Your clients' case files are kept on your computer, not on ours. JUDLEX is a desktop application: splitting, foliation, reading and reporting all happen on your machine, on a copy the application keeps in the vault you choose. We receive no copy of your case files, and we have no way of reading them.

There are five exceptions, and this policy sets out every one of them: the licence (section 5), AI-assisted reading (section 6), your account and billing (section 4), only if you turn it on, a usage summary of the application itself (section 7) and, only if you switch it on, JUDLEX AI (section 6.4). With your own key, none of them brings us any data from your case files: AI-assisted reading takes it to your provider, not to us. With JUDLEX AI, what the AI needs to read goes through our server on its way to Anthropic, without staying there.

The application also downloads from our server what it needs in order to work: new versions, the public list of AI prices and, when you ask for them, the voice model for dictating to the chat (section 6.1) or the tool that plays audio and video. Those downloads bring us only what any request from the application carries —your IP address, the version and a header saying it is JUDLEX—: the application receives, it sends no data from your firm, and the audio you dictate never leaves your computer.

• Controller: CONSULTING EVEREST LLC
• Incorporated in: the State of New Mexico, United States — Business ID 7479255, incorporated on 27 December 2023
• Address: 8206 Louisiana Blvd NE, Ste A #2970, Albuquerque, NM 87113, United States
• Contact: support@judlex.com

Anyone who visits judlex.com, opens an account, or contracts and uses a JUDLEX licence. It is not addressed to your firm's clients: as regards the data you process in your case files, you are the controller. Section 8 explains what our role is there: none over your case files, however you configure the application.

We process the browsing data strictly necessary to serve the page and keep it secure, plus whatever you give us if you write through the contact form: your name, your email, the subject you pick and what you tell us.

There is no chat on this website, no analytics and no third-party cookies. Nor is there an assistant that answers before you buy: the one that existed until 25 August 2026 has been withdrawn, and with it the installation fingerprint and the IP address that were stored to cap free messages. They are no longer collected.

• Legal basis: your consent when you write to us; legitimate interest in keeping the site secure and available.
• Retention: correspondence, for as long as the enquiry is live and thereafter for the limitation period of any actions that may arise.

What we process: email address, password —stored encrypted, never in the clear—, the billing details you give us and the organisation you belong to.

What for: giving you access to your account, issuing the licence, invoicing and providing support.

Legal basis: performance of the contract, and compliance with tax and accounting obligations as regards invoicing.

Payment is processed by Stripe. Your card details are entered directly into a Stripe component and never pass through our servers: we neither see nor store them.

Retention: for as long as the relationship is live and, thereafter, for the applicable legal and tax limitation periods.

For the licence to work, the application communicates with our server:

• On activating a seat: the licence key and an identifier of the machine.
• While you work: a periodic heartbeat that renews permission to use it. The application keeps working offline for up to 72 hours, and up to 14 days in travel mode.
• When checking for a new version.

No data from your case files, clients or parties leaves your computer in any of those communications. Nor do any figures on what you spend with your AI key. What goes in each of them is the licence key, an identifier of the machine, the operating system and the installed version.

The machine identifier used by the licence and the pseudonym of the usage summary in section 7 are different and cannot be linked, not even by us: the second is worked out on your computer from a value that never leaves it.

There are two ways of using it. The usual one is with your own key: you configure the key of your own AI provider —Anthropic, OpenAI, Google, xAI, DeepSeek or MiniMax— and the request goes straight from your computer to that provider. We do not step in, we do not see it and we do not log it. The processing relationship is between your firm and that provider, under whatever contract you have with them.

What each task needs travels to that provider, with your key: the text of the pages and, for scanned ones, the image it looks at; when you ask about an email, the email with its attachments. Every request is logged under Settings → History, on your own computer.

The other one is JUDLEX AI, optional and switched off by default, explained in section 6.4. Until 25 August 2026 there was another mode in which the request went out through a server of ours on its way to the provider; it was withdrawn entirely, and the data it relied on went with it: the responses kept so that a retry would not be charged twice no longer exist.

