YouTube clickbait detection

Desktop app (Tkinter + Ollama) that reads a video’s captions, summarizes them, and compares that summary to the title. There is no browser and no local web server — only a normal window.

Prerequisites

1. Install Ollama so the ollama command is on your PATH (default API http://127.0.0.1:11434). When you run the app, it tries to start ollama serve automatically if nothing is already listening; closing the app stops only that child process (if the app started it).
2. Pull at least one model, for example: ollama pull llama3.2 or ollama pull gpt-oss:20b (GPT-OSS on Ollama).
3. Install uv.

The window shows Ollama status (reachable or not) and refreshes every ~12 seconds. Analysis jobs go through a FIFO queue: only one Ollama generation runs at a time; the next job starts when the current response is finished. This avoids overlapping requests that often trigger “max retries exceeded” / connection errors.

If Ollama shows “not reachable”

1. Start Ollama (desktop app or ollama serve). In a terminal, ollama list must work.
2. In the app, set Ollama URL to http://127.0.0.1:11434 or http://localhost:11434 and click Test connection. If one works, the dialog can suggest switching the URL.
3. HTTP_PROXY / corporate proxy: this project uses requests with proxy env vars ignored for Ollama so localhost is not sent through a proxy. If it still fails, check VPN/firewall and that port 11434 is allowed for local traffic on Windows.

Setup

uv sync

Run the desktop UI

uv run youtube-clickbait

Or:

uv run python app.py

Paste a YouTube URL and click Analyze (add to queue). You can queue several URLs; they run in order.

Videos without captions cannot be analyzed (this project does not download or transcribe audio).

Environment (optional)

Variable	Meaning	Default
OLLAMA_HOST	Ollama base URL	http://127.0.0.1:11434
OLLAMA_MODEL	Default model in the UI	llama3.2
OLLAMA_AUTO_START	Set to 0 to disable auto ollama serve	1 (enabled)
OLLAMA_EXE	Full path to ollama / ollama.exe if not on PATH	(unset)

If the app says ollama was not found but Ollama is installed, Windows often puts the binary at %LOCALAPPDATA%\Programs\Ollama\ollama.exe. Either add that folder to your user PATH, or set OLLAMA_EXE to that file once (PowerShell: $env:OLLAMA_EXE = "$env:LOCALAPPDATA\Programs\Ollama\ollama.exe").

Build a Windows .exe

From the repo root (installs dev deps including PyInstaller):

uv sync --group dev
scripts\build_exe.bat

The executable is written to dist\youtube-clickbait.exe. Copy that file anywhere; Ollama must still be installed and running on the machine.

iOS

There is no iOS app in this repository. Python + Tkinter + Ollama’s local API do not map to an App Store–friendly stack. A real iOS client would be a separate native app (Swift/SwiftUI) that calls Ollama on your Mac or over the network.

Add packages with uv add <package>.