← all repositories
haziqahmedqasmi/PeakCut

Self-hosted clip finder that works while the stream is still running

PeakCut turns long videos and live streams into captioned vertical clips on your own machine, with whatever AI provider you already pay for.

★1.7k stars Python Other AI
Too new to rank. Velocity needs a few days of history.
not enough history yet
star history

What it does

You paste a YouTube, Twitch or Kick link (or upload a file), and PeakCut transcribes the audio, asks a chat model which moments would hold a stranger’s attention, then renders those moments as 9:16 clips with word-by-word captions in one of three styles. It can also record a live stream into a rolling buffer and cut highlights while the stream is still going, using loudness, chat activity and the transcript. Finished clips can be queued or auto-posted to YouTube, Instagram and TikTok, with spacing and daily caps.

The interesting bit

The AI is entirely pluggable: Gemini, OpenAI, Grok, Groq, OpenRouter, Claude, a local Ollama model, or anything speaking the OpenAI chat format — and with no key at all it falls back to picking clips by loudness. Speech-to-text runs locally via sherpa-onnx with the Parakeet model, so audio never leaves your machine; only the transcript goes to the AI, and a one-hour video costs roughly four requests.

Key highlights

  • Live clipping from a rolling buffer, driven by loudness, chat activity and transcript
  • Per-creator workspaces with channels, clip settings, posting rules and an activity log
  • Runs fully offline for transcription; you pay your AI provider directly, with a usage page showing requests and tokens per video
  • Sensible multi-user behavior: per-user queues, self-monitoring of CPU/memory/disk, and admin roles that only the local database can grant
  • Plain HTML/CSS/JS frontend with no build step; FastAPI backend, ffmpeg and yt-dlp for the heavy lifting

Caveats

  • The first start downloads a ~480 MB speech model, and Windows users must keep the install path short (the 260-character path limit breaks model loading).
  • It checks GitHub once a minute for updates and sends two commit ids; you can disable this with PEAKCUT_UPDATE_CHECK=0.
  • Clipping other people’s content is your problem — the README is explicit that rights and site terms are on you.

Verdict

Worth a look if you produce clips regularly and want to own the pipeline instead of renting it from a SaaS clipper. Skip it if you need polished auto-editing magic — the AI only picks moments and writes titles; the rendering is straightforward ffmpeg work.

Frequently asked

What is haziqahmedqasmi/PeakCut?
PeakCut turns long videos and live streams into captioned vertical clips on your own machine, with whatever AI provider you already pay for.
Is PeakCut open source?
Yes — haziqahmedqasmi/PeakCut is open source, released under the MIT license.
What language is PeakCut written in?
haziqahmedqasmi/PeakCut is primarily written in Python.
How popular is PeakCut?
haziqahmedqasmi/PeakCut has 1.7k stars on GitHub.
Where can I find PeakCut?
haziqahmedqasmi/PeakCut is on GitHub at https://github.com/haziqahmedqasmi/PeakCut.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.