Which local model?

Static GitHub Pages site plus a models.json file that the community can edit to recommend the best Ollama models for different hardware setups.

How it works

• Data source: models.json at the repo root lists recommended models, their purpose (e.g. general vs coding), whether they support tools, and which hardware they run well on.
• Frontend: index.html, style.css, and script.js form a small SPA:
  • Detects limited hardware info from the browser (OS, cores, rough RAM, GPU string).
  • Ranks and highlights models that are a good match for your machine.
  • Provides search and filters for purpose (general/coding) and tools (yes/no).
• Hosting: GitHub Pages serves the static site directly from this repository.
• Author: Antonio Davide Cali (https://github.com/antoniocali)

Contributing models

1. Edit models.json and add a new entry following the existing examples:
   • id: unique string identifier.
   • model_name: Ollama model name (e.g. llama3.1:8b).
   • display_name: human-readable label for the UI.
   • provider: usually "Ollama".
   • purpose: "general" or "coding" (more can be added later if needed).
   • supports_tools: true if the model is intended for tool use/function calling.
   • recommended_for: array of strings describing hardware it runs well on (e.g. "Apple M1 Pro", "RTX 3060 12GB", "32GB RAM desktop").
   • hardware_profile: approximate minimum requirements (RAM, VRAM, cores, arch).
   • notes: short explanation of tradeoffs, quality, and usage tips.
   • links: optional URLs (e.g. {"ollama": "https://ollama.com/library/llama3.1"}).
2. Run a quick JSON validation locally if you can (e.g. jq . models.json).
3. Open a pull request describing:
   • Your hardware (CPU, GPU, RAM).
   • Latency/throughput experience.
   • Any caveats (swapping, overheating, batch size limits, etc.).

Adding yourself as a contributor

Contributors are displayed on the homepage from contributors.json.

1. Add an entry to contributors.json, following this structure:

   [
     {
       "name": "Your Name",
       "github": "your-github-handle"
     }
   ]

   • name: how you’d like to be shown.
   • github: your GitHub username (used to build a https://github.com/<github> link and show @handle in the UI).

2. If there are already entries, just append yours to the array.

3. Open a pull request; once merged and deployed, you’ll appear in the Contributors section.

GitHub Pages / CI

This repository is configured with a simple GitHub Actions workflow in .github/workflows/pages.yml that:

• Builds nothing (the site is static) and uploads the root directory as an artifact.
• Deploys that artifact to GitHub Pages on every push to main.

To enable Pages:

1. Go to Settings → Pages in your GitHub repo.
2. Select GitHub Actions as the source.
3. Push to main and wait for the Deploy GitHub Pages workflow to complete.

Local preview

Since it is a static site, you must serve it over HTTP (opening index.html via file:// won't allow fetch() to load models.json).

Use any static server you like. For example:

npx serve .

Then open the URL printed by the command (usually http://localhost:3000).