Commercial use, privacy, and data
The answers evaluators and procurement teams ask for before building on Flat's Optical Music Recognition API.
Commercial use
Can I use OMR commercially and resell the output?
Yes. You can use the OMR API to process sheet music uploaded by customers who own the content or otherwise hold the necessary rights, and provide the resulting MusicXML and MIDI files to those customers outside of Flat.
Specifically:
- No attribution is required for the generated files.
- Nothing prevents you from building a commercial transcription product on the API.
- You are not required to keep the result in a Flat library. Create the job with
output: "musicxml"and the result never enters one. See Do I have to create a Flat score? - One account can serve all your end users. They do not need Flat accounts of their own. See Can my backend use one account for all my end users?
As always, you and your users remain responsible for having the appropriate rights to the source material and for complying with applicable law.
Can I store and reuse the MusicXML?
Yes. The output is yours. Store it, reuse it, and serve it to as many of your users as you like. Deduplicating conversions on your side is also the easiest way to reduce credit spend. See Can I cache and reuse the output?
Do I need a custom contract?
Not for standard API usage with credit packs. For larger-volume deployments, custom pricing, or specific contractual requirements, email developers@flat.io and we can put a business agreement in place.
Third-party apps
Do my users need a Flat account?
No, and most apps are built without one. The usual setup is to resell the conversion:
- You hold one Flat account. Your backend calls the API with its personal access token for all your users.
- You buy the credits. Every scan is paid from your account's credit packs, with volume pricing for larger deployments.
- You set your own price. Charge your users per scan, by subscription, or include scanning in your app's price. Flat is not involved in that billing.
- Your users never see Flat. No sign-up, no Flat credits to buy, and no attribution in your app.
Keep the token on a server you control. A token shipped inside a mobile or desktop app can be extracted, and anyone holding it can spend your credits. See Can my backend use one account for all my end users?
Choose per-user Flat accounts, below, only if your users already use Flat or you cannot run a backend: each user then needs a Flat account and their own credits.
Can my app use the OMR API with its users' own Flat accounts?
Yes. An app can have each user sign in with their own Flat account through OAuth2 and run OMR jobs on that account. Native and mobile apps use a public client with PKCE. Each scan is then paid with that user's credits. See Whose credits are spent?
There is one requirement: tell your users what a scan costs before they start it, as Flat's own apps do. Before a scan, show:
- that it uses credits from their Flat account,
- how many:
costPerPagefromGET /omr/capabilitiestimes the number of pages, taken when the job starts and refunded if it fails, - how many credits they have left:
remainingCreditsfrom the same call (theomrscope is enough to read it).
No "Powered by Flat" mention or attribution is required.
Do my users accept Flat's Terms?
Yes. Your users sign in on Flat's own page, and creating a Flat account requires accepting Flat's Terms of Service. For individual accounts, sections 10.6 and 10.7 of the Terms of Service make the user responsible for clearing the rights of anything they scan, and state that Flat does not check scanned content for copyright. Flat for Education accounts are covered by the Education terms.
Who is responsible for the output once my app has it?
Your app and its users. Flat provides the conversion. Once you have downloaded the MusicXML or MIDI, what your app does with it (storing, exporting, or letting one user send a score to another) is governed by your own terms, not Flat's. We recommend that your terms make your users responsible for having the right to scan and share what they do, and that you give rights holders a way to contact you.
Content kept in Flat is different. A job created with output: "library" creates a score in the user's Flat library, and sharing or publishing that score on Flat falls under Flat's Terms of Service, including its rules of acceptable use.
What happens if a rights holder complains?
Flat's notice and takedown procedure (section 12 for individual accounts, abuse@flat.io) applies to content available through Flat.
Output kept in your app is not on Flat. With output: "musicxml", Flat keeps a job's files only until you call DELETE or until its retention.expiryDate (see How long is my data kept?), and once they are erased nothing is left on Flat. A complaint about content in your app is therefore yours to handle under your own terms.
Privacy and model training
Do you train your ML models on my files?
No. Flat's OMR machine-learning models are trained entirely on synthetic data we generate ourselves with our own engraving engine, and on public-domain sheet music we source independently.
Files you submit to the OMR API, and the outputs generated from them, are not used to train or improve our models.
Where is the data processed, and who are the subprocessors?
