C:\> HOME / PROJECTS / ..

Open source · Live streams to text / Local AI

99talks

99talks turns YouTube Live streams and recorded videos into timestamped transcripts on your own computer. It uses faster-whisper on the CPU and shows the text in a browser as the talk plays.

Catch the talk. Keep the text.

Each transcribed segment is written to disk as it arrives. Past sessions stay available for export, and a live capture can reconnect after a dropped stream. Whisper supports multiple languages, with automatic detection or an explicit language choice.

From a stream to your AI tools

Export Markdown with the title, channel, source URL, date, description, and timestamped paragraphs. TXT, SRT, and VTT exports are available too.

Bring the Markdown into NotebookLM, ChatGPT, Claude, or your own agents to work on notes, summaries, or a knowledge base. 99talks produces the transcript; importing it into another AI tool is a separate step.

Local transcription, not a cloud service

The Python/FastAPI app runs on your computer with a local browser interface. Speech recognition uses your CPU, without a transcription-service API key. Fetching video streams and downloading a model require network access; transcription and saved sessions stay local unless you choose to share an export.

Run it locally

Install ffmpeg and yt-dlp, then follow the repository setup. The documented quick start uses Python 3.12:

git clone https://github.com/rfdslabs/99talks.git
cd 99talks
python3.12 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python server.py

Open http://localhost:8000, paste the stream or recording URL, select the language, and press Start. The first run downloads the Whisper model. YouTube is the tested input source.

Know the limits

One capture runs at a time. Text follows the speaker with a processing delay, and there is no speaker diarization: the transcript does not identify individual speakers. Transcription errors and chunk-overlap repeats can still occur. Check the repository for current requirements and setup details.

MIT licensed. This portfolio provides project information and the source link, not a hosted transcription service. Run 99talks locally; the static Cloudflare site does not capture audio or process video.

C:\> LOCAL TERMINAL / NOT A SHELL