social-knowledge-base turns all of a creator's videos into a knowledge base you can search. It works on Instagram today. TikTok and YouTube support are still in progress, according to the repo.
It runs in four steps. First it downloads the account's videos. Then it transcribes them in bulk with faster-whisper. Next it writes one markdown note per video, using the transcript plus frames and OCR. Last, it builds an OVERVIEW file that groups the notes by topic, and query.py lets you search the set.
The job is resumable. If a run stops partway, the manifest records what is done, so you do not start over.
The verdict in my tracker is validated by repo design. That is the stamp I logged, and I am keeping it at that level. The repo is public and the README matches the design, step by step.
Cost: free. The code is open source under the MIT licence, and transcription runs on your own machine. A GPU is recommended at scale, and the notes step uses your own LLM, so that usage is on you.
For me, this is the pattern behind my youtube-watch idea: build notes from a creator I learn from once, then ask questions across all of their videos instead of rewatching them one by one.