Jan: Open, cross-platform AI app with useful proprietary models
Generative Artificial Intelligence (AI) today (2026) It is still mostly used by the majority of ordinary users (non-advanced IT professionals) as online tools provided by large global information technology companies. However, for advanced IT professionals (HelpDesk, SysAdmin, Devs, among others) or expert professionals from other business or commercial areas with high incomes, and therefore, with good hardware resources (computers with modern CPUs, robust GPUs, plenty of RAM and disk space) at their disposal, The use of local IAG tools is usually much more frequent and safer when usingAnd in these cases, as with the LM Studio and AnythingLLM applications, "Jan" is one of the best available options to consider..
And if you've never read, heard, or seen this Jan app before, it's worth noting from the outset that, besides being a open-source alternative to Claude and ChatGPT which works completely offline on your computer, also constitutes a suite of products with a family of its own models.
AnythingLLM App, an open and all-in-one AI app available for Linux
But, before we delve into this Interesting post about a new Generative Artificial Intelligence application called "Jan"which, like many others, is designed to harness the power of IAGs directly on our computers, enhancing our privacy, anonymity, security, and economy; we recommend exploring one previous related post, upon finishing reading it:
As an IT teacher and computer engineer, I often test, use, and recommend many software tools (professional, work-related, and educational). Some of these are Google Notebook LM and Gemini (both from Google). However, since those are online, I usually recommend other AI tools for offline use, such as AnythingLLM, Bavarder, Chatbox, Clippy Desktop Assistant, ComfyUI, GPT4All, Jan IA, Koboldcpp, LM Studio, Local AI, Ollama, Msty, Newelle, Pinokio, PyGPT, NetxChat, and Witsy, among others.
Jan: Open, cross-platform AI app with useful proprietary models
What is Jan?
Analyzing the Official website from "Jan"This open-source development and Generative Artificial Intelligence (GAI) tool can be briefly described as follows:
Jan is an open-source alternative to Claude and ChatGPT. It includes built-in base models and an application ecosystem that runs entirely on your hardware, keeping your data private and giving you complete control over your AI. Therefore, it's both a suite of products and a family of internally trained models. It's also designed to work wherever you are, so it's available for your desktop, browser, and even your mobile device, allowing you to use it however suits you best.
In addition, in his official section of GitHubSome of its many essential functions are summarized below, such as the following:
- Includes OpenAI compatible API: Available through a local server at localhost:1337 for other applications.
- Prioritize local, private, and safe use above all else.Therefore, everything is capable of running locally whenever the user decides.
- Availability of a Model Context Protocol (MCP)This facilitates integration with MCP for agent capabilities.
- Implementation of personalized assistantsProviding the ability to generate specialized AI assistants for various tasks.
- It allows the use of local AI modelsFacilitating the download and execution of LLM models from different providers from the HuggingFace website.
- It offers versatile integration with IAG in the cloud.: Allowing connection to GPT models via OpenAI, and Claude models via Anthropic; and others such as Mistral, Groq and MiniMax.
Installation Requirements (Hardware and Software)
On Windows, macOS and Linux
In summary, the most important points are that for Jan to function optimally, the following is required:
- Operating SystemsWindows 10/11, macOS 14.0 or higher and modern GNU/Linux distributions such as Debian/Ubuntu, Fedora/RHEL, Arch, openSUSE.
- CPU: Intel/AMD processors with 8 cores, with AVX2 instructions (Intel Haswell / +2013), AMD Excavator / +2015). On macOS, any Apple Silicon chip (M1, M2, M3, M4).
- RAM8 GB minimum, but 16 GB is recommended, since most powerful LLMs consume a large amount of operating system memory.
- GPUA dedicated graphics card with at least 6 GB of VRAM is required. However, 8 GB to 12 GB or more of VRAM is ideal.
Local LLM Engines
On Intel/AMD computers
On Intel/AMD computers, use llama.cppwhich is an engine that runs AI models locally. That is, it's the essential software that takes an AI model file (GGUF) and makes it run on your hardware, whether it's your CPU, graphics card, or Apple Silicon chips (M1 and later). It was...Originally created by Georgi Gerganov, while now maintained by ggml-org/llama.cppTherefore, pFor the community, llama.cpp is designed to run large language models efficiently on consumer hardware without the need for specialized AI accelerators or cloud connections.