What you spend with your key is recorded on your own computer, so that you can see it: provider, model, tokens and cost of each call, in the application's spending log. That table never leaves your machine and you can empty it whenever you like.

If you turn on the usage summary in section 7, the only thing that travels about artificial intelligence is how many requests you made that day: a number. Not the amount, not the model, not the provider, and certainly not the request.

You can dictate your questions instead of typing them. The audio never leaves your computer: it is transcribed there, with a voice model that works offline, and discarded as soon as the text is written. The text stays in the chat box and only leaves if you send it, like any other question.

To make that possible, the first time the application downloads that voice model from our server (between 141 MB and 1.5 GB, depending on the one you choose). That download brings us the same as any request from the application —your IP address, the application version and a header saying it is JUDLEX— and no data from your firm: the application receives the model; it sends nothing.

Besides the providers on the list, you can add one of your own: a compatible service that is not in our catalogue, or a server of your firm. Before the application makes any call to it —not even to ask for its list of models— we ask you to accept a notice, because the rules change:

• The text and images the artificial intelligence works on go to the service you have configured, with your key. We have not reviewed it: that service processes them under its own terms, and if it is an intermediary it may pass them on to other companies.
• If the server is on your own machine or on your office network, what is sent does not go out to the internet, unless that server forwards it.
• It still does not pass through us: the request goes straight from your computer to the address you entered.

That address and the names of its models are stored on your computer, together with the key, and are not sent to us.

If your papers are not enough to answer, the chat may ask the search tool of your own AI provider —Anthropic or OpenAI—, with your key, to look something up on the internet. The switch «Search the internet when the papers are not enough» comes switched on and is turned off in the chat menu.

• What goes out is only the text of that search: never your papers or the conversation.
• Before it goes out, the application cleans it, and refuses it if it contains an email address, a number plate, an IBAN, a phone number, a figure of five digits or more, a code of letters and numbers, the name of a client or of a party, or two proper names in a row that appear in the papers of that question.
• It is announced beforehand, with the text that will go out, and afterwards we show you what the provider actually searched for. With OpenAI, it is also asked not to store the response.
• Your provider charges for it per search, on top of what the request costs, and it is logged under Settings → History.

The fee quote uses the same route, with the same switch, to look up the average rate for an item that has no price in your fee list: only a general phrase goes out, without the client, the parties, the case number or any amount, and the chat says so before each search.

JUDLEX AI is the application's artificial intelligence with no keys and no contract with a provider: we provide it, with Anthropic's models, and it is paid with the licence's credits. It comes switched off. It is switched on in Settings → Connections and, before the first call, the application asks you to accept a notice that says the same as this section. The accepted version is recorded on your computer and on our server.

• What travels: the same as with your own key —the text each task needs and, from scanned pages, the image it looks at—, but from your computer to judlex.com and from there to Anthropic.
• What we keep: none of the content. Not the text, not the documents, not the answers: they are forwarded and not stored. For each call we record only the licence, the computer, the model, the tokens, the credits spent and the time, because that is what is charged.
• What Anthropic keeps: it receives it to answer, does not use it to train its models and deletes it within 30 days at most. If its safety system flags a request as a possible breach of its usage policy, it may keep it for up to 2 years.
• If the chat searches the internet (section 6.3), the search —only what is searched for, already cleaned— also goes through the search engines Anthropic works with, in the United States.

Legal basis: the contract, to the extent needed to provide the service you switched on. For your clients' data travelling in the requests, we act as your processor (section 8). Retention: the record of each call, for as long as the licence lasts and, after that, whatever tax and accounting rules require; the content, none.

To stop using it, just go back to your own key in Settings → Connections. From then on the request goes straight from your computer to your provider again, and nothing is deleted from your computer.

This is switched off unless you turn it on. The first time you open JUDLEX we ask whether you want to send us a usage summary. Until you say yes, the application does not write a single row of this section on your computer: it is not that nothing is sent, it is that nothing is collected. You can check it, and you can turn it off later in Settings → Data, under «Usage summary», with a switch. When you turn it off, whatever was pending on your machine is deleted and the pseudonym is regenerated, so that if you ever turn it on again it will be a different installation for us.

