Terms of Service

Last updated 3 September 2026. Pincite Audio is in an open beta; §2 explains how these terms are accepted and what they currently govern.

Section 1

Who these terms are between

These terms are between you and the operator of Pincite Audio ("we", "us"). They govern your use of pinciteaudio.com and, if your firm holds an account, your use of the transcription service itself.

If you are using Pincite Audio on behalf of a law firm or other organisation, you confirm you are authorised to accept these terms for it, and "you" means that organisation.

Section 2

What these terms currently do

Pincite Audio is in an open beta, and you can open an account for your firm yourself. Accounts are created in the application at app.pinciteaudio.com/signup, not on this website.

So, concretely: browsing this website is governed by these terms. Your use of the service is governed by the acceptance you give at sign-up, by an affirmative act — you tick a box stating that you accept these terms before an account can be created, and we record when you ticked it and which version of this page was published at that moment, identified by the "Last updated" date at the top. That recorded acceptance, rather than your browsing of this page, is what forms the agreement covering your use of the service. A new firm starts on a one-time grant of free minutes, which expire thirty days after that firm's first upload; past that grant the service is prepaid, at the rate published on the pricing page. You buy minutes in advance and each delivered transcript draws that balance down. Minutes you buy do not expire, there is no invoice, and nothing is charged to you except a purchase you make yourself.

We publish the terms here, outside the sign-up flow, for one reason. An attorney evaluating a vendor for client material has a professional obligation to understand what that vendor undertakes before sending anything, and "terms available after you sign up" makes that impossible. You should be able to read what we commit to before you have committed to anything.

Section 3

The service, and what it is not

Pincite Audio transcribes audio you supply. It produces a transcript, a coverage measurement against the source audio, and a digest record over the produced text and the audio submitted for transcription.

It is not a law firm and does not provide legal advice. It does not review your material for privilege or responsiveness, does not decide what is discoverable, and does not certify that any transcript is admissible, authenticated, or adequate for any purpose in any forum. Those are your professional judgements and remain entirely yours.

A transcript is a machine-produced document that we design to disclose its own gaps. It is not a certified transcript, and no part of the service is a substitute for a human transcriptionist or court reporter where one is required.

Section 4

Confidentiality, and what we do with your material

This section exists because your professional-responsibility rules require it of you. A lawyer engaging a technology vendor is expected to make reasonable efforts to ensure the vendor's conduct is compatible with the lawyer's own confidentiality obligations, and to confirm that the vendor's terms require confidentiality. Ours do.

We will not disclose your audio, transcripts, or matter information to anyone, except to the subprocessors named in section 5, as you direct, or where we are compelled by law. If we are compelled by law, we will give you notice before disclosing unless we are legally prohibited from doing so.

We will not use your material to train models, and we build none. We will not sell it, and we will not use it to advertise to you or anyone else.

Your material is yours. You retain all rights in the audio you supply and in the transcripts produced from it. The limited licence you give us is only the one we need to operate the service for you: to process, store and transmit your material in order to produce and deliver your transcripts.

Technical measures, stated so you can verify them rather than trust them: transcript content is stored as ciphertext under a key held per matter, so backups are ciphertext too; deleting a matter destroys that key. See what the system does and the privacy notice.

Section 4a

What we retain, and why it does not weaken section 4 or section 9

Section 4 says what we will not do with your material and section 9 says how it is destroyed. Neither is worth much unless you can also see what running the service leaves behind. So this section says what we keep, in the four kinds it comes in.

Operational records — what the system did. We keep the state each recording moved through, the error codes it produced, the technical shape of the audio (its container and codec, sample rate, channel count and duration) and the metering that bills you: seconds processed, calls made to the transcription service, and what they cost. Our server logs are the same kind of record, and are held to the same rule — no transcribed text goes into them. The record of the account notices we sent is the same kind again: which notice went to your firm and when, held so no notice is ever sent twice — it stores no second copy of your email address, which is looked up at the moment of sending, and none of the notice's text, which never varies.

Quality measurements — numbers about a recording. Processing produces the coverage ratio, the un-transcribed gap seconds, the seconds of detected speech that did not reach the transcript, word and utterance counts, and per-word confidence scores. These are counts, ratios and durations. No words of yours appear in them, and none of them can be turned back into words.

Your content — the words themselves. Transcripts, summaries, file names, matter names and any speaker names you record exist in our database only as ciphertext under that matter's key. The audio you supply is held as objects in cloud storage. Deleting a matter destroys the key and deletes those objects, which is section 9.

