AI Use Statement
Last updated: August 14, 2026
01Summary
CueSyncuses automated systems in two places, and it is worth being precise about which two, because "AI" is used loosely enough in this industry that the word alone tells you nothing:
- The support chat on this website answers with a language model, operated for us by ClientFlow. It tells you so before you type.
- The AI Music Analysis add-on is an optional one-time purchase that unlocks model-backed analysis inside the desktop application.
The core show engine is not one of them. Cue detection, beat tracking and the timing that runs a show are deterministic signal processing, not a model, and nothing about how your show behaves is decided by a language model.
We do not train models on your content, we do not use a model to decide anything about your account, and where a fact on this page depends on a third party's configuration rather than our own code, it is marked "To be confirmed with ClientFlow" instead of guessed.
02The Support Chat Assistant
The chat widget in the corner of this site is an automated assistant. It is operated for us by ClientFlow, our support and customer-relationship platform, and it composes its replies with a large language model.
The transparency disclosure
Article 50(1) of the EU AI Act requires that a person interacting with an AI system is informed of that fact, unless it is obvious to a reasonably well-informed person. The widget satisfies this properly rather than nominally, and this is the part of our AI handling we are confident is done right:
- The disclosure is displayed in the chat panel before you type, not buried in a policy or revealed after the first reply.
- It says what it is and what you can do about it: "You're chatting with an automated assistant. Ask for a human anytime."
- It is localised to the visitor's detected language by the server and refreshed with each reply, so a non-English speaker gets the disclosure in a language they read — which is the point of the obligation, and the part that is usually missed.
- It renders from a built-in default if the server's localised version cannot be fetched, so an outage cannot leave the widget silently undisclosed.
03What Happens to What You Type
The path a chat message takes is short, and it does not run through our servers:
- Your browser sends the message directly to ClientFlow, along with your IP address. It does not pass through CueSync's API.
- ClientFlow stores it as part of a conversation record, removes recognisable personal details, and sends the message on to a third-party language-model provider to compose a reply.
- The reply comes back to your browser through ClientFlow.
- A copy of the conversation is kept in your own browser for 24 hours so a page reload does not lose it. The "Clear conversation" button in the chat header deletes that copy and the session it belongs to.
- If you choose to leave an email address for a follow-up, it is stored with the conversation and becomes a record in the CRM.
Because the conversation lives with ClientFlow rather than with us, its retention is governed by ClientFlow's configuration. As stated in our Privacy Policy, support chat conversations and CRM records are retained until deleted on request — they have no automatic expiry. Ask us and we will have them deleted.
Practical advice, which is not a disclaimer of our obligations: do not paste licence keys, passwords, card numbers, or a production's confidential material into a chat window. Use the contact form for anything sensitive.
04Model Provider, Training & Retention
These are the questions a serious reader asks about any AI feature, and they are the ones we cannot answer from our own code: the model runs behind ClientFlow's API, not ours. Rather than write a reassuring paragraph we cannot stand behind, here is the actual state of each:
| Question | Answer |
|---|---|
| Which company provides the model? | To be confirmed with ClientFlow. ClientFlow publishes its own sub-processor list, which is the authoritative source; we link to it from the Sub-processors page. |
| Which model? | To be confirmed with ClientFlow. |
| Is chat content used to train the provider's models? | To be confirmed with ClientFlow. Business API tiers at the major providers exclude API content from training by default, but we have not verified which tier and which contractual term applies here, and a default is not a commitment. |
| How long does the model provider retain the message? | To be confirmed with ClientFlow. Providers commonly retain API inputs for a limited abuse-monitoring window; the specific window that applies here is not something we can read off our own systems. |
| Where is the model hosted? | To be confirmed with ClientFlow. |
If you need any of these answered before you use the chat — a venue's IT review, a procurement questionnaire, a data protection impact assessment — write to privacy@cuesync.live. We will get the answer from ClientFlow, send it to you, and publish it here.
05Removing Personal Details Before the Model Sees Them
ClientFlow strips recognisable personal details — phone numbers, email addresses and card-like number sequences — out of a message before it is sent to the model provider. That is a real mitigation and we describe it in the Privacy Policy.
We want to be straight about its limits, because pattern-based redaction is the kind of control that is easy to over-claim:
- It runs on ClientFlow's servers, not ours. We cannot inspect the message after redaction, and there is no CueSync-side test that proves a given message was redacted before it left.
- How the redaction is validated — the rule set, its test coverage, and how a miss is detected — is to be confirmed with clientflow.
