Every Creator's Mind. One Conversation.

Every answer is generated only from the channel's actual video transcripts and is cited back to the specific videos and timestamps. The AI is instructed to answer from the source material rather than guess — click any [mm:ss] citation in the answer and the cited moment plays in-frame so you can verify the claim against the original video in one click.

Yes — that's what Projects are for. Curate videos from any channels into a focused project, then chat across the whole bundle. Citations carry across channels seamlessly. Perfect for cross-creator research (“what do Huberman, Attia, and Rhonda Patrick say about cold plunges?”) or building a knowledge base from a specific topic across many shows.

For every video you ingest, we auto-generate a one-click insights panel: a 2–3 sentence TL;DR, 5–8 key points with timestamps you can click to jump to, top questions a viewer would ask, viral hooks (short-form opening lines), and content angles to repurpose. Cached after first generation so re-visits are instant. Click any timestamp in the insights panel and the embedded player seeks there.

Channel-wide ingest is parallelized — a 500-video channel typically finishes in 5–10 minutes. For a curated set, use Bulk Import: paste up to 50 YouTube URLs at once, optionally have the system create a new project from the batch in one click. Videos already in our shared canonical library are added to your account instantly with no re-fetch.

Yes. Channels you add are auto-refreshed weekly by default — Friday cron pulls new uploads, the worker transcribes them, and your chat picks them up automatically. Every Sunday you get an email digest listing the new videos across every channel you follow with one-click jumps into their AI Insights. You can toggle auto-refresh off per channel or trigger a manual refresh anytime.

We have a two-stage fallback chain. The primary fetch tries TranscriptAPI for native YouTube captions. If those don't exist, we fall through to Supadata — which can AI-generate a transcript from the audio itself. Videos that genuinely have nothing to transcribe (music-only clips, silent shorts, etc.) are gracefully skipped. Everything else is indexed in full, no length cap on long podcasts or lectures.

Yes. Every channel detail panel has a Download all transcriptsbutton — gives you a single clean TXT (or JSONL) with every transcript ready to feed into your own LLM, RAG pipeline, or research archive. We don't lock the data; it's yours.

Chat and insights use modern frontier models routed automatically — if the primary provider is overloaded or refuses the content, we fall back to a secondary so you always get an answer. Embeddings use OpenAI's text-embedding-3-small. Model choice is constantly tuned for cost and quality; we don't lock you into any specific vendor.

TubeBase works with publicly available transcripts and metadata from public YouTube channels — the same captions anyone can view on YouTube. We don't rehost videos. Transcripts are used to power search and Q&A for your own research. You're responsible for how you use the output.

Yes — if it's not a fit, email us within 7 days of your purchase at hello@tubebase.io and we'll refund you.