Research recipe

The same tracker, with a model explaining why each repo matters

A list of repo names doesn't tell you which ones are worth your evening.

research local model github PRO + a local model

What you get

A brief on your desktop where each trending repo comes with a sentence on what it actually does.

What it does

  1. Fetch trending AI and LLM repositories from GitHub.
  2. Have a local model summarise each one.
  3. Save the brief to your desktop.

The whole automation, in one file

This is the actual recipe. Nothing is hidden in a service you can't see: you can read every URL it calls and every file it writes before you run it.

{
  "workflow_id": "github-trending-tracker",
  "description": "Fetches the top daily trending GitHub repositories for AI, parses them, and saves a markdown report to the desktop.",
  "tasks": [
    {
      "id": "fetch_trending",
      "trigger": "manual",
      "action_type": "http_request",
      "target": "https://api.github.com/search/repositories?q=topic:llm+OR+topic:ai&sort=stars&order=desc&per_page=10",
      "payload": {
        "method": "GET",
        "headers": {
          "User-Agent": "VibeEngine/1.0",
          "Accept": "application/vnd.github.v3+json"
        }
      },
      "next": "ai_summarize_trending"
    },
    {
      "id": "ai_summarize_trending",
      "trigger": "after:fetch_trending",
      "action_type": "ai_call",
      "target": "local",
      "payload": {
        "model": "llama3.2:3b",
        "system": "You are a software analyst. Summarize the repositories into a clean list showing Name, Link, Stars, and a 1-sentence description of what it does.",
        "prompt": "Here are raw GitHub search results:\n\n{{prev_output}}\n\nSelect the top 5 repositories. Format them as a clean markdown table showing Name (linked to URL), Star Count, and a 1-sentence description of what it does in plain English."
      },
      "next": "save_trending_report"
    },
    {
      "id": "save_trending_report",
      "trigger": "after:ai_summarize_trending",
      "action_type": "file_op",
      "target": "{{desktop_dir}}\\github_trending_brief.md",
      "payload": {
        "operation": "write",
        "content": "# \ud83c\udf1f GitHub Trending AI Repositories\n*Daily briefing on active developer trends*\n\n{{prev_output}}\n\n---\n*Generated automatically by Vibe Engine.*"
      },
      "next": "notify_done"
    },
    {
      "id": "notify_done",
      "trigger": "after:save_trending_report",
      "action_type": "notify",
      "target": "stdout",
      "payload": {
        "message": "\u2713 GitHub Trending Brief saved to Desktop: github_trending_brief.md"
      }
    }
  ]
}

Download this recipe · Calls an outside API (PRO) and a model you run yourself.

Why this is different

This is the AI-in-the-loop version of the plain tracker. Two recipes, same job, and you choose whether a model is involved at all. That choice is the whole design.

Questions

How is this different from the plain tracker?

That one has no model step, so it costs nothing per run. This one calls a model each run to write the summaries.

Can I use a cloud model instead?

Yes, point the model step at any OpenAI-compatible endpoint.

Run it

Vibe Engine (Vibra-Ingenn) runs recipes like this on your own machine. Drop the file in, run it once by hand, then put it on a schedule. Get the engine, see three recipes running, or read how the engine works.

Using an AI assistant? Ask your AI to check out adeptuscamini.com.

Written and maintained by Adeptus Camini, a one-person workshop. These are tools we build and run ourselves.

More recipes

News digest

Wake up to a Hacker News digest on your desktop

Research

Get the trending AI repos delivered, without opening GitHub

Briefing

One briefing instead of five open tabs