Skip to content
LiveClip

About LiveClip

An app that finds clip candidates from long VODs with clear reasons.

Start for freeView roadmapSend feedback

Spend less time finding highlights in hours of VOD

Candidates and reasons reduce the time spent wondering what to clip.

AI provides candidates, reasons, and suggested ranges. It is not a tool for mass-producing finished videos.

You make the final call. LiveClip only prepares the material needed for that decision.

What LiveClip does now

A place to review candidates, reasons, and context before editing.

The final call is yours

Review candidates and reasons, then accept, reject, or adjust the range yourself.

The video file is not uploaded

For the MVP, LiveClip handles only the needed audio data and thumbnails.

It does not replace editing software

LiveClip focuses on candidate selection and decision support, not full editing.

It reads livestream context

LiveClip is designed for solo Japanese streamers, including reactions and conversation flow.

Workflow

LiveClip creates the material needed for review from the VOD on your device.

Step 1

Choose a VOD

Select a long VOD from your device in the browser.

Step 2

Prepare in the browser

Create the audio summary and transcription audio. The video file is not uploaded.

Step 3

See the whole stream

Use transcript and heatmap context to review what happened around each moment.

Step 4

Review reasoned suggestions

AI provides candidates, reasons, and suggested ranges.

Step 5

Make the final call

Accept, reject, adjust the range, then export from the original video.

How we handle your data

What LiveClip takes in, what it sends to the server, when it gets deleted, and why none of it is used to train AI — explained based on each provider's official documentation and the code that actually runs.

About the videos you bring in

LiveClip only works with video files you pick on your device. Pasting a URL to pull videos from YouTube or Twitch isn't supported: platform terms don't permit unauthorized downloading or automated access, so we chose not to build on that. LiveClip never accesses an external site to fetch a video.

Read YouTube's Terms of ServiceRead Twitch's Terms of Service

What we send and how long we keep it

We keep the data leaving your browser to a minimum — and everything that does leave has a destination and an expiry.

Never sent

Your video file

Audio extraction and Short exports all happen on your device. Your footage never reaches our servers.

Deleted right after

Compressed audio for transcription

Sent only for transcription and deleted as soon as the analysis finishes, whether it succeeds or fails.

Kept with an expiry

Transcript and volume data

Stored for your plan's retention period (Free: 7 days / Pro: 90 days) so you can revisit results, then deleted.

Suggested moments only

Thumbnails

We store still images only for the moments the AI suggested, so you can tell candidates apart.

Why your data won't train AI

For every route your data travels, you can verify the no-training guarantee in official documentation.

Groq

Transcription

Transcribes your audio. Its data processing agreement rules out using inputs and outputs for model training, and the data it processes isn't retained by default.

View official documentation

Vercel AI Gateway

Route to the suggestion AI

Relays requests to the AI that drafts clip suggestions. Vercel doesn't retain prompts or responses, and doesn't use them to train its own AI.

View official documentation

OpenAI and other model providers

Generating suggestions

API inputs and outputs are not used for training unless you explicitly opt in — LiveClip has not. On top of that, every request is restricted to routes covered by no-training agreements.

View official documentation

LiveClip itself

This service

We never train our own AI on your data. Our privacy policy makes the same promise.

Read the privacy policy

The actual code

The relevant code, shown exactly as it is in the actual implementation.

Enforcing no-training routes

Every suggestion request carries a flag that routes it only to providers that don't train on your data. If no such route exists, the request simply isn't sent.

Source: src/lib/llm/propose-clips.ts
providerOptions: {
	gateway: {
		disallowPromptTraining: true,
		...(AI_GATEWAY_FALLBACK_MODELS.length > 0
			? { models: AI_GATEWAY_FALLBACK_MODELS }
			: {}),
	},
},

Immediate audio deletion

The audio used for transcription is always deleted, whether the analysis succeeds or fails.

Source: src/lib/analysis/run-analysis-request.ts
} finally {
	await deleteTranscriptionChunksSafely(validChunks, ownerId);
}

Who it is for

  • Solo creators who lack time to rewatch long streams
  • Part-time VTubers and independent streamers
  • Clip editors who want faster candidate selection
  • People who remember a good moment but not the exact timestamp

Not a fit / still weak for

These use cases are not the center of LiveClip today.
  • People who want publishing without a review step
  • People who need a full editing timeline
  • People who only need captions or text effects
  • People who want to replace their whole video asset manager

Try it with your own VOD first

Try it with a real stream. If anything feels off or missing, tell us.
Start for freeView roadmapSend feedback