Audio workspace privacy

Updated 5 November 2026

This service is operated by INCE.IO LTD, trading as MAP Audio. This notice describes audio handled by this workspace and supplements the MAP Audio privacy policy.

Using it without signing in

The player, the analysis, PAM, the preset lists and MFX on its built-in loops work without a MAP Audio account. Signed out, ChatGPT may send an anonymous identifier with a request; we use a one-way digest of it, or the IP address when there is none, only to apply request limits, and we store neither.

What you share

When you ask MAP Audio to open an attached recording, ChatGPT supplies a temporary download URL and file details. We copy the audio and its filename into a private Cloudflare R2 bucket so the player can read it. When ChatGPT asks only for an analysis of an attachment, including a comparison or a reference check, our server reads each file from its URL once and does not store it. Maximum file size is 32 MiB. We do not collect payment details.

Signing in

Controlling and exporting from PAM, MFX on your own recordings, mastering and the Music API ask you to sign in with your MAP Audio account. map.audio's sign-in gives us your account ID and which of PAM and MFX it holds a licence for, which we record only as a usage label; it does not decide which tools you can use. It also tells us whether your account's email address is confirmed and, when it is, the address: we pass these only to the MAP Audio Music API, to find your Music API account when you use its tools, and we do not store them or include them in a tool result or usage record. The connection lasts up to 30 days or until you disconnect it in ChatGPT. When ChatGPT opens PAM while you are signed in, our server keeps a one-way digest of your account ID with that PAM for 24 hours, so that only your account can control it.

Access and deletion

Each recording has an unguessable ID. Anyone with that ID or its download link can access it while it is active. Keep links private. Access expires one hour after import. A cleanup task runs every 15 minutes to remove expired copies; a separate one-day storage lifecycle rule provides a fallback if cleanup is delayed. Copies retained by ChatGPT or downloaded to your device are governed separately.

Processing

Waveform and spectrogram display run in your browser. MAP Audio's analysis of tempo, key, loudness, tonal balance, stereo and sections, comparisons and reference checks, and the spectrogram image ChatGPT can ask for, run on our server. Their result, numbers, short notes and that image, goes to the conversation and is kept only in the server's memory so that the player and ChatGPT can share one analysis. The player shares the analysis summary, playback position, selected regions and annotations with the conversation. We do not use uploaded recordings to train models, sell them, or send them to a separate AI-generation service.

What tool results return to ChatGPT

ChatGPT shows each tool result to you and passes it to its model, except the parts for the player only. A result holds only what that request needs:

• Recordings: the filename, size, file type, audio ID and download link. An imported or exported recording’s ID and link stop working after one hour.
• Files from ChatGPT’s file viewer: the name, the file type and the file reference that ChatGPT supplied.
• Analysis: the recording’s format and length, its tempo, key, loudness, tonal balance, stereo and section measurements, short notes about them, and MAP Audio presets that fit it; with a comparison, how the two files differ; with references, where the mix leaves them; with a spectrogram, the image and its levels over time.
• Presets: the names, descriptions, categories, tempos and variation counts of MAP Audio’s own presets.
• PAM: the preset, variation and tempo, whether it is playing, its position and meter, a random ID for that opening of PAM, whether it can be controlled from ChatGPT, parameter values ChatGPT asked for or set, the cell, name, length and format of a recording loaded into it, an exported file’s name, size and download link, and a fixed sentence when PAM refuses a command.
• CM-FX and CMI: get_cmfx and get_cmi return a cell’s CM-FX effect or CMI instrument program with its name and control values, set_cmfx and set_cmi the name and control values they wrote, and these four and check_cmfx the compiler’s verdict on a program: its problems, warnings and processing cost.
• MFX: the preset and settings used, and the built-in loop when one was played.
• Music API: the presets’ names, authors, descriptions, tempos and tags, a job’s ID and status, each output’s role, Music API file ID, length and sample rate, the render’s seed and render-minutes, a pattern’s genre, tempo and cell roles, what the render’s licence allows, and, when asked, the render-minutes left, the allowance and when the period ends.
• For the player only: the recording’s audio data, one-time links for storing a processed MFX copy or a PAM export and for loading a recording into PAM, and the address on which PAM receives ChatGPT’s commands, made from a random key.

Tool results never include your name, email address, account ID or anything derived from it, licence keys, payment details, or the usage records and tracking data described under Service operation. Whether you are signed in shows only as whether a tool can run.

Files from ChatGPT’s file viewer

In the ChatGPT desktop app, a WAV, FLAC, MP3 or M4A file opened from a conversation is read by the player in your browser through ChatGPT. We receive its name, and we do not store a copy of the file.

MFX

When you ask MAP Audio to process a recording with MFX, your browser reads the stored copy, loads MFX's engine and the preset from map.audio and renders the result itself. The processed copy is then stored in the same private bucket, under its own ID, and expires after one hour like an imported recording. A preset played on one of MFX's built-in loops renders in your browser, and nothing is stored.

PAM

When ChatGPT opens PAM, your browser loads PAM’s interface, audio engine, presets and samples from map.audio. PAM runs in your browser. The preset, tempo, variation and whether it is playing are shared with the conversation; the sounds and patterns that you play are not sent to us unless ChatGPT exports them. While PAM is open and you are signed in, it keeps a connection to our server, so that ChatGPT can change its tempo, variation, preset, transport and parameters: those commands and PAM’s replies pass through without being stored. When ChatGPT exports audio or a preset, your browser records or saves it and uploads it once to the same private bucket, where it expires after one hour. When ChatGPT loads a recording into a cell, your browser downloads the stored copy once through a one-time link.

Music API

The Music API tools send your request to the MAP Audio Music API, which acts for your MAP Audio account under the Music API terms you accepted. Renders it makes are kept by the Music API under its own retention, outputs for 7 days on the free tier (Music API docs), as well as in this workspace’s copy, which expires after one hour like any other recording.

Service operation

Cloudflare processes requests and stores audio for us. IP addresses and account IDs are used for request limits; IP addresses are not stored with usage records. To find faults and understand use, we record usage events: which tool or page was used, whether it worked, error categories, timings, whether the request was signed in, the ChatGPT surface (such as web or desktop), the platform ChatGPT reports (such as iOS or Windows), which of PAM and MFX the signed-in account holds a licence for, and the country Cloudflare reports. A request made while signed in also carries a pseudonymous account code, a one-way hash of the account ID, so we can count how many accounts use the workspace and how often they return; it is never included in a tool result. The players also report anonymous events, such as when the player opens, when PAM has loaded, when the player analyses or plays a recording, or when an upload fails. PAM’s events carry a random ID made each time it opens, so we can count how many openings reach sound; it is not saved in your browser or linked to you. These records contain no audio, filenames, download URLs, audio IDs, email addresses or messages. Our application logs contain the same categories and cleanup counts. Cloudflare’s privacy information describes its own security processing. No advertising trackers or analytics cookies are installed by this workspace.

Contact

For privacy questions or to request earlier deletion of a stored recording, email support@map.audio. Include its audio ID only if needed to identify the copy; do not send private recordings or passwords.