What deletion leaves. Three things, and it is easier to name them than to imply there is nothing. The billing record survives, because we have to be able to account for what was charged — it holds seconds and dollars, and we remove the matter and recording identifiers from it. An activity log survives, holding identifiers, timestamps and the names of actions taken, never text. And the matter itself survives as a marker: your own matter reference, its jurisdiction and its dates remain, so a deleted matter is still recognisable to you, while its name is overwritten with a value no key can decrypt. Where an activity-log entry records a transcript digest — a one-way hash computed over the text, which the certificate also prints — that hash remains; it cannot be turned back into the transcript. The quality measurements above do not survive: they are stored against the recording, and the recording is deleted.

How we improve the service. We work from the operational records and quality measurements described above, and from recordings we obtain for evaluation ourselves — public court audio and licensed speech datasets. We do not develop the product from your audio or your transcripts.

Anything beyond what this section describes — using a firm's own recordings to measure accuracy, for example — happens only under a separate written agreement with that firm. It is never the default, and nothing in these terms grants it.

Section 5

Subprocessors

You cannot assess whether to send us client material without knowing who else touches it, so we name them rather than making you ask.

Google Cloud Speech-to-Text performs the transcription. Audio you submit is transmitted to it in order to be transcribed.
Google Cloud Storage holds audio and derived files during processing.
Google Cloud Secret Manager holds the per-matter encryption keys.
Google Vertex AI (Gemini) produces the short per-recording summary. Once a recording has been transcribed, its transcript text and the speaker labels are sent to it in order to generate that summary. Your audio is not sent to it, and neither are any speaker names — the labels it receives are machine designations such as Channel 1 or a positional Speaker 2, never a name anyone at your firm typed. This runs automatically on every recording that transcribes successfully; it is not something you switch on.
Resend delivers the emails this product sends, where a deployment has email delivery configured. There are nine. Seven go to you: the message asking you to confirm your address when you sign up — the one email we send to an address that has no account yet, on the say-so of whoever typed it into the form; the notice that your account was created; the notice that uploads are open for your firm; the notice that a recording finished processing (ready to review, or flagged for you to check); the account-security notice sent to a user who changes their own password; a notice that your minute balance is running low; and a notice that your free minutes are about to expire. Two go to us, at one address. The first is an internal alert telling the people who run this service that a firm has signed up, has finished its first transcript, or has bought minutes for the first time — so that a human starts setting your firm up promptly rather than when they next happen to look. The second tells us that a firm has reported a problem with one of its transcripts, and whether it asked for the minutes back. It receives the email addresses of the people we notify and the bare fact behind each notice — an address was entered at sign-up, an account was created, uploads opened, a recording finished, a password changed, a balance ran low, a free-minutes window is closing, a firm reached one of those three points, a report was filed. The two balance notices carry one figure each, because the message would be useless without it: how many minutes are left, and the date the free ones stop counting. It receives no audio, no transcript text, and no matter name — the notices are deliberately written without them, and the code that sends them is built so it cannot obtain them. The internal alerts do not name your firm either: each carries an internal record number and nothing else, so your firm's name never leaves our servers on that path. What you write in a report stays on our servers — it is encrypted under your matter's own key like the transcript is, and the notice that a report exists cannot carry a word of it.
Stripe processes payments. When your firm buys minutes, the card details are entered on Stripe's own hosted page and go to Stripe directly — they never pass through our servers, which store only opaque references into Stripe's records. We do not keep a payment method on file; each purchase is its own transaction. Stripe receives what you enter on that page and an internal identifier for your firm. It receives no audio, no transcript text, and no matter name.
Anthropic (Claude) receives transcript text — but only if your firm chooses to connect Pincite to Claude, and only what you ask for. Pincite Audio can be added inside claude.ai as a custom connector. Nothing is sent until you do that, and nothing is sent on a schedule afterwards: each time you ask Claude a question that reaches your matters, the tool call comes to us, we decrypt what that call asks for, and we return it. What goes back is the transcript passages you requested, the matter names those passages belong to, any speaker names your firm has recorded against them, and the coverage and gap figures for the recordings involved. Your audio is not sent — no Anthropic surface accepts an audio file, which is why the connector exists at all. The connector asks for one permission, to read transcripts, and every tool behind it is a read: there is no tool that writes to a matter, edits a transcript, or changes a speaker label. Once a passage is in a Claude conversation it is in your firm's Anthropic account, not ours, and it is governed by whatever agreement your firm has with Anthropic — a question we do not answer for you and would rather you asked them. On our side we record which tool was called and how much came back; the code that writes that record accepts a fixed list of eight fields, every one of them a tool name, a count or a flag, and rejects anything else, so it cannot hold a word of the transcript or of the conversation.

