Liam Search

Search words/phrases from every Liam VOD after July 2023. Inspired by yardsear.ch.

Currently using these playlists:

• https://www.youtube.com/playlist?list=PL-dR2WR6nR_YPhchO2f48lssHKENSMAYy
• https://www.youtube.com/playlist?list=PL-dR2WR6nR_ZYPbJhACQKEiyb43DXunBy
• https://www.youtube.com/playlist?list=PL4p5tSr0nlvikGvf0bhqFuQoFAH7Iw9Ay

Official URLs

• Frontend: https://liamsear.ch/
• Backend: https://api.liamsear.ch/

Development

Prerequisites

• Bun
• uv
• ffmpeg shared (I'm using ffmpeg 8)
• CUDA Toolkit 12.8.1
• Meilisearch
• An NVIDIA RTX GPU with at least 6 GB VRAM (i think)

Caution

If you have problems with downloading videos, use yt-dlp 2026.03.03 and set the binary path in backend/config.toml

Backend

1. Navigate to the backend directory
2. Install dependencies with bun i
3. Copy .env.example to .env and configure your Meilisearch details
4. Download your YouTube cookies while signed-in with Get cookies.txt LOCALLY (chrome, firefox) and put them in the backend directory as cookies.txt (liam-search/backend/cookies.txt)
5. Modify config in config.toml if needed
6. Navigate to the backend/src/transcriber directory
7. Install dependencies with uv sync
8. Navigate back to the backend directory and run with bun dev

Tip

Modify the playlists array in backend/config.toml for testing.

Frontend

1. Navigate to the frontend directory
2. Install dependencies with bun i
3. Copy .env.example to .env
4. Run with bun dev

Roadmap

• Rework the job system: jobs api endpoint, batch jobs, complete rewrite
• Twitch clips: index all Liam Twitch clips before the oldest VOD in the current index
• Better transcription: i hate whisperX and python >:(
• Config file: allow for more configuration without modifying the code
  • GPU/CPU switch: switch between GPU mode and CPU compat mode
• Better error handling