On the same screen there is a button that shows you the exact text that is going to be sent, as it is, before sending it. There is no long version for us and a short one for you.

One summary per day and per installation, and nothing outside this list:

• A pseudonym of the installation (installId): thirty-two hexadecimal characters that your own application works out from an identifier of your machine and a random value —a «salt»— that stays on your computer and is never sent. Without that salt, the pseudonym cannot be reversed or linked to your account, your licence or your machine, and we do not have it.
• Technical context: JUDLEX version, operating system and its version, interface language.
• What you do for a living, if you stated it in My profile: lawyer, jurist, court representative, accountant, administrative agent, expert or other; and whether you are a firm or a self-employed professional. Not answering is an option, and it is what travels if you leave it untouched.
• How much you used it that day: how many times you opened the application from closed, how many sessions, how many active minutes.
• Which screens you visited and how many times, from a closed list of nine: Clients, Matter, Case file, Search, Chat, Computation, Settings, Profile and News.
• How many times you did each thing, from a closed list of ten: matters created, case files created, documents added, pieces split, pages read, AI requests, exports, deadline computations, backups and searches.
• Which sections you opened that day: Personal, Professional or both. Only which ones, nothing of what is inside.
• How many errors there were that day and, for each error: its type —an unexpected crash, a screen error, an error in an internal operation or a network error—, a code, at which point of our own program it happened, a fingerprint to group repeated ones, and a technical message.
• Your country and city, which we work out from the IP address the request arrives from. The IP address is not stored: it is used for that and discarded on the spot; there is no field where it could remain.

All of the above, except the technical message of an error, consists of numbers and values from closed lists. Anything not on those lists is discarded without being stored.

Not the name of a client, nor the title of a matter or a case file, nor a party, nor a date of a proceeding, nor a fragment of text, nor a file path, nor the name of your computer, nor your email address, nor your user identifier, nor your IP address. It is not that we promise not to look: it is that the field to put it in does not exist, and we have an automatic check that prevents publishing a server version that adds it.

The technical message of an error is the only free text in the whole system, and that is why it is cleaned twice —on your computer before it leaves, and again on arrival— to remove file paths, quoted text, email addresses and runs of four or more digits, and it is cut to 200 characters.

To know which parts of the program are used and which are not, to find out that something breaks without depending on you writing to us, and to decide what we build next. Nothing else. There is no advertising, we do not sell it or give it to anyone, and it is not cross-referenced with your account.

Your consent (art. 6.1.a GDPR), asked for before anything is collected and withdrawable at any time from the application itself, with the same effort with which it was given.

We explain it because it deserves an explanation:

1. Article 22.2 of the Spanish LSSI requires your consent to store data on your device and retrieve it, and requires it whether or not personal data are involved afterwards. Our summary builds up during the day in a database that lives on your computer: that is storage on your terminal equipment. That is why the application asks before writing the first row, and does not merely ask before sending. It is also why the switch starts off: a consent ticked by default is not consent.
2. We could have relied on legitimate interest (art. 6.1.f) for the subsequent processing —measuring the use of one's own product is a recognised interest, the data are minimal and do not allow a person to be singled out—. We have decided not to, for two reasons. The first is legal: when access to the terminal equipment has already required consent under article 22.2, basing the subsequent processing on a different basis empties that consent of meaning, and that is how the supervisory authorities see it. The second is one of substance: you are a firm and your case files are the most sensitive thing you handle; asking your permission and accepting a «no» without consequences is the only position consistent with the rest of this policy.
3. Saying no changes nothing. The program works exactly the same, with all its functions, and we do not ask you again.

