Privacy Policy

This policy covers the AINSOF music agent — the connector you add at https://mcp.ainsof.io — and this website. AINSOF is operated by Or Chausha AINSOF, a sole proprietorship registered in Israel. Contact: info@ainsof.io

What the agent can reach

The agent has sixteen tools, and every call records the operational request log described under “What we record”. That is why all of them are disclosed to clients as changing private first-party state, even when the requested action only searches or computes a result. The tools cannot inspect your files, accounts, conversations or other connectors. They receive only the catalog query, track ID, public or authorized reference, or video that you explicitly provide.

Some tools also create private service artifacts. score_my_video — if you give it a link to a video, it downloads that video, renders a copy with our music mixed onto it, and gives you a link to the copy. analyze_video creates private analysis state, and deliver_score may create a private delivery archive. get_upload_link — if the video is on your own machine instead, it hands back a one-time address you upload it to; the file goes straight from your machine to our storage and never passes through the conversation. search_by_reference may create private reference-analysis state. feedback stores one line about how an answer landed, and sends it only after you explicitly consent. fit_to_length cuts a track to the length you ask for and stores the edit and a download package on our infrastructure for seven days, then deletes the files. We keep the record of what was requested. Listening to the edit needs no account. Downloading the package needs you to sign in to an AINSOF account with an active plan; each download records which edit it was, your account ID and email, your country and your browser’s user agent. Video is described separately below, in full, because it is the part of this service that handles something of yours.

If you tell us an answer was wrong

When you react to something we returned — "none of these fit", "that one is perfect" — the agent may send that reaction to us through feedback. We store the sentence you said, which tool it was about, and the track or brief it concerned, so that the people who choose and tag the music can see where we are getting it wrong. We store nothing else from the conversation. The ordinary request log may associate the feedback call with an AINSOF key identifier, client application and pseudonymous caller hash, as described below. Nothing is sent unless you actually said something about a result, and you can tell the agent not to send feedback at all.

The agent cannot:

• read, open or list any file on your computer
• see your conversations, projects or documents
• reach any other connector, account or service you use
• change or delete anything on your computer or in another account; files you intentionally upload to AINSOF are removed under the schedule below
• fetch any video other than the one whose link you send it

It receives only what you send it in a request — a written brief, a video length, a reference link, a track ID, or a link to a video you have asked it to score.

If you send us a video

Nothing here happens unless you ask for it yourself. Two tools reach a video, and only when you hand them one: score_my_video, which puts music on it, and analyze_video, which only measures it. analyze_video downloads the video and reads its length, frame rate and cuts. It also takes up to twelve still images from it and a transcript of any speech in its first two minutes, and returns those to the AI assistant you asked. It renders nothing. We keep those stills, the transcript, the one-sentence description of what you wanted (if your assistant passed one) and a reduced-resolution copy of the video for 30 days, so our team can review how well our tools served your edit. They are then deleted automatically. Ask us sooner and we will delete them sooner. Without the video or any image from it, we keep usage statistics for each request: date, country, which app and channel sent it (MCP or API), its length and edit pace, whether it has speech and in which language, the kind of project and a few words on the music wanted (if your assistant passed them), and whether the result was scored, watched or downloaded. If you used an AINSOF upload link, the storage original remains available for the score step until that link's two-hour window ends. When you call score_my_video:

• We download that video to our own machine and read it — its length, frame rate and where it cuts — so the music can be matched and placed.
• We render a new copy with the music on it. The temporary copy on our processing machine is deleted when the operation ends. If you used our upload link, that uploaded original is scheduled for immediate deletion after the verified render and delivery files are safely stored. A failed deletion is retried. If processing fails, the uploaded original remains available for retry until the original two-hour upload window ends, then is scheduled for deletion. An upload you never start follows the same window.
• The render is stored on our infrastructure. Each time its AINSOF link is opened, the link creates a temporary signed storage address. The AINSOF link itself can be opened again until the render is deleted. It is not listed anywhere and its identifier is generated randomly.
• The render is scheduled for automated deletion after seven days. A failed deletion is retried. Ask us sooner and we will delete it sooner.
• Your video is never used to train anything. We do not send it, or the stills and transcript taken from it, to any AI service other than the assistant you are already using — the measuring and mixing are ordinary audio and video processing on our own machine. Supabase, our storage processor named below, handles those files under the deletion schedules described here.

On ainsof.io, analyzing a video and downloading need you to be signed in, so each video you analyze and each file you download is recorded against your account. If you use Sync to video, your video is kept in My videos on our own machine for 30 days after you last use it, so you can come back to it, try music on it and download it; then it is deleted. You can delete it yourself at any time from My videos. A downloaded synced video stays available to you for seven days, then is deleted.

