It needs no account or API key, and it can be self-hosted for teams.
Alexa+ is Amazon's generative-AI version of Alexa for Echo devices, the Alexa app and Alexa.com in the US. It is included with Prime, a free text chat tier is open to everyone, and it costs $19.99 a month without Prime.
ConverseNow is a voice AI for restaurants that handles phone ordering with upselling, multilingual support and analytics. It is sold by consultation; no prices or free plan are published.
Hostie is a virtual concierge for restaurants that answers calls, texts and emails, books reservations and takes takeout orders. Plans are Essential, Premium and Hospitality Plus; no prices are published.
Need the outcome rather than the software?