We say it with the right word: the daily summary is pseudonymised, not anonymous. The installId is stable for as long as you keep it turned on, and that makes it possible to know that the summaries of two different days come from the same installation —which is exactly what is needed to measure whether people come back—. The fact that we cannot get from that pseudonym to you does not make it anonymous, and calling it so would be inaccurate. It is therefore treated as personal data, and you have over it all the rights in section 11.

The daily summaries and the errors, thirteen months, which is what makes it possible to compare a month with the same month of the previous year. After that period they are deleted automatically. From them, some aggregated monthly totals are kept with no time limit —how many installations, from which countries, with which versions— which no longer contain any pseudonym or allow any particular installation to be traced.

There are none. These data are not used to build profiles, nor to take any decision that affects you, nor to segment you commercially. They are not cross-referenced with your account, your licence or your billing; in fact, by the design of the pseudonym, they could not be cross-referenced even if we wanted to.

With your own key, we process no data from your case files. JUDLEX is a program that runs on your machine: the files stay on your computer and AI requests go straight to the provider you hired. You are the controller of that data, and your AI provider is your processor —under the contract you have with them, not with us.

With JUDLEX AI switched on, we do. What the AI needs to read goes through our server, and for that we are your firm's processor (Article 28 GDPR), with Anthropic as subprocessor. The terms of that processing are in clause 12.1 of the Terms of use, and are accepted when JUDLEX AI is switched on.

As regards your own data —your account, billing and support— we are the controller, and sections 4 and 5 cover that.

We disclose data only to those needed to provide the service:

• Stripe, for payment, including JUDLEX AI credits.
• The hosting provider of our servers, which are in Germany.
• With JUDLEX AI switched on, Anthropic, which answers the requests, and the subprocessors Anthropic lists in its trust centre (trust.anthropic.com).

There is no email provider on this list because the site does not send email today: what you write in the contact form is recorded in our system and we read it there.

With your own key, your AI provider is NOT on this list, and that is not an oversight: it is not our processor. You hire it directly, the request goes from your computer to theirs without passing through here, and the processing contract for that data is the one you have with them.

An up-to-date list of processors and sub-processors is provided on request at support@judlex.com, and notice is given before a new one is brought in.

We say this first because it is what you want to know before deciding anything: CONSULTING EVEREST LLC is a United States company, and its servers are in Germany, inside the European Union. What sits in the United States is the owning company, not your data.

With your own key, your AI requests do not pass through us (section 6): they go from your machine to the provider you chose, and that relationship is between the two of you — including where that provider's servers are, which is set out in your own contract with them.

With JUDLEX AI, the requests go to Anthropic, which is in the United States. That transfer relies on the European Commission's standard contractual clauses that Anthropic incorporates into its data processing addendum. What passes through our server on the way is not stored.

What does reach our servers in Germany is your account, your licence, your billing and —if you have turned it on— the usage summary in section 7. The country and city are worked out by that same server with a database it holds locally: no third party is consulted and nothing leaves Germany to find them out.

You may exercise your rights of access, rectification, erasure, objection, restriction and portability by writing to support@judlex.com, proving your identity. You may also withdraw consent where you gave it, without affecting the lawfulness of prior processing.

If you believe we have not handled your request as we should, you may complain to the Spanish Data Protection Agency (www.aepd.es).

We apply appropriate technical and organisational measures: encryption in transit, access control, logging of the communications that leave your machine, and deliberate minimisation of what is stored —the design of the application keeps the content of your case files from ever reaching our systems.

And what is up to you, said plainly: your case files live on your computer, and JUDLEX does not encrypt them. What protects them on a stolen machine is the system's own disk encryption —FileVault on macOS, BitLocker on Windows—. The application checks whether it is on and tells you in Settings → Data; if it cannot find out, it tells you that too instead of assuming it is. Your AI providers' keys are stored in the application's internal folder with permissions restricted to your user account, and the backup JUDLEX makes does not include them: you can take the backup to an external disk without exposing them.

This policy is revised whenever the way the product works changes, not merely once a year: it is anchored to a technical fact sheet that is updated at the same moment the application changes. Substantial changes are communicated to users holding a licence in force.

Last updated: 3 November 2026