In ChatGPT, Claude or another app, and through the API without a key, working with a video first asks you to sign in once on ainsof.io with a link shown in the chat. We record that this sign-in link was opened from that app and confirmed by your account (account ID and email), and which video tools then used it, with your country and the app’s user agent. The link stops working after 30 days.

If you subscribe to a plan or buy a single-track license

Plans and single-track licenses are sold by Paddle (Paddle.com), our reseller and Merchant of Record. You enter your payment details with Paddle, in Paddle's own checkout form; they never pass through AINSOF, and we never receive or store your full card number. From Paddle we receive your name, email address and country, your plan and billing cycle (or, for a single-track license, the track and the license you chose), the amounts charged, and Paddle's transaction, subscription and customer IDs. Paddle handles your payment details under its own privacy notice.

Because a license lasts for productions published while you were subscribed, we keep the evidence of it: your payment and subscription records, the version of the Terms you accepted and when, and a record of each track you downloaded: which track and version, when, and under which plan or single-track license. A single-track purchase is kept the same way as a subscription record, because the production it licensed stays licensed for good. We use these records to show what you are licensed for (for example, to help release a copyright claim), and for our accounting and tax.

When you send us a copyright claim or a Help Center request, we keep what you send: your name and email, your message, the links, and any screenshot or file you attach, and for a claim, the license details we checked on your account. Screenshots and files are held in private storage, and reach our team by email (through Resend and Google, listed below). To get a claim released, we may pass the link, the tracks, who filed the claim and your screenshot to the music distributor that registers our catalog with the platforms, and to the platform itself, never your payment details. We keep these records for as long as we need them to handle the request and as evidence of your license. Ask us, and we will delete your screenshots and files once the request is closed.

What we record

We keep a log of requests. We do this to run the service, to enforce fair-use limits, and to know how the catalog is being used.

What	Why
The search text or link you sent	To run the search, and to improve matching
The video link you asked us to score, and what we measured from it (length, frame rate, shot count)	To run the render, and so a repeat request re-uses the same work instead of downloading your video twice
For analyze_video: the still images, speech transcript, your one-sentence request and a reduced-resolution copy of the video — kept 30 days, then deleted	To review how well our tools served your edit and improve them
Which tool was called, and when	Service operation and troubleshooting
Track IDs returned or requested	To know which tracks are in demand
Which AINSOF listen, view or download link was requested or opened	To understand which links people choose and to troubleshoot delivery. This records the link event, not proof that playback or a download completed.
Client application name and user-agent string	Compatibility and support across Claude, Gemini, Grok, Cursor, ChatGPT and other MCP clients
Country	Licensing territory
If you subscribe or buy a single-track license: payment, subscription and purchase records received from Paddle, the Terms version you accepted and when, and each download (track, version, time, plan or single-track license)	Evidence of your license, Content ID claim releases, accounting and tax
If you send a copyright claim or a Help Center request: what you wrote, the links, your screenshot or files, and for a claim the license details we checked	To handle your request, to get a claim released, and as evidence of your license
A pseudonymous SHA-256 hash derived from your IP address	Rate limiting and request grouping. The raw IP itself is not stored in the AINSOF application log. This hash is pseudonymous data, not anonymous data.

If you hold an API key, we also hold the name and email address you gave us when the key was issued, and a one-way hash of the key. We never store the key itself — it is shown once and cannot be recovered.

What we do not do

• We do not sell or rent your data, and we share it with no one for their own purposes, except that Paddle, as the seller of our plans, also uses payment data for its own legal, tax and fraud-prevention obligations.
• We do not use advertising or analytics trackers.
• This website sets no cookies of its own. The payment form on the checkout page is Paddle's, and Paddle may use cookies or similar technology inside it to process the payment and prevent fraud. The Contact Sales and Help Center request forms show Cloudflare's bot check (Turnstile), which works the same way inside its own box.
• We do not build advertising profiles.
• We do not store your IP address in a form that identifies you.

Where it is held

Request logs and the catalog are held on Supabase infrastructure in the European Union (Frankfurt, eu-central-1). Search requests are processed by systems we operate. To understand a written brief and rank the catalog for it, we send the words of the brief to the AI services of two companies, Google and TypeSafe; a brief document you attach is read by the AI service of Anthropic or Google. They receive only that text or document, never your name, email address or IP address. Your video is not sent to them — the measuring and the mixing run on our own hardware.

