DeepSeek is a strong reasoning model, but you cannot hand it an episode. It will listen to you ask a question, and it will read a PDF you attach — yet a two-hour podcast or an mp4 has nowhere to go, and the API exposes no transcription endpoint. The missing piece is a conversion step: turn the recording into text first, then let DeepSeek do what it is good at. BibiGPT is that step — paste a link or a local file, get a timestamped transcript, chapter summaries and notes. Since DeepSeek open-sourced its agent harness on 2026-08-13, you can also install BibiGPT into it as a skill, so the agent runs both halves on its own.
Paste a podcast, lecture or video link — the transcript and chapter summaries come back in seconds, ready to hand to DeepSeek. The live BibiGPT app loads inline; no redirect.
DeepSeek cannot transcribe audio or video on its own — there is no upload for it and no transcription endpoint to call. Convert first, then hand it the text. BibiGPT does the converting: paste a link or a local file, get a timestamped transcript, chapter summaries and notes. And since DeepSeek open-sourced its agent harness on 2026-08-13, BibiGPT can be installed into it as a skill, so the agent does the fetching, the converting and the reasoning in one go.
Nothing here is a knock on the model. It is a plumbing gap: the input DeepSeek accepts and the format your content is in do not line up.
DeepSeek's chat app has no audio or video upload, and its API exposes no transcription endpoint. Whatever you want it to reason about has to arrive as text.
Pasting a YouTube or podcast URL into a chat gives the model a string of characters, not the two hours of talking behind it. It will happily guess from the title — which is exactly the failure you want to avoid.
People try longer instructions, role-play, even asking the model to "listen carefully". The fix is upstream: convert the recording to text, then send the text.
One paste turns a recording into the three things a reasoning model actually needs: the words, where they were said, and the shape of the whole thing.
Export as SRT or VTT when you need the timing, plain text when you just need the words, or Markdown when it is going into Notion or Obsidian.
A long recording comes back split into sections with a summary per section, so you can feed DeepSeek the part you care about instead of an undifferentiated wall of text.
YouTube, Bilibili, podcasts, TikTok, X and Xiaohongshu links go in directly, and so does an audio or video file sitting on your disk.
DeepSeek Harness (dsh) is DeepSeek's open-source agent harness, MIT-licensed and currently a developer preview. Launch it with npx @deepseek-ai/dsh web. The steps below add BibiGPT to it — one command, nothing to clone.
dsh plugin --profile web add "github:JimmyLv/bibigpt-skill#path:/dsh-plugin" — keep the quotes, since # starts a comment in the shell. dsh hands the argument to pnpm, and pnpm can install a package straight out of a subdirectory of a git repo, so there is nothing to publish and nothing to clone. Use --profile tui or --profile headless to target those instead.
The harness also reads skills off disk: drop the skills/bibi folder into ~/.agents/skills/ and it is picked up with zero dependencies. That directory is shared with other SKILL.md-aware agents, so one copy serves Claude Code and DeepSeek Harness at the same time.
Restart the profile, open the command palette and look under Skills — /bibi is there. Give it a link and ask for a summary, a transcript or chapter notes; it fetches, converts and hands the text back into the conversation.
Both paths end with DeepSeek reading a transcript. They differ in how many windows you touch to get there.
| Step | BibiGPT skill inside DeepSeek Harness | Manual, no skill |
|---|---|---|
| Getting the recording | Paste the link into the conversation | Download the video or rip the audio yourself |
| Turning it into text | The agent calls the skill and waits | Upload to some transcription service, wait, download the file |
| Getting it into DeepSeek | Already in the conversation | Open the transcript, copy it, paste it, hope it fits |
| Chapters and timestamps | Come back with the transcript | Rebuild them by hand, or go without |
| Repeating it for 20 episodes | Same one instruction, in a loop | Twenty rounds of the above |
The common thread: the reasoning was never the bottleneck — getting the recording into a readable shape was.
A ninety-minute lecture becomes a timestamped transcript split into chapters. Feed DeepSeek the section you are stuck on and argue with it about that, instead of pasting an hour of raw text and hoping.
Pull the transcript off a short video, then have DeepSeek restructure it into a script, a thread or a newsletter. The transcript keeps the phrasing that made the clip work, which a summary would have flattened.
With the skill installed, an agent running in DeepSeek Harness can take a link, get the text and act on it without a human copying anything. That is what turns a one-off into a nightly job over a whole feed.
Everything on this page about DeepSeek Harness comes from the project itself.
DeepSeek Harness — DeepSeek's open-source agent harness, MIT-licensed, developer preview, built on Cordis, published 2026-08-13.
deepseek-ai/deepseek-harness on GitHub ↗The BibiGPT skill and the harness bundle that registers it, including the ~/.agents/skills install path.
JimmyLv/bibigpt-skill on GitHub ↗No. DeepSeek's app has no audio or video upload and its API has no transcription endpoint, so there is nothing to send a recording to. To get audio into DeepSeek you convert it to text with a separate transcription tool first, then paste or pipe that text into the conversation.
DeepSeek Harness, or dsh, is DeepSeek's open-source agent harness, published on GitHub on 2026-08-13 under the MIT licence and labelled a developer preview. It is built on Cordis and organised so that everything is a plugin, which is how outside capabilities such as reading a video get added to it.
A skill is a folder holding a SKILL.md file whose frontmatter tells an agent what the skill does and when to reach for it. DeepSeek Harness discovers skills from disk, including the shared ~/.agents/skills directory, so the same folder can serve several SKILL.md-aware agents at once.
Why people use BibiGPT to turn videos into text every day.
Trusted by 50,000+ users worldwide
“I paste a link and get clean captions in seconds — it saves me hours of retyping every single week.”
Maya R.
Content Creator · Repurposes short videos
“Exporting the transcript lets me review new words at my own pace instead of pausing the video constantly.”
Daniel K.
Language Learner · Studies with real videos
“Accurate, timestamped text I can quote directly. It has quietly become part of my daily workflow.”
Priya S.
Researcher · Cites public talks
FAQ'S
Ask us anything!
1 Best Bilibili AI video summary tool 2026? Paste a link for a free AI recap, mind map, and transcript-style takeaways on 30+ platforms — no login needed.
2 Install a video-summary skill into DeepSeek Harness in one copy-paste — SKILL.md matches Claude Code. Or skip dsh: paste a Bilibili or YouTube link in the browser and get a timestamped summary.
3 Reverse any video free with no app and no download, right in your browser. Step-by-step for iPhone, Android, and PC, plus how to reverse audio (2026).
Paste a link or drop a local file — transcript, timestamps and chapter summaries come back in seconds, ready to hand to DeepSeek. Free to try, no signup needed.
