Snapshot 279597
Normalized text
Scripts and page chrome removed; this is what change detection compares.
Skip to content
DATA PROCESSING AGREEMENT
Data processing agreement
When an organisation buys VoxisLive for its team, the organisation is the data controller and VoxisLive is the data processor. This page is the template we sign, published so your legal and security people can read it before anyone asks you for a purchase order.
Read this first. The text below is a template, published for review. It is not the executed agreement. If you need a data processing agreement in force, ask us for a signed copy through the enterprise channel — the version both parties sign is the one that governs, and it overrides anything written on this page.
Roles: who is the controller, who is the processor
In enterprise use the arrangement is the ordinary one. Your organisation is the data controller: you decide which meetings and which audio get translated, who in your team is given a seat, and how long you keep the transcripts that land on your people's machines. VoxisLive is the data processor: we process on your documented instructions, which in practice means we run the service you bought and nothing beyond it. The processor is VoxisLive Limited, a company registered in England and Wales (company no. 17500296), registered office 71-75 Shelton Street, Covent Garden, London WC2H 9JQ, United Kingdom. If your members use live caption sharing, the same split applies to the captions they share, with two exceptions that stay ours as controller; see the annex "Live caption sharing".
One consequence is worth stating plainly: the people whose voices are translated are frequently not your employees. The other side of a call is a data subject too, and the lawful basis for translating what they say is yours to establish, not ours.
For an individual consumer account there is no controller/processor split in this sense — that relationship is governed by the Terms and the Privacy policy. The DPA is for organisations.
Subject matter, duration, and the data involved
Subject matter and nature of the processing. Real-time speech translation: capturing audio on the user's own machine, transmitting it to a translation engine, and returning translated speech and text. Plus the account and usage records needed to run a paid service — who has a seat, and how many minutes were used. Optionally, relaying the live translated captions of a meeting to the people a member shares a link with (live caption sharing; see the annex).
Duration. For as long as the organisation plan is live, plus the retention periods described under deletion below. Processing of any single session's audio lasts as long as the session does.
Categories of data subjects. Your members who hold a seat; and any person whose speech is captured during a session — the other participants in a meeting, or the speakers in whatever media is being translated. Where a member uses live caption sharing, the same people are data subjects of the shared captions; the listeners who read them are data subjects of ours (see the annex).
Category Examples Where it is processed
Account data Sign-in e-mail, licence tier, the organisation a licence is linked to Our own server
Usage records Minutes consumed, session start and stop, the plan they draw from Our own server
Session audio The speech captured during a live session, transmitted while the session runs The translation engine provider for the target language — see the sub-processor list
Transcripts The text of a session, written to the user's Documents folder, one folder per session The user's own machine
Shared captions (optional, live caption sharing) The translated caption lines of a meeting, while a member shares them Our own server in the United Kingdom, in memory only and not stored; for extra languages, the text-translation provider on the sub-processor list
Problem reports A scrubbed application log and, if the user ticks the box, a transcript they choose to attach Our own server
Payment data Billing details for the purchase itself Our payment provider, which acts as merchant of record
Special categories. We do not ask for special-category data and the product has no feature that solicits it. But live speech is live speech: if your people translate a clinical conversation, health data passes through the pipeline because the meeting contained it. That is a decision for your controller assessment. We have no BAA and make no healthcare-specific claim — see the trust center.
Sub-processors, and what happens when the list changes
We publish our sub-processors by name, with what each one is used for, on the sub-processor page. It is not a category list — you can read the actual vendors and take them to your own review board.
Two points a reviewer normally asks about. Which provider receives the live audio depends on the target language, because engine routing is per target; the sub-processor page says which is which. And when the list changes, we update that page and notify enterprise customers, so a new processor does not appear in your supply chain without you hearing about it.
The website assistant on voxislive.com is a separate, optional surface: questions typed into it go to Anthropic, listed on the same page. It handles no customer audio and no account data.
Security measures
These are the controls we actually implement. We would rather describe a control you can verify than show you a badge.
Encryption in transit. Traffic between the application and our server, and between the application and the translation engine, runs over TLS.
Keys encrypted on the device. If your organisation uses its own engine API key, that key is stored encrypted at rest on the machine using Windows DPAPI, bound to the account and the install.
Work done on the device, not sent anywhere. Speech detection and speaker-change detection run on the user's own CPU. Transcripts are written to the user's own disk, not to us. Live caption sharing, if a member turns it on, relays captions through our server in memory and does not store them.
