NTIndex

NTIndex is a static site generator for NeonTeam model-swap videos.

Site: https://ntindex.kro.kr

Use it to search NeonTeam swaps like A as B without digging through YouTube titles manually.

It collects video metadata, stores the normalized source data in SQLite, and builds static JSON and HTML files for browsing and searching.

SQLite is the source of truth. Files under dist/ are generated artifacts.

Requirements

• Python 3.11+
• uv

Setup

uv venv
uv sync --dev

Common Commands

Crawl Videos

Default crawl uses yt-dlp against the configured NeonTeam channel.

uv run ntindex crawl

Default values:

• --db: ntindex.sqlite3
• --channel-id: UCI4No3r3X66tSQbVgXse_MA

Use a different channel ID:

uv run ntindex crawl --channel-id UC_other

Use a channel URL directly:

uv run ntindex crawl --channel-url "https://www.youtube.com/@somechannel/videos"

Use YouTube RSS/Atom for recent uploads only:

uv run ntindex crawl --recent

Use a specific feed URL:

uv run ntindex crawl --feed-url "https://www.youtube.com/feeds/videos.xml?channel_id=UC_other"

Import local JSON:

uv run ntindex crawl --input examples/videos.json

Expected JSON item shape:

{
  "title": "Furina as Nahida | Genshin Impact Model Swap",
  "link": "https://www.youtube.com/watch?v=example",
  "published_at": "2026-01-01T00:00:00Z"
}

Only titles matching this pattern are indexed:

A as B | Game Model Swap

Build Static Site

uv run ntindex build

Default output directory:

dist/

Use a different output directory:

uv run ntindex build --dist public

Generated files include:

dist/
├── index.html
├── search.json
├── style.css
├── app.js
└── game/
    └── <game-slug>.html

Update

Run crawl, then build:

uv run ntindex update

Use RSS for recent uploads:

uv run ntindex update --recent

Use a custom output directory:

uv run ntindex update --dist public

Merge

Merge commands use canonical IDs. Rows are not deleted. Before applying changes, merge commands print a summary and ask for confirmation.

• id identifies the actual stored row.
• canonical_id identifies the representative row used for search and build output.
• Merge updates video rows to the canonical target ID.
• Build output still includes merged names as searchable aliases.

Important exception:

• characters.game_id is not rewritten by merge game.
• Character rows keep their original game row so alias names remain traceable.
• Search and build output still resolve those games through games.canonical_id.
• This avoids UNIQUE(name, game_id) conflicts when both merged games already contain the same character name.

In short, video references are normalized after merge, but character alias rows keep their original game ownership.

Merge Games

uv run ntindex merge game <old_id> <new_id>

Example:

uv run ntindex merge game 12 3

This makes game 12 resolve to the same canonical game as 3.

Skip the confirmation prompt:

uv run ntindex merge --yes game 12 3

Merge Characters

uv run ntindex merge character <old_id> <new_id>

Example:

uv run ntindex merge character 51 8

Characters can be merged only when they belong to the same canonical game.

Skip the confirmation prompt:

uv run ntindex merge --yes character 51 8

Merged names remain searchable through search.json alias fields:

• source_names
• target_names

Parse Failures

NTIndex records titles that were crawled but could not be parsed into the A as B | Game Model Swap shape.

List unresolved parse failures:

uv run ntindex failures list

Filter by review status:

uv run ntindex failures list --status unreviewed
uv run ntindex failures list --status ignored
uv run ntindex failures list --status needs_parser

Include resolved failures:

uv run ntindex failures list --all

Limit the number of rows:

uv run ntindex failures list --limit 20

Output JSON:

uv run ntindex failures list --json

Set review status:

uv run ntindex failures review <failure_id> ignored
uv run ntindex failures review <failure_id> needs_parser --note "parser candidate"
uv run ntindex failures review <failure_id> unreviewed

Review statuses:

• unreviewed: not checked yet
• ignored: intentionally outside the current A as B data model
• needs_parser: likely worth supporting with parser changes

Development

Run tests:

uv run pytest

Current tests cover:

• title parsing
• SQLite insertion and duplicate skipping
• static site build output
• CLI crawl/build/update paths
• RSS feed parsing
• yt-dlp metadata parsing
• canonical merge behavior

Project Notes

• SPEC.md is the product specification.
• README.md is the usage guide.
• dist/ is generated and ignored by git.
• ntindex.sqlite3 is local runtime data and ignored by git.
• uv.lock should be committed for reproducible installs.