ModelFit

Find local AI models that fit your computer's memory budget.

ModelFit detects your hardware and presents an interactive catalog of language, image, video and audio models. Adjust memory and context settings to compare estimated fit before downloading large model files.

Interface	Runtime	Accounts required	License
Local browser app	Python standard library	None	MIT

What it does

• Detects CPU, system RAM, GPU and VRAM where supported.
• Lets you enter memory values manually when hardware detection is incomplete.
• Filters models by category and other catalog properties.
• Recalculates fit estimates as you change the planned context window and available memory.
• Runs on a loopback-only server; the bundled catalog does not require a cloud service.

ModelFit estimates capacity. It does not download, run or benchmark the models it lists.

Requirements

• Python 3.10 or newer.
• A modern browser.
• Git, or Code → Download ZIP.

No Python packages are required for the app itself. Hardware detection includes platform-specific code; the release smoke test was performed on Windows. Use manual memory inputs where detection is unavailable.

Install and run

git clone https://github.com/CaptainKeat/modelfit.git
cd modelfit
python source/modelfit.py

Your browser opens automatically. If it does not, open the local address printed in the terminal. The port is selected automatically.

1. Review the detected hardware at the top of the page.
2. Correct RAM or VRAM values if needed.
3. Choose a model category and adjust the filters/context setting.
4. Use Quit app to stop the server. Closing the browser tab alone leaves it running.

On macOS or Linux, use python3 if that is your Python command.

Edit or package the app

File	Purpose
source/core.py	Python server and hardware-detection template
source/ui.html	Editable interface and model catalog
source/modelfit.py	Generated standalone entry point
build.py	Combines the template and interface

After editing the template or UI:

python build.py
python source/modelfit.py

To build an optional executable locally:

python -m venv .venv
.\.venv\Scripts\python.exe -m pip install pyinstaller
.\.venv\Scripts\python.exe -m PyInstaller --onefile --name ModelFit source/modelfit.py

The result is under dist/. No executable is distributed with this source release.

Troubleshooting and limitations

• Blank VRAM: enter the GPU memory manually; detection varies by GPU and driver.
• Browser does not open: copy the address printed in the terminal into the browser.
• Fit looks optimistic: memory estimates do not account for every runtime, quantization format or competing process.
• Catalog looks dated: it is an August 2026 snapshot. Model descriptions, quality scores and license notes have not been revalidated for this release. Check the model publisher before relying on them.

The standalone app was regenerated and passed Python syntax and local HTTP page checks. Contributions to detection, estimates, accessibility and catalog maintenance are welcome.

See SECURITY.md and LICENSE. Model weights retain their own licenses.