Problem reports are scrubbed before they leave. Redaction of tokens, keys, e-mail addresses and user paths happens on the device, and it is fail-closed: if scrubbing does not complete, the report is not sent.
Access. Administrative access to the server holding account and usage records is limited to the people who operate the service.
Verification. Part of the audio path is published as a read-only source excerpt, so a security reviewer can check the claims above against code rather than prose — see licensing for what that excerpt is and is not.
What we do not claim. We hold no ISO certification, no SOC 2 report, no third-party attestation, and no BAA. We will not put a compliance badge on this page that we cannot back. If your procurement process requires one of those documents, we do not have it, and it is better that you learn that here than three weeks into a review.
Data-subject rights and deletion
As processor, we assist you in answering the requests you receive: access, rectification, erasure, restriction, portability and objection. Route the request to us through your enterprise contact and we work it with you rather than answering your data subjects directly.
Much of what people ask for is not in our hands at all, which is worth knowing before you draft a response. Transcripts sit in the user's own Documents folder, one folder per session — deleting those is something the member does on their own machine, and no request to us is needed. Session audio is not kept as a file for us to go and find.
Send the request through your enterprise contact, identifying the account or the licence it concerns.
We confirm receipt and tell you what we hold for that account — typically account data and usage records.
Account data is deleted or anonymised within 30 days of the request. Usage records that must be retained for tax and accounting are anonymised rather than removed, so the figures still add up without naming a person. One record is not removed by this route: if the member confirmed the live caption sharing notice, the record of that confirmation is kept for six years after the account is deleted, without the email address. We keep it as controller, for legal claims; see the annex.
We confirm back to you in writing, so you have something to close your own case file with.
Deletion of an individual consumer account follows the same 30-day route, described in the Privacy policy.
Where processing happens — and the limit we will not paper over
We do not guarantee EU-only data residency. The translation engines are operated by large cloud providers and we do not contractually pin a session to an EU region. If your requirement is that no personal data ever leaves the EU, VoxisLive does not meet it today, and no clause we could sign would change the underlying routing.
Our own server, which holds account and usage records and relays shared captions, is in the United Kingdom.
If residency is a hard requirement, tell us at the start of the conversation — it is a short call either way. International transfers otherwise rely on the standard mechanisms, and the providers involved are named on the sub-processor page so you can check their own transfer terms directly.
Annex: Live caption sharing
This annex is part of the data processing agreement between your organisation (the controller) and VoxisLive Limited, a company registered in England and Wales, company no. 17500296 (the processor, "we"). It covers one optional feature, live caption sharing, when it is used from a seat under your organisation plan. Where this annex and the rest of the agreement differ about this feature, this annex applies. "Data protection law" means the UK GDPR and the Data Protection Act 2018 and, where it applies to your processing, the EU GDPR (Regulation (EU) 2016/679).
A1. What the feature does
A member in a Meeting-mode session of the desktop app can let other people read the live translated captions of that meeting on their own phones or computers, through a link, a QR code or an 8-character code (voxis.live/<code>). Readers ("listeners") need no account. Sharing is off unless the member turns it on, and each share is started by the member.
A share carries the translation of what the member says. It can also carry the translation of what the other people in the meeting say, but only if the member confirms this at the start of that share; without that confirmation the other side's words are not sent to our live-sharing server. Audio is not shared, and the original-language text is not shared. The member's name, email address and account details are not shown or sent to listeners.
The member can offer listeners up to 5 more languages to read in. For those languages the app on the member's device sends the text of the shared captions (completed sentences only) to an AI text-translation service and passes the translations on.
A2. Roles
You are the controller, we are the processor, for the shared caption content: the caption lines (the translation of what your member and, where confirmed, the other people in the meeting say), the language of the captions, which side of the meeting a line comes from, and a speaker number (S1, S2) that a line from the other side can carry. Your member acts under your authority when they start a share. Making the captions available to listeners outside the meeting is a disclosure that you make, through your member; we carry it out on your instructions.
We are a controller in our own right, not your processor, for the following, because we process it to run and protect the service and to meet our own accountability duties, not on your instructions:
the technical data of listeners: IP address, a random reconnect ticket (we hold only a one-way fingerprint of it), the language a listener reads in, a random two-digit waiting number and the member's decision when approval is on;
the record that a member confirmed the sharing notice (section A8);
the share log and the usage diagnostics (section A3, last two rows);
the sign-in and plan check that decides whether a seat may share.
