Private footage
With Ollama for the LLM, nothing leaves your machine except the AssemblyAI call. Unreleased episodes, client VODs and internal talks stay in-house.
SupoClip is an AGPL-licensed OpusClip alternative you can run yourself. Give it a podcast, talk or stream VOD and it transcribes it, has an LLM of your choice pick three to seven clip-worthy moments, scores them, reframes to 9:16 with face tracking, burns in word-synced captions and exports for TikTok, Reels or Shorts. There's a hosted version from $9.99, an MCP server in the repo, and a REST API.
SupoClip's pipeline is the one every AI clipper uses: transcribe, let a language model pick the moments, score them, reframe, caption, export. What's different is that the code is public under AGPL-3.0, so you can run it on your own machine, swap the LLM, point it at a local Ollama model so no footage leaves your network, and change anything you don't like.
If you'd rather not run Docker, the same team hosts it at supoclip.com with YouTube-link input and an iOS app, priced by hours of video per month.
LLM= in the .env picks Gemini, GPT, Claude or Ollama. Transcription is AssemblyAI (API key needed).mcp/ folder and there's a REST API with keys, so Claude Code or a pipeline can drive it. See clipping MCP servers.Progress streams live to the browser while the ARQ worker runs each stage.
AssemblyAI returns a transcript with a timestamp on every word. That precision is what lets captions sync word by word later, and what lets the LLM reason about exact cut points.
The model you configured reads the transcript and returns the segments it thinks will stand alone as shorts. Swap Gemini for Claude or a local model and the picks change; that's the knob proprietary tools don't give you.
Every candidate gets hook, engagement, value and shareability scores that roll up to a 0–100 virality number. Sort by it, or by the one dimension your audience cares about.
Face detection keeps the speaker centred in the vertical crop as they move, so talking-head and podcast sources come out usable without manual repositioning.
Pick a caption template and font, add an AI hook title, optionally overlay Pexels stock B-roll and transitions, then export with the TikTok, Reels or Shorts preset. Trim, split and merge in the built-in editor first if a cut lands wrong.
You need Docker, an AssemblyAI key and one LLM provider key (or Ollama running locally). First start takes a few minutes while images build.
git clone https://github.com/FujiwaraChoki/supoclip.git && cd supoclip.env in the repo root (snippet opposite).docker-compose up -d, then docker-compose logs -f to watch it come up.localhost:3000, API at localhost:8000, docs at /docs.make test runs pytest, Vitest and Playwright.Stack: FastAPI + ARQ worker (uv), Next.js (pnpm), Docker Compose. Optional B-roll, analytics and email need extra config; the repo has a troubleshooting guide.
# transcription (required)
ASSEMBLY_AI_API_KEY=your_assemblyai_api_key
# clip selection: pick one provider
LLM=google-gla:gemini-3-flash-preview
GOOGLE_API_KEY=your_google_api_key
# alternatives:
# LLM=anthropic:claude-… ANTHROPIC_API_KEY=…
# LLM=openai:gpt-… OPENAI_API_KEY=…
# LLM=ollama:llama3 (local, no key)Monthly prices from supoclip.com, October 2026. Yearly billing gives two months free. No free hosted tier is listed.
Self-hosted costs aren't zero in practice: AssemblyAI bills per audio hour and the LLM bills per token. For a two-hour podcast expect a few dollars in API usage, or near-zero with Ollama and a local transcription swap.
With Ollama for the LLM, nothing leaves your machine except the AssemblyAI call. Unreleased episodes, client VODs and internal talks stay in-house.
No monthly hour quota. Clip fifty VODs a month for API costs only. For campaign clippers, that changes the maths in the earnings calculator.
MCP server and REST API in the box. Point Claude Code at it and build the loop you want. See what is a clipping agent.
The moment-selection prompt is in the source. Tune it for your niche (gaming callouts, finance hot takes) instead of guessing what a black box rewards.
Docker, API keys and a .env are a wall for some. The hosted plan removes it but caps hours and, on Pro, resolution at 720p.
No scheduler in the open-source build; hosted auto-posting is "coming soon". OpusClip or StreamClipping post for you today.
The code is free under AGPL-3.0 and self-hosting costs nothing beyond your hardware and API usage (AssemblyAI for transcription, plus your LLM provider unless you use Ollama). The hosted service starts at $9.99 a month.
AGPL-3.0 allows commercial use. The catch is that if you modify it and offer it as a network service, you must publish your changes under the same licence. Clipping your own videos with it is fine. Contributions grant the maintainer relicensing rights, per CONTRIBUTING.md.
The hosted plans do (90 min on Pro, 3 h on Creator). The README for the self-hosted build describes file input; check the repo for the current state of URL import, or download the source first with permission.
Almost. Clip selection can use a local Ollama model, but transcription is AssemblyAI by default, which is a cloud call. Swapping in a local Whisper is a code change, not a config switch.
FujiwaraChoki/supoclip is the upstream project (about 1.2k stars); most others on GitHub are forks. The hosted service at supoclip.com and the iOS app are run by the same project.