- Pattern matching cannot catch what has no pattern. A name, a venue, a production title, an unusual rig description, or a sentence that identifies someone by context will pass through it. Treat the redaction as reducing risk, not eliminating it.
This is the honest reason the advice above is "do not paste sensitive material into the chat" rather than "it is safe to, we strip it".
06Reaching a Human
You can leave the assistant at any point. There are three routes, and none of them requires you to argue with a bot first:
- Ask in the chat. The disclosure invites it explicitly. Ask for a human and the widget offers you the handoff form.
- Take the handoff the assistant offers. When the assistant cannot answer — or when its knowledge base is unreachable, which it distinguishes and says so rather than blaming your question — it offers a short form that reaches the team by email. Sending it does not require an account.
- Skip the chat entirely. The contact form and contact@cuesync.live both go to people.
Timing
A handoff or a contact-form message reaches a person, not a queue that an assistant answers on our behalf. We publish a target of a response within 24 hours on our support and contact pages, and refund requests are answered within 2 business days under the Refund Policy.
We do not operate a 24/7 support desk and do not promise one. There is no service-level commitment attached to these targets; if you need a contractual response time for a production, ask us before you rely on one.
07Automated Triage of Tickets
When you send a contact form, a bug report, a feature request or a support ticket, it is recorded in ClientFlow, which runs its own automated triage over the incoming record — categorising it and prioritising it in the queue a human then works through.
Triage sorts the queue. It does not decide anything about you, it does not close your ticket, and it does not compose a reply that goes out under our name. Every substantive answer to a ticket is written by a person.
08The AI Music Analysis Add-On
AI Music Analysis is an optional one-time purchase that unlocks model-backed music analysis inside the CueSync desktop application, upgrading the deterministic signal processing the engine otherwise uses. It is an add-on, not an edition: it changes nothing about which protocols or features your licence includes.
The analysis is described in the entitlement contract between the website and the desktop application as local, model-backed analysis — a model that runs on your own machine, on the audio already loaded there. On that design, your audio is not uploaded to us or to anyone else in order to be analysed, and the only thing that crosses the network is the entitlement check that confirms you bought the add-on.
We have not re-verified the desktop implementation as part of writing this page, and this website cannot observe what the desktop application does with a local file. If you need "no audio leaves the machine" confirmed in writing for a venue or a client, ask us and we will confirm it against the shipped build rather than against this paragraph.
Buying the add-on, and a refund revoking it, follow the ordinary Refund Policy.
09Where We Do Not Use AI
A statement about AI use is only useful if it also says where AI is absent. It is absent here:
- Your show content is never training data.Nothing you upload to Studio — audio, cues, comments, specs — is used to train any model, ours or a third party's. The licence you grant us in the Studio Terms does not permit it.
- The show engine is deterministic. Beat tracking, cue firing and timecode handling are signal processing and arithmetic. A model does not decide when a cue fires.
- The waveform data Studio computes is not AI. It is a decode-and-summarise pass over your uploaded audio, done so the editor can draw the waveform and read duration.
- No AI-written policy or product copy is passed off as fact. The documents in this legal section are written and reviewed by people against what the software actually does.
10Automated Decisions About You
We do not make decisions producing legal effects concerning you, or similarly significantly affecting you, based solely on automated processing — the thing Article 22 of the GDPR is about.
The place this could plausibly happen is licence anti-piracy. Automated detectors do score signals such as an unusual number of devices, repeated activation handoffs, dispersed network locations, or a clock that moved backwards. Those scores populate a review queue that a person works through. The policy layer that would turn a score into an automatic suspension exists in the codebase but is not connected to anything and does not run; suspending a paying customer's licence on a heuristic is the one failure this product cannot afford, and the design starts with a warning and a self-confirmation link rather than a lockout.
If your licence is ever restricted, a person decided that, and you can contest it by writing to contact@cuesync.live.
11Accuracy & What the Assistant Cannot Do
The support assistant can be wrong. It answers from a knowledge base about CueSync and it can misread a question, miss a recent change, or state something with more confidence than the underlying source supports.
- An answer from the assistant is not advice, and it is not a contractual statement. Where it conflicts with the pages in this legal section, these pages govern.
- The assistant cannot access your account. It cannot see your subscription, issue a refund, reset a licence, or change anything. Those all go through a person or through your account dashboard.
- Do not use it as a check on a show-critical configuration the night before a get-in. Test the rig.
12Contact
Questions about this statement, or a request to have the "To be confirmed with ClientFlow" items answered for your review: privacy@cuesync.live.
Related documents: Privacy Policy, Sub-processors, and the Studio Terms.