How we handle that data is described in our Privacy Policy, section 12.
A3. The processing
Subject matter and nature. Receiving caption lines from the member's app, holding them briefly in memory, and relaying them to the browsers of the listeners who hold the link or code; where the member offers extra languages, having the caption text translated and relaying the translations.
Purpose. To let the people your member chooses read the meeting's captions in real time, in the language of the captions or in one of the extra languages offered.
Duration. Each share lasts until it ends, and the caption content is processed only during that time. A share ends when the member stops it, stops the meeting, closes the app or signs out, when the app has not contacted the server for about 10 minutes, or after 8 hours at the latest. The annex applies for as long as your organisation plan is live and the feature is switched on for your seats.
Data subjects.
your members who hold a seat and start a share;
the other participants in the meeting, whether or not they belong to your organisation and whether or not they use VoxisLive, where the member confirmed sharing the other side;
any other person whose speech reaches the captions: someone speaking near the member's microphone, sound from the member's speakers reaching the microphone, or other speech playing on the member's computer during the meeting;
any person who is talked about in the meeting;
listeners (for their technical data we are the controller, see A2).
Categories of data.
Data Who is controller Where it is processed How long
Shared caption lines in each offered language, with caption language, side of the meeting and, for the other side, a speaker number You Our live-sharing server in the United Kingdom, in memory only Up to the last 30 completed lines per language and the lines being translated at that moment, until the share ends. Not written to disk or to a database, so not in backups
Caption text sent for translation into extra languages (only if the member offers them) You The AI text-translation service (sub-processor, section A5) Not kept by us. The provider logs requests and responses for 55 days for abuse detection only and does not use them to train its models
Listener technical data: IP address, ticket fingerprint, chosen language, waiting number and decision We Our live-sharing server, in memory only IP address for rate limits: minutes; with the ticket: until the share ends. Everything else: until the share ends. The live-sharing service does not write the IP address to its logs
Share log: that a share was opened and closed, its caption languages, which sides were shared, whether approval was on, number of extra languages, duration, highest number of listeners, counts of caption updates and of join requests accepted, declined or timed out. No caption text, no codes, no tickets, no IP addresses, and it does not name the account We Our server's system log Up to 30 days
Usage diagnostics reported by the member's app: that the feature was used, whether a share was paused, whether the other side was included, listener and request counts in ranges. Never the captions We Our own server, with the member's account As for other usage records
Record of the member's confirmation of the sharing notice We Our own server Section A8
Special categories. We do not ask for special-category data. As with the rest of the service, a meeting contains whatever is said in it; if special-category or confidential information is spoken while a share is on, it is in the captions and reaches the listeners. Whether a given meeting may be shared is a decision for you as controller.
A4. Your instructions
You instruct us to process the shared caption content as described in this annex whenever a member of your organisation starts a share from a seat under your plan. A member starting a share, choosing whether to include the other side, offering extra languages and admitting or declining listeners are each an instruction given on your behalf. We process the caption content only on those instructions, unless the law requires otherwise; in that case we tell you first, unless the law forbids it.
If you do not want your members to share, tell us (section A7) and tell your members. If we think an instruction breaks data protection law, we will say so.
A5. Sub-processors and transfers
For the shared caption content we use two sub-processors, both already named on the sub-processor page:
Sub-processor What it does for this feature Data Location
IONOS Cloud Ltd (hosting of our own server) Hosts the server that relays the captions. We administer the server ourselves The caption lines while they are in the server's memory United Kingdom
Google (Gemini API) Translates the caption text into the extra languages a member offers. Used only if the member offers extra languages, and only for a language that at least one listener is reading Caption text only. No audio, nothing about the listeners, no account details Any country where Google or its agents have facilities; the data can leave the United Kingdom and the European Economic Area
You give us general authorisation to use these sub-processors for this feature. The notice and update rule of the main agreement applies: when the list changes we update the sub-processor page and notify you at the contact address on your plan. We have a written contract with each sub-processor that places data protection duties on it, and we remain responsible to you for what they do with the caption content.
Transfers. The live-sharing server is in the United Kingdom. Caption text sent for translation into extra languages may be processed outside the United Kingdom and the EEA; that transfer relies on the transfer terms in our agreement with that provider. As the main agreement says, we do not guarantee UK-only or EU-only processing. If no extra languages are offered, nothing is sent to that provider for this feature.