Two places handle a job's content:
| Where | What it holds | How long |
|---|---|---|
| Flat's object storage | The uploaded files, recognition results and exports, and processing logs | Until you call DELETE, or 30 days at most |
| Modal, the GPU provider that runs our recognition models | The page images, the recognition result and processing logs of each run | Up to 7 days, then Modal deletes them automatically |
Modal processes pages worldwide and keeps this data in the US. Transfers rely on Modal's DPA. The full list of infrastructure providers, subprocessors and hosting regions is in the data and infrastructure article.
Are OMR files sent to OpenAI, Anthropic, or another AI provider?
No. Recognition runs only on our own models, in GPU containers on Modal. The uploaded files, the text extracted from them, and the results are never sent to OpenAI, Anthropic, or any other AI provider.
OpenAI and Anthropic appear in our subprocessor list because they power a different feature, the assistant in Flat's editor. The data and infrastructure article breaks providers down per feature.
Is EU-only processing available?
No. Modal processes pages worldwide and keeps them in the US for up to 7 days, as described above.
Retention and deletion
How long is my data kept?
Jobs created with output: "musicxml" keep their uploaded files and recognition results for 30 days by default, then they are erased.
The window exists for two reasons:
- Re-downloads. You can fetch a finished job's results and exports again, for example after a failed download or to get another format, without paying for a new run.
- Support. If you contact us about a job, its files and processing logs let us replay what happened and find the cause. Once a job is deleted or expired there is nothing left to investigate: only its billing record remains.
Read the exact window from retentionDays on GET /omr/capabilities, and each job's own deadline from retention.expiryDate. The deadline is fixed when the job is created, so a later change to the retention window never moves it.
Jobs created with output: "library" have no retention object at all: their result is a score in the user's library and follows the library's own lifecycle.
Full detail, including the API calls, is in Retention and deletion.
Can I delete a job before the deadline?
Yes, and you should as soon as you have collected what you need:
bash
curl -X DELETE "https://api.flat.io/v2/omr/jobs/$JOB" -H "Authorization: Bearer $TOKEN"The call is idempotent. Downloads for an erased job then return 409 (OMR_JOB_EXPIRED). The same action is the Erase files now button on the job's page in the developer dashboard.
Can I get a shorter retention window?
There is no per-account retention setting: every account has the same 30-day window. DELETE is how you shorten it, job by job. An integration that calls DELETE as soon as it has downloaded the result keeps nothing on Flat's side beyond that call.
The trade-off: a deleted job can no longer be re-downloaded, and we can no longer investigate it if you report a problem with its result. If your users report recognition problems, keep the source file on your side so you can send it with the report, or delay DELETE until your user has checked the result.
What survives deletion?
The originals, recognition results, and processing logs are all stored under a single dedicated prefix in our object storage. At expiry, or when you call DELETE, that prefix and everything in it is permanently deleted. We do not maintain backups or replicas of this data, so once it is gone it is gone.
What remains is a stripped job record kept for accounting and for listing your job history:
| Kept | Not kept |
|---|---|
| Job id, creation and expiry dates | The uploaded files |
| Status the job finished with | Recognition results and exports |
| Page and file counts | Filenames and checksums |
| File size and MIME type | Any document content |
| Billing information | Processing and debug logs |
The job stays listable with retention.expiredDate set, so you can still audit what you ran. Filter on it with GET /omr/jobs?status=done&expired=false.
Does a subprocessor keep a copy after I delete a job?
Yes, for a limited time. Modal, the GPU provider that runs recognition, keeps the page images, the recognition result and the processing logs of each run for up to 7 days, then deletes them automatically.
- Modal sets that window for all its customers. We cannot shorten it.
- It does not depend on
DELETE: deleting a job erases Flat's copy right away, and Modal's copy still expires on Modal's schedule. - Flat has no access to Modal's copy and does not read it back.
Modal's retention is documented in its security and privacy guide.
Terms and agreements
Which terms govern the OMR API?
The OMR API is covered by Flat's Terms of Service and Privacy Policy.
Is a DPA available?
Yes. You can rely on our published Data Processing Addendum for your use of the OMR API, with Flat as processor of the files you send. The document is written for Flat for Education; until terms dedicated to developers are published, it also covers OMR API customers.
For specific contractual requirements beyond it, email developers@flat.io.