Privacy Policy
Effective 19 October 2026. This policy describes what SivqAI collects, why, who processes it, and how to have it removed.
Before publishing: replace the operator details in “Who we are” with your real legal entity, address and contact address, and set the governing law in the Terms. Have a lawyer review both documents — Google’s OAuth verification reviewers read them closely, and this draft describes the software accurately but is not legal advice.
Who we are
SivqAI (“SivqAI”, “we”) operates the clipping automation service at sivqai.com. Operator: [legal entity name], registered at [registered address]. Privacy enquiries: privacy@sivq.ai.
Two kinds of people use SivqAI
Clippers hold an account and use the dashboard. Creators are streamers who authorize a clipper to clip their channel. A creator does not need an account, and we hold far less about them. Where the two differ, this policy says so.
What we collect from clippers
| Data | Why |
|---|---|
| Email address and password | Your account. Authentication is handled by Supabase; we never see your password. |
| Twitch account identity and access tokens | To look up channels, read chat for the channels you are authorized to monitor, and create clips. Tokens are encrypted at rest. |
| Google account email and YouTube channel identity | To identify which channel uploads go to and to upload the clips you approve. |
| Campaign details you enter | To check your clips against the rules you record. |
| Clips, transcripts and generated metadata | To produce, store and publish your videos. |
| Processing logs | Diagnostics and an audit trail of what was done with authorized content. |
What we collect from creators
- The Twitch channel ID, login and public channel metadata of the channel being authorized.
- Your approval decision, the exact permissions you granted, the version of the terms you accepted, and when.
- How your control of the channel was verified, and the identity that verified it (your Twitch account ID and login).
- Your Twitch access token, encrypted, where you granted clip management. It is used only to create and download clips from your own channel, and is destroyed when you withdraw permission.
- Your IP address and browser user agent at the moment you approved or declined, as evidence of the decision.
- Content guidelines you choose to write.
Chat data
To find clip-worthy moments, SivqAI reads public chat for channels it is authorized to monitor. It computes activity measures — messages per minute, how many distinct people are talking, how much reads as reaction, and spam indicators. Individual chat messages are processed in memory and are not stored. What persists is the resulting numbers and a plain-language reason such as “chat activity increased 4.8×”.
Video content
SivqAI only holds video it has a specific, recorded permission to hold. Footage reaches us one of three ways, each tied to a creator authorization: through Twitch’s official clip-download API using permission the creator granted; uploaded by the creator; or uploaded by a clipper where the creator permitted that. SivqAI does not download whole livestreams or whole VODs — Twitch’s API does not offer that and we do not attempt it.
How AI is used
- Speech recognition runs on our own infrastructure using self-hosted Whisper models. Your audio is not sent to a speech API.
- Hooks, titles, descriptions and campaign rule extraction send the clip’s transcript text, or the campaign requirements you paste, to Anthropic’s Claude API. Video and audio are never sent there — text only.
AI output is drafted, not verified. A human reviews it before anything is published.
Who else processes your data
| Processor | What they receive |
|---|---|
| Supabase | Database, authentication and encrypted media storage. |
| Anthropic (Claude API) | Transcript text and campaign requirement text. No video or audio. |
| Twitch | API requests made with your or a creator’s granted permission. |
| Google / YouTube | The video and metadata you approve for upload. |
| Vercel | Serves the website. |
We do not sell your data, your footage or your transcripts, and we do not license them to third parties.
Google user data: Limited Use
SivqAI’s use of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements.
We request these scopes and use them only as described:
| Scope | Why |
|---|---|
openid, userinfo.email | To identify which Google account is connected. |
youtube.readonly | To identify the YouTube channel uploads would go to, and to read back the status of videos SivqAI uploaded. |
youtube.upload | To upload a video you have explicitly approved, to your own channel. |
Specifically: Google user data is not transferred to others except as necessary to provide the service, is not used for advertising, is not sold, and is not used to train generalised AI models. SivqAI does not upload anything to your channel without a per-video approval recorded against your account.
How long we keep things
- Account data — while your account exists, then removed on request.
- OAuth tokens — until you disconnect the account or revoke access with the provider. Disconnecting deletes the stored token, not merely flags it.
- Authorization records — retained after revocation, because the record of who permitted what, and when, is the evidence that processing was lawful.
- Footage and clips — retained after a revocation unless deletion is requested. Revocation stops all new processing immediately; it does not by itself erase what is already stored. We say so plainly to creators at the point of revocation.
Your choices
- Clippers can disconnect Twitch or YouTube at any time in Settings.
- Creators can withdraw permission from the link given at approval, or by revoking SivqAI in Twitch’s own Connections settings. Either stops all new processing.
- Anyone can ask for access to, correction of, or deletion of their data, including stored footage, by emailing privacy@sivq.ai.
Deletion requests are handled by email today rather than by a self-service button. We would rather say that than imply a button exists.
Security
OAuth tokens and upload session URLs are encrypted at rest. Capability links — approval, upload and revocation links — are stored only as keyed hashes, so the database alone does not yield working links. Access is scoped per user at both the application and database layers. No system is perfectly secure, and we do not claim otherwise.
Children
SivqAI is not intended for anyone under 13, or under the minimum age required by Twitch and YouTube in your country, whichever is higher.
Changes
Material changes to how creator content is processed are versioned. A creator who approved under earlier terms is asked to review and approve again rather than having the new terms applied to their existing permission.