Listeners can be anywhere in the world. Sending captions to a listener in another country is part of the disclosure your member makes on your behalf; it is not a transfer to a sub-processor of ours.
A6. Security measures for this feature
These add to the measures in the main agreement and describe what is implemented today.
Encryption in transit. Captions travel over an encrypted (TLS) connection from the member's app to our live-sharing server and from there to the listeners' browsers.
No storage. The live-sharing service holds captions in memory only, does not write them to disk or to a database, and removes every language from memory when the share ends. A listener who joins late does not receive what was said before they joined.
Access by link or code. A share is reached through a link, a QR code or an 8-character code. A code that has closed is not issued again for 24 hours. Each listener gets a random 128-bit ticket; the server keeps only a one-way fingerprint of it.
Optional approval. The member can switch on "ask me first" (off by default). A person who opens the link then receives no captions in any language until the member lets them in. The member sees a random two-digit number and the waiting time, nothing else; this is not an identity check.
Limits. Up to 100 listeners per share. Limits apply per IP address to the speed and number of connections, to repeated wrong codes (followed by a temporary block) and to repeated declined requests. A share ends by itself after about 10 minutes without contact from the app, and after 8 hours at the latest.
The other side needs a confirmation every time. The other participants' words are shared only after the member confirms, at the start of that share, that everyone in the meeting knows. The confirmation is not remembered from one share to the next.
Confirmation before first use. A member cannot open a share until they have ticked the responsibility box of the in-app notice and that confirmation has been recorded on our server (section A8).
Microphone safeguard. The app can pause the sharing of the member's side when it detects sound from the speakers in the microphone. This is a safeguard, not a guarantee: it does not catch every case and does not detect another person speaking in the room.
Logs without content. The share log holds no caption text, codes, tickets or IP addresses and does not name the account.
Listener page. The caption page sets no cookies and loads no analytics or advertising scripts.
What these measures do not do. Anyone who has the link, the QR code or the code can read the captions while the share is on, and can copy the text or take a screenshot. We do not know who the listeners are and cannot control what they do with what they read. We do not review the captions.
A7. What you are responsible for
As controller you are responsible for:
the lawful basis for sharing what is said in your meetings, including what participants who are not your members say;
telling the people concerned. Everyone in a meeting must be told, before or when a share starts, that what is said is being translated and that the translation is being shared with others; and their permission must be obtained wherever the law or the circumstances call for it (for example where everyone's agreement is required before a conversation is passed on, or where the meeting is confidential). The app asks each member to confirm this; we rely on that answer and cannot check it;
your own rules. Deciding whether, for which meetings and with which audience your members may share, telling your members, and meeting any workplace information or consultation duties that apply to you. Under our Terms of Service (section 11) a member who uses a seat under an organisation plan must follow the organisation's rules about sharing its meetings;
who gets the link. The member chooses who receives the link, and whoever receives it can pass it on.
Switching the feature off for your organisation. You can ask us at any time, in writing through your enterprise contact or at [email protected], to switch live caption sharing off for the seats under your plan. We act on the request without undue delay and confirm in writing. There is currently no setting for this in the organisation panel and no view there of whether members have shared; we make the change on our side.
A8. The record of a member's confirmation
Before a member's first share, and again whenever the in-app notice changes, the app shows the notice and asks the member to tick a box confirming that they are responsible for what they share, will tell the people in the meeting and will get any permission that is needed. Sharing does not start until that confirmation has been recorded on our server.
The record holds: the member's account and licence identifiers, the version of the notice, a fingerprint (SHA-256) of the notice text in the language it was shown in, the version of the Terms in force, the date and time (server time), the app version, the distribution channel and the app language. It holds no captions, no share code, no IP address, no device identifier, no name and not the member's email address. It does not hold an identifier of your organisation. The box a member ticks at the start of a share to include the other side's words is not recorded on our server.
We keep this record as a controller, for our legitimate interest in being able to show what was agreed and in establishing or defending legal claims. We keep it for as long as the member's account exists and for six years after the account is deleted, and longer only while a claim or complaint about a share is open; after that it is destroyed at our next routine clean-up, within six months. When the account is deleted we add to the record a one-way, keyed fingerprint (HMAC) of the member's email address and the date of deletion, so that the record can be matched to a person only if that address is given to us.
