Home/Questions and answers
Questions and answers
We have collected the questions asked by lawyers, firm partners and security departments. If your question is not here, write to us and we will answer substantively.
| What we are comparing | A general-purpose AI assistant | Manual review of a volume | Cassation |
|---|---|---|---|
| A reference to a case sheet | from the model's memory | exact, done by hand | inserted by code, with a verification flag |
| A second pair of eyes | no | a colleague, if available | a sceptic, opposing counsel and a judge |
| Where the files are stored | on someone else's server | with you | with you, or in an isolated space |
| Time for 4 volumes | fast, but with each claim checked | days | hours for the first review |
| Deadlines and dates | computed by the model, gets it wrong | by hand | computed by deterministic code |
| Model spend | opaque | — | an estimate before it runs, a breakdown by case |
Three differences, and all three are verifiable. First: the reference to volume, page and paragraph is inserted by deterministic code from a closed list of fragments — the model physically cannot name a page that is not in the case. Second: built-in debates, in which a sceptic, opposing counsel and a judge review the analysis, and the judge never coincides with the author of the answer. Third: the environment — the files may never leave your server at all.
If a debate role, the recognition engine or a model provider fails, the platform says so directly in the answer and in the interface. There is no silent substitution of a model or quiet truncation of a result that could easily be mistaken for a complete one.
No, and that is a principle. Every generated document is created as a draft and changes status only by an explicit decision of the lawyer: an edit, an approval, a confirmation. Deadline dates are computed by code from data confirmed by a human, not by the model.
Yes, for closed environments. The app finds the server on its own network and works only inside the perimeter. The most recent files and answers are also available offline — in the courtroom that matters.
Two values: seats, and the storage volume per deployed environment. The hosting is one of two options — our server or your firm's hardware; the functionality is the same, and only the storage location and who is responsible for the hardware differ.
Model calls are counted separately: you pay the provider with your own key, or us for a prepaid balance. Infrastructure is not the issue here — it costs cents to install, and the real cost of the product is determined by model spend.
The project is needed when the platform is deployed in your perimeter and connected to a model stack as your security department requires: commercial models under your control, a hybrid stack, or a fully local stack with no external calls.
The project includes an audit of the infrastructure and security requirements, deployment of the platform, model integration and routing by role, local OCR arbitration, connection of legal databases and registries under your contracts, team training, documentation and acceptance against criteria. The final price depends on the number of environments, the scale of the local stack and the set of integrations.
Both options coexist and can be switched in the admin area. Your own keys suit large firms that already have contracts with providers. Our balance suits most: there is no need to set up keys with several providers and monitor quotas. You can mix them: your own key for one role and our balance for the rest.
Usually not. Our gateway enables caching of case files, batch processing when volumes are uploaded, and cheap models in cheap roles — this cuts the bill on input tokens several-fold. The service fee is stated openly; there is no hidden markup: in a legal setting, it harms trust more than its size would.
Work stops in different ways, and that matters. Balance exhausted — the lawyer sees what the operation would have cost and can ask the administrator for a top-up. Limit triggered — an administrator or partner raises it on the spot; this is an internal policy, not a purchase. Provider key rejected — a clear message instead of a network error. Viewing files, search and answers already produced always work.
Not necessarily. A Mac mini with 8 GB of memory is a working configuration, and if you have no hardware of your own, we supply a prepared node or deploy the system on a VPS at your provider. Extra spare memory is needed only for a fully local environment with models on board.
No. The server itself establishes an outbound connection — any router and almost any corporate network allows outbound traffic on 443. The app tries the local address and the relay at the same time and uses whichever answers first. There is no need to configure the router or forward ports.
Yes. Moving between environments is a routine operation, and files are exported on request. You are not locked in with us: the firm's data remains the firm's data.
The app from the store can sign in to an already paid account and works with our platform by default — that is what the stores' rules require. Purchases and balance top-ups are made only on the website: if we were to sell inside the app, the store would keep 15–30%, and in our economics that would mean paying extra for every client transaction.
Did not find an answer?
Write to us — we will answer substantively, including questions from your security department and legal department.