Running AI on your own computer sounds simple until you try to choose a model.
A person may have 16 GB of RAM, an NVIDIA GPU, or an Apple Mac, but model pages often do not give a direct answer to the question that matters:
Will this model run on my computer, and what should I use to run it?
I built MamiLens to make that decision easier.
What MamiLens asks
The tool starts with information that most people can find:
- Operating system
- System memory
- Graphics card and GPU memory
- The job they want the model to do
- The amount of text they want to process
- Their preferred balance between speed and capability
It then shows three model options instead of sending the user into a catalogue with dozens of cards.
The three options have different purposes:
- A balanced recommendation
- A lighter model for a faster start
- A more capable model when the computer has enough memory
Each recommendation explains why it was selected. The user can compare model size, memory needs, expected fit and available evidence before confirming a model.
It also recommends the software
Choosing a model is only part of the problem. The user still needs software such as Ollama, llama.cpp, LM Studio or another compatible engine.
MamiLens connects the confirmed model to the software step. It considers the operating system, model format, technical comfort, intended use and number of people who may use the system at the same time.
The user can accept the recommended engine or choose another compatible option.
What the tool does not promise
Local AI performance depends on more than RAM.
GPU support, memory bandwidth, quantization, context length, runtime settings and concurrent users can all change the result. MamiLens therefore uses conservative memory planning and labels recommendations that still need a real test.
It does not claim an exact speed when there is no benchmark from the same device.
The final installation plan helps the user download the right software, use the listed model artifact and report whether the setup worked. Those reports can improve future recommendations.
Why the name changed
The first name was LocalLens. Several established products already used that name, so I renamed the product MamiLens.
The new name comes from "Mami Kein," meaning "big mother" in my local language. It represents guidance, care and responsibility. The lens represents looking through many model choices to find a practical answer.
I need technical feedback
MamiLens is public and free:
https://locallens.deviceterra.com
If you run local models, I would value feedback on three areas:
- Does the memory estimate make sense for your hardware?
- Does the engine recommendation match how you would deploy the model?
- Is the explanation clear enough for a beginner without hiding important technical facts?
Please include your operating system, RAM, GPU, selected model and engine when reporting a problem. That will make the feedback useful.
Top comments (0)