This is an exception to the 30-day deletion in the main agreement. A request from you to delete a member's account data does not remove this record before the six years have passed. The record shows that the member confirmed the notice. It is not a statement by your organisation and does not bind your organisation.
A9. Requests from data subjects
We help you answer the requests you receive, as the main agreement describes: send the request through your enterprise contact, identifying the account or licence it concerns.
For this feature there will usually be nothing for us to find. We do not keep the captions, and after a share ends we keep nothing that identifies a listener or a meeting participant. The share log does not name the account. If a participant asks you what was shared about them, the answer rests with the member who shared and with what your listeners kept, not with data we hold.
Requests about data for which we are the controller (the confirmation record, listener technical data) are ours to answer. If such a request reaches you, pass it to us; if a request that is yours to answer reaches us, we pass it to you without undue delay and do not answer it ourselves.
A10. Personal data breach
If we become aware of a personal data breach affecting shared caption content that we process for you, we notify you without undue delay at the contact address on your plan. We tell you what we know at that point: what happened, the kind of data and roughly how many people are concerned, the likely consequences, and what we have done or propose to do; we add to this as we learn more. We help you meet your own duty to notify a supervisory authority or the people concerned. Deciding whether to notify them is yours as controller.
Captions exist on our server only while a share is on and only in memory, which limits what a breach on our side can expose. A link or code passed on by a listener, or a listener copying what they read, is not a breach of our systems; if you learn of misuse of a link, the member can end the share at once, and you can report it to [email protected].
A11. Other assistance, and when the feature ends
We give you the information you reasonably need to show that this processing meets data protection law, and we help with a data protection impact assessment for your use of the feature where you ask for it; the facts in this annex and in the Privacy Policy, section 12, are the basis.
When a share ends, the caption content is removed from our server's memory; there is nothing to return. When the feature is switched off for your organisation or your plan ends, no caption content remains with us. The record in section A8 stays as described there.
We may change, limit or withdraw the feature, and we may end a share or switch the feature off for an account if we believe it is being misused, if we receive a complaint that appears well founded, to protect the service, or if the law requires it. If we change how the feature processes the caption content in a way that matters to you, we update this annex and tell you.
Getting a signed copy
Write to us through the enterprise page or contact form with the legal entity name, the address, and the contact who will sign. We return an executable copy of this text. If your legal team has a preferred form of DPA, send it — we read it, and we say which clauses we can sign and which we cannot, rather than signing something we would then breach.
Seats and pooled minutes are set up on the same channel; see enterprise plans for how the pool works and pricing for the rest.
What this page is
A template, published so it can be reviewed before it is requested. It is not legal advice, it is not in force by being read, and it is not a substitute for your own assessment as controller. The document both parties execute is the agreement that governs; where it differs from this page, it wins.
FAQ
Common questions
01Do you have a DPA we can actually sign, or is this just a web page?
There is a signable agreement. This page is its text, published so your legal and security reviewers can read it without asking first. Request the executable copy through the enterprise channel with your entity name and signatory, and we return it.
02Can you commit to processing our data in the EU only?
No. We do not guarantee EU-only data residency. The translation engines run on large cloud providers and we do not pin a session to an EU region. If that is a hard requirement for you, it is better to know now than after signature.
03Are you ISO 27001 certified or SOC 2 audited? Can you sign a BAA?
No to all three. We hold no certification, no third-party attestation and no BAA. What we offer instead is a named sub-processor list, described controls, and part of the audio path published as a read-only source excerpt so a reviewer can check the claims against code.
04Will you tell us if you add a new sub-processor?
Yes. Sub-processors are published by name on the sub-processor page, and when the list changes we update that page and notify enterprise customers, so a new vendor does not appear in your supply chain unannounced.
05One of our employees asked for their data to be deleted. What do you actually hold?
Account data and usage records on our server. Send the request through your enterprise contact and account data is deleted or anonymised within 30 days; usage records kept for tax and accounting are anonymised rather than removed. Transcripts are not part of this — they sit in the member's own Documents folder and they delete those themselves. If that member confirmed the live caption sharing notice, the record of that confirmation stays for six years after deletion, without their email address.
06Do the other people in the meeting count as data subjects?
Yes. Anyone whose speech is captured during a session is a data subject, including participants who are not your employees. Establishing the lawful basis for translating what they say is the controller's job, which in an enterprise deployment is you, not us.