Your audio goes to Google and nowhere else. Your transcripts go to Google — to Vertex AI, for the summary — and, if and only if your firm connects the connector described above, to Anthropic, at your direction and in answer to your own questions. Nobody else in this list receives either one. We remain responsible to you for every subprocessor's handling of what it receives; the Anthropic connector is the one path here your firm switches on itself, and what happens to a passage once it is in your Anthropic account is between your firm and Anthropic.

If you need a data-processing agreement, or the subprocessor's own terms, before sending client material — ask first. That request is reasonable and we would rather answer it than have you assume.

Section 6

Disclaimer of warranties

THE SERVICE IS PROVIDED "AS IS" AND "AS AVAILABLE", WITHOUT WARRANTIES OF ANY KIND, WHETHER EXPRESS, IMPLIED OR STATUTORY. WE SPECIFICALLY DISCLAIM THE IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NON-INFRINGEMENT.

WE DO NOT WARRANT THAT ANY TRANSCRIPT IS ACCURATE, COMPLETE, OR SUITABLE FOR USE IN ANY PROCEEDING. Automatic speech recognition makes mistakes, including on audio that sounds clear. The coverage measurement is a completeness tripwire and not an accuracy measure: it can pass on a transcript that is wrong. You are responsible for reviewing any transcript before relying on it.

WE DO NOT WARRANT THAT A TRANSCRIPT CAN BE REPRODUCED BY RE-RUNNING THE SERVICE. The speech-recognition service exposes no version pinning, so resubmitting identical audio later may produce different text. Reproducibility comes from the stored artifact and its digest record.

NO SERVICE LEVEL OR TURNAROUND TIME IS WARRANTED. Transcription is asynchronous and carries no vendor service-level agreement that we could pass through to you.

Section 7

Limitation of liability

TO THE MAXIMUM EXTENT PERMITTED BY LAW, WE WILL NOT BE LIABLE FOR ANY INDIRECT, INCIDENTAL, SPECIAL, CONSEQUENTIAL, OR PUNITIVE DAMAGES, OR FOR ANY LOST PROFITS, LOST DATA, OR LOSS OF GOODWILL, ARISING OUT OF OR RELATING TO THE SERVICE.

OUR TOTAL AGGREGATE LIABILITY ARISING OUT OF OR RELATING TO THE SERVICE WILL NOT EXCEED THE GREATER OF (A) THE FEES YOU PAID US IN THE TWELVE MONTHS BEFORE THE EVENT GIVING RISE TO THE CLAIM, OR (B) ONE HUNDRED US DOLLARS.

These limits do not apply to our breach of the confidentiality undertakings in section 4, our gross negligence or wilful misconduct, our fraud, death or personal injury caused by our negligence, or any liability that cannot be limited or excluded under applicable law. Some jurisdictions do not allow the exclusion of certain warranties or the limitation of certain damages, so some of the above may not apply to you.

Section 8

Your obligations

You are responsible for having the right to submit the audio you submit, and for complying with the recording, wiretap, consent and evidence-handling laws that apply to you. Rules on recording and on handling correctional-facility audio vary by jurisdiction, and we are not in a position to assess your situation.

You will not use the service to break the law, to infringe anyone's rights, or to attempt to gain unauthorised access to the service or to another customer's material.

Section 9

Termination, and what happens to your material

You may stop using the service at any time, and you may delete a matter at any time. Deleting a matter destroys that matter's encryption key, which renders its stored content unreadable including in backups that already exist. That is not reversible, and we cannot recover the material afterwards — including for you.

We may suspend or terminate access for material breach of these terms, or where we are required to by law. Except where the law requires otherwise, we will give you a reasonable opportunity to export your transcripts before terminating.

Section 10

Changes to these terms

We may change these terms. The date at the top of this page is when they last changed. If a change materially reduces your rights or increases your obligations, we will give notice before it takes effect and it will not apply retroactively to anything that already happened.

We will not apply a changed version of these terms to you by silently editing this page and treating your continued browsing as agreement.

Section 11

Governing law and disputes

These terms are governed by the laws of the State of Washington, United States, without regard to its conflict-of-laws rules. The state and federal courts located in King County, Washington have exclusive jurisdiction over any dispute arising out of or relating to them, and you and we each consent to that jurisdiction.

There is no arbitration clause and no class-action waiver in these terms. If either is added later, it will be by the notice process in section 10, conspicuously, and not quietly.

Section 12

General

If any provision of these terms is held unenforceable, the rest remain in effect and the unenforceable provision is limited to the minimum extent necessary. Our failure to enforce a provision is not a waiver of it. You may not assign these terms without our consent; we may assign them to a successor in connection with a merger, acquisition, or sale of assets, on notice to you.

These terms, together with the privacy notice, are the entire agreement between us about their subject matter.

Questions about these terms belong on the support page.