About Apple Silicon computers (M1, M2, M3 and M4)
On Intel/AMD computers MLXThis is an inference engine for Apple Silicon (M1 and later) that uses "Metal GPU" acceleration for fast and efficient local inference, available in macOS 14 and later. Currently, it's an experimental implementation that will improve over time. However, it still has limitations: embedding is not available, so llama.cpp must be used for embedding/RAG workflows; reasoning is not yet integrated, so reasoning output is not displayed separately; and some newer model architectures are not typically loaded, so if a model fails to start, it should be tested on llama.cpp (GGUF).
Therefore, even for maximum model compatibility and the broadest feature set, llama.cpp should be used. In other words, it's only worth trying. MLX on Apple Silicon if you want Metal-optimized performance for a compatible model.
Usable models
Application In its "Hub" tab or section, it labels the LLM models available according to the compatibility detected with the computer hardware used, as follows:
- It does not fit/adapt: Probably exceeds available memory.
- It may run slowly.It will run, but it leaves little memory available.
- It adjusts/adaptsIt should work without problems on your device.
Installation processes
- Installation and use on LinuxIts deployment is extremely clean and ideal for Linux environments, as it is officially distributed through a standalone AppImage package and is available for the following architectures: x86_64However, it also offers DEB installers and Flatpak installations. Explore installation tutorial
- Installation and use on WindowsIts deployment is quick and easy through a classic installer file (.exe), available for x86_64 architectures. Explore installation tutorial
- Installation and use on macOSIts deployment is versatile and efficient, both through a classic installer file (.dmg) and through the Homebrew system, and available for both Apple Silicon (M1/M2/M3) and Intel processors. Explore installation tutorial
NotaCurrently, its current version is version Jan v0.8.3 dated the end of June 2026. And for more information you can access their Official documentation y Official Manual.
Screenshots of its Graphical User Interface (GUI)
1. Initial execution: Request to download the first model Jan
2. New Project Creation Screen
3. Chat Information Search Screen
4. LLM Model Search Screen on HuggingFace (Hub)
5. Settings Menu
A. Section: General
B. Section: Interface
C. Section: Assistants
D. Section: Attachments
E. Section: Local API Server
F. Section: HTTPS Proxy
G. Section: Shortcuts
H. Section: Hardware
Section I: Privacy
Integrations: MCP Servers
Integrations: Claude Cloude
Model Providers: Local – Llama.cpp
Main Screen: New Chat
Modern GGUF files, like the older and still-current GGML files, are file formats used to store AI models for inference. This is especially true in the context of Large Language Models (LLMs) such as the Generative Pre-trained Transformer (GPT). And of course, also for AI models known as Retrieval Augmented Generation (RAG) models. However, GGUF, being more modern than GGML, is clearer, more extensible, more versatile, and capable of incorporating new information without compromising compatibility with older models. It's worth noting, though, that the development of the GGUF format is based on its predecessor, GGML.
Summary
In summary, "Jan", Like LM Studio and AnythingLLM are great IAG tools that serve as ideal and essential solutions so that anyone (enthusiast, student, worker, and IT professional) can get started using LLM Models, RAG Systems, and local AI Agents—that is, without an internet connection. Above all, because they are cross-platform and have extensive or good multilingual support for Spanish. Furthermore, all three offer excellent integration with the website of hugging face, in order to offer a greater number of LLM models (downloadable GGUF files) available for automated use.
And of course, Jan and the other two favor the Privacy policy (our conversations never leave the computer or device used), a Caffordable (thanks to the possibility of not having to pay monthly subscription fees or API costs), an excellent vspeed response (which will depend on our hardware, and not our internet connection), and a larger cuntroll of our data and operations (since we can choose exactly which models to run and how they should behave). And as usual, we invite you to leave your opinion in the comments about this or other similar AI tools.
Lastly, remember visit our «homepage» in SpanishOr, in any other language (by simply adding 2 letters to the end of our current URL, for example: ar, de, en, fr, ja, pt and ru, among many others) to find out more current content.