← Back to the work
LOCAL AI / DATA SOVEREIGNTYWindows, macOS & Linux · $11.11

LLMLifeline.

Why should access to your tools depend on a subscription?

I wanted an AI workspace I could keep using on my own terms. LLMLifeline started on 1 September 2026 as a local chat app. Ten weeks and 37 releases later, it is a private AI system for a whole home: chat, image and video generation, documents, and an assistant that can build things, all running on hardware you own.

It runs local language models through llama.cpp and image models through ComfyUI. Models bigger than the graphics card load only the layers that fit, which took a 27-billion-parameter model from 0.6 to 3.3 tokens a second on my 11 GB card. Replies appear as they are written, conversations are organized into projects that remember, and uploaded documents are indexed so the assistant can quote them.

It fits the computer it is on. Every install comes with uncensored chat and image models in three sizes, large, medium and small, and on its first start LLMLifeline picks the biggest one that runs entirely on your graphics card. Put in a bigger card and it moves up a size on its own. Video works the same way: Wan video models in three sizes make short clips from a description, and the size your card holds downloads the first time you switch video on. A video can start from a picture or carry on from an earlier one, run up to five minutes, and comes with a scene-by-scene sheet that keeps later videos consistent. Each person in the home gets a profile that learns who they are and how they like answers, and it notices when someone else sits down at the keyboard.

What it ships with is a baseline, sized to the computer you have today, and you are not limited to it. As you upgrade, you can move to bigger and more capable models. A chat model downloaded as a folder of files becomes a model LLMLifeline runs: it converts the folder with llama.cpp’s own converter, on your computer, while you keep working. Any image or video model ComfyUI can run joins the lists through a workflow you build once in ComfyUI, including quantized video models, and LLMLifeline fills in your words, a fresh seed and the starting picture each time. The whole app also speaks 25 languages besides English, picked from your computer’s or phone’s own language, and your own model can translate it into any other.

With computer use turned on, the assistant works like a coding assistant on your own machine. It writes files in folders you share, runs commands and downloads pages, and shows its plan as a live checklist. You approve each new action and can revoke any permission. Whatever it is allowed to do, it can never upload your data. That rule is enforced in code rather than left to the model.

One strong computer can share its model with the rest of the home. Other computers connect with LLMLifeline; phones, tablets and Chromebooks just open a web page, with no app to install. Each device is approved with a matching six-digit code and remembered, and the sharing computer can rename, boot, block or forget it. Photos and documents sent from a phone go with the message and are kept in that device’s own files folder.

If you allow it, an approved phone can drive the sharing computer’s assistant from another room: give it a task, watch its checklist, stop it, or send it an update partway through. It works under the same permissions it has on the computer itself.

All of it stays on your network. Connections are encrypted and answer only local addresses, the models run on your own machine, and there is no account and no telemetry. LLMLifeline contacts my store only to activate a purchase and when you ask it to check for updates, and it sends only the key and an anonymous ID for the computer. It never updates itself or checks in on its own.

LLMLifeline is $11.11, paid once, for Windows, macOS (Apple Silicon) and Linux. A key activates up to three computers, and phones, tablets and Chromebooks connect free through the browser. Windows is where I build and use it every day; the macOS and Linux builds pass automated checks on every release, and if it doesn’t run on someone’s computer, I work on a fix and ship it as an update.

Inside the project

  • Uncensored chat and image models in three sizes, picked to fit your hardware
  • Video from a description or a picture, up to five minutes, with scene sheets
  • Projects that remember, with document retrieval
  • A profile for each person, that learns who they are
  • Computer use with your approval and a hard no-upload rule
  • One model shared across your home network (LAN)
  • Phones, tablets and Chromebooks through the browser
  • Remote control of the assistant from an approved device
  • No account, no telemetry, no subscription
  • Never updates on its own: you see what’s new, choose, and can switch back
  • Use your own models: model folders become .gguf files, and any ComfyUI workflow joins the lists
  • In 25 languages, and any other your own model translates
  • 37 releases in ten weeks, for Windows, macOS and Linux