Who else handles it

These are the only parties who see any part of a request, and each sees it solely to deliver the service you asked for:

Who	What reaches them
Supabase — hosting, database and audio storage	Everything described above. EU (Frankfurt).
Netlify — the address mcp.ainsof.io runs through	Every request passes through it on the way to us. Netlify is a global network, so a request may be handled outside the European Union before it reaches our systems in Frankfurt.
GitHub Pages — serves AINSOF listen and video watch pages	When you open one of those pages, your browser connects to GitHub directly, so GitHub sees your IP address. Nothing about your search reaches it.
Google, TypeSafe and Anthropic — their AI services	The words of a written brief, so we can understand it and rank the catalog for it, and a brief document you choose to attach. Not your name, email or IP address, and never your video.
Paddle: reseller and Merchant of Record for plans and single-track licenses	Only if you subscribe or buy a single-track license. Your payment details, which you enter with Paddle directly, and your name, email address, country and billing address as Paddle asks for them. Paddle processes payments, invoices and sales tax under its own privacy notice, and sends us the order details described above — never your full card number.
Resend: sends our emails	Only when we email you or our team (a receipt, a request confirmation, a claim update, the copy of a request our team receives): your email address and the content of that email.
Google: our team's mailbox	Only if you write to us or send a request: your message, as it arrives in our mailbox.
Cloudflare: the bot check on our forms	Only when you open the Contact Sales or Help Center request form: the signals Cloudflare's Turnstile needs to tell a person from a bot.
Our music distributor and the platform	Only if you send us a copyright claim: the link, the tracks, who filed the claim and your screenshot, so the claim can be released. Never your payment details.
Apple, Deezer, Spotify and YouTube	Only if you search by reference. To work out what record you named, we send that link or that "artist — title" to them from our server, through each one's own public or authorized interface — Apple's and Deezer's published search APIs, Spotify's API under our registered application, and YouTube's oEmbed endpoint, which returns a video's title and channel name and no audio. Your own browser is not involved and your IP is not passed on. One exception: when a written brief names an artist, that name may be looked up in Deezer's public artist search, to find the style it means.

Each of these acts on our instructions and for no purpose of its own — except Paddle, which as the seller of our plans is also responsible for payment data in its own right, and a platform that handles a copyright claim under its own policies. We use no advertising network, no analytics provider and no data broker.

How we reach anything that is not ours

We take nothing from a platform that the platform does not offer. Every request we make to someone else's service goes through that service's own published or authorized interface, and where a platform does not permit something, we do not do it by another route — we refuse and say so.

• We never download a recording. To match a reference we use only the 30-second preview clips Apple and Deezer publish for exactly this purpose, through their public search APIs. We do not download whole commercial tracks from anywhere. The clips are written to temporary storage for the length of the comparison and cleared automatically within hours; they are never added to the catalog and never used to train anything.
• YouTube gives us a title, never audio. A YouTube link is read through oembed, the endpoint YouTube publishes so other sites can embed its videos; it returns the video's title and channel name. That is the entire extent of our contact with YouTube.
• Spotify is read under our own registered application, with credentials issued to us, for track metadata only.
• We refuse the platforms that do not allow it. We will not pull a video off YouTube, Vimeo, Dailymotion or Twitch — their terms forbid it whoever owns the video, and we cannot know whose it is. The agent says so and offers the ways that are allowed: a direct file link, Dropbox, Google Drive, or a one-time upload link straight from your machine. A SoundCloud reference is refused too, because SoundCloud publishes no clip we are permitted to analyze; a TikTok link is refused because what TikTok publishes about it is the caption, not the record, and guessing from a caption would be worse than saying no.
• A video reaches us only because you sent it. We fetch nothing you did not hand us, we do not follow a link into a private network, and we never go looking for anything beyond the one address you gave.

How long we keep it

Request logs currently have no fixed automatic deletion date and may be retained indefinitely for operating and improving the service, unless deletion is requested. Account details for API-key holders are kept while the key is active and removed on request. Payment and subscription records, the record of which Terms version you accepted, and download records are kept for as long as the law requires us to keep accounting and tax records, and for as long as they are needed as evidence of a license that is still in force.

Your rights

Email info@ainsof.io and we will tell you what we hold about you, correct it, or delete it. If you hold a key, we will revoke it and remove the associated record. Records we must keep by law, or as evidence of a license you hold, are kept but used for nothing else. We will answer within 30 days.

What we never collect

AINSOF does not collect, solicit or process payment card data (you enter it with Paddle, never with us), health information, government identifiers, or any access credentials, passwords or API keys for other services. An AINSOF key is stored only as a one-way hash, as described above. We do not request location data. We do not read, reconstruct or infer your chat history — the agent sees only the brief, link or track ID you send with a request.

Support

For any question, request or complaint about this policy or the service, contact info@ainsof.io, or send a request from the Help Center. We aim to answer within two business days.

Children

AINSOF is a professional music licensing service and is not directed at children under 16.

Changes

If this policy changes we will update the date at the top of this page. Material changes will be communicated to active key holders and subscribers by email.