Hob: Free open-source app puts local AI on any PC with zero setup

What it's about

Until now, running an AI language model on your own computer took some work. You had to find a model that suited your hardware, set up a runtime such as Ollama or llama.cpp, and pick a user interface to go with it. If you also wanted the AI to read or edit your own files, you had even more setting up to do. Hob, a free open-source app, aims to handle all of that for you. According to a detailed review by the Japanese tech site Gigazine, Hob checks your machine, chooses a suitable model, downloads it and then opens a chat window right away.

Hob comes from the company FrigadeHQ, and the source code is on GitHub. The developers call it a "free, open-source alternative to Claude and ChatGPT" that runs open-weight models such as Google's Gemma or OpenAI's gpt-oss locally. Hob runs on Windows, macOS and Linux, and the developers say it works privately and offline. You can find it at hob.chat.

How installation works

Gigazine installed Hob on a Windows 11 PC with an Nvidia graphics card. You click "Download for Windows" on the official site and run the downloaded file "Hob-windows-x64.exe", which installs itself. Hob then checks how much RAM you have and which graphics card is installed, and suggests a model to match. One click on "Download and start" is all it takes. The chat opens as soon as the download is done.

Gigazine says Hob supports 32 languages. Ask a question in Japanese and you get an answer in Japanese. The source doesn't say exactly which 32 languages are included, but the supported model families, such as Gemma, are multilingual.

More than chat: files, charts, projects

Hob does more than answer questions. The app can write and run its own Python code to process data. Gigazine attached a CSV file of sales figures and asked Hob to total revenue by month, turn the totals into a bar chart and explain the trend. Hob wrote a suitable script and ran it, and the chart and analysis appeared right in the chat.

The app can also read PDFs and images, for example to summarize documents or check what's in a screenshot. In the test, Hob correctly summarized a PDF in three bullet points. For images, it also downloads a small image-recognition model called Qwen3-VL. That's a vision-language model, an AI that handles images and text together.

The Projects feature is useful for real work. In the sidebar, you create a project via "New Project" and link it to a folder. The AI can then work with all the files in that folder together.

Hob also offers:

• web search,
• local voice input, which turns your speech into text on your own computer,
• chat history export as Markdown, a simply formatted text file,
• saved "skills", reusable sets of instructions,
• custom Python tools you can add yourself,
• connections to outside services via MCP (Model Context Protocol). MCP is a shared standard that lets an AI call functions in other apps and services. Depending on the service, you'll need to sign in and be online.

Offline and privacy-friendly

Privacy is the main reason to run AI locally. Because the language model runs on your own computer, Gigazine says no private data goes to anyone else's servers. Once the model is downloaded, you can disconnect from the internet and still chat and work with your saved files. Hob only needs internet access for outside services via MCP and for downloading extra Python packages.

If you want to be completely safe, turn off "Web access" under "Settings" → "Privacy". That blocks web search and any internet access from code the AI runs. It's a sensible step, because an AI that runs its own code could in theory send data out to the web.

Reactions and limits

On the tech forum Hacker News, opinions on Hob were mixed. One user wrote:

"Just having something that simply works makes getting started so much easier."

Others said the models were "not particularly smart". People also complained that there's no easy way to uninstall the model once it's installed, and it takes up around 18 GB. If you're short on disk space, keep that in mind before you install.

Gigazine points out a deliberate limitation. You can switch models and choose how hard the model should think. But there are no finer controls, such as temperature (how creative or random the answers are), context length (how much text the model can keep in view at once) or how much of the graphics card it uses. Hob is built for people who just want to use local AI, not for tinkerers who want to squeeze every last bit out of their hardware.

The bigger picture

Hob joins well-known programs such as LM Studio, Jan and Ollama with a graphical interface. Two things set it apart. Hob picks a model to match your hardware automatically, and features like Python execution, file projects and MCP come built in. That makes it feel more like the ChatGPT or Claude desktop apps than a tool for tech enthusiasts. For beginners, people handling sensitive data (from accounting, medicine or legal work, for example) and anyone who wants to use AI without a subscription or the cloud, it's an easy way in.

Keep your expectations realistic, though. Local models that run on ordinary PCs are much weaker at hard tasks than today's cloud models, and the Hacker News comments suggest the same. The sources don't say how well Hob runs on weaker machines without a dedicated graphics card, which model it picks there, or how reliably its own Python scripts handle complicated files. The privacy claims come from the developer and from Gigazine's test. But because the code is open on GitHub, anyone can check it.