Law firms
AI for law firms that cannot let a single document leave the office
Search across case files, contracts and pleadings with a plain-language question and get the answer with a reference to the document — on a server in the firm, with nothing passing through an external vendor.
Professional privilege is not an organisational preference: it is an obligation. This page explains how AI-assisted document search can sit inside that obligation instead of putting it in question.
The daily problem is not legal research, it is searching your own archive
Legal databases already cover statute and case law. What no database covers is your archive: the pleadings filed, the contracts negotiated, the opinions given, the matters closed five years ago.
That is where the time goes. “How did we frame that objection”, “which contract did we accept that indemnity clause in”, “what did we tell the client on that point”: the answers exist, they are in one of the firm's documents, and finding them means remembering which.
Semantic retrieval answers from the meaning of the question rather than its exact words, and returns the passage the answer came from. The checking still belongs to whoever signs, but it starts from three paragraphs instead of four folders.
Why the perimeter question is different for a firm
A company evaluating a hosted service weighs a risk. A law firm also weighs a professional and contractual duty to a client who as a rule consented to nothing, and who is often bound by confidentiality undertakings of their own.
- Documents stay on the firm's server: there is no transfer to a third party to disclose to the client or to justify.
- After the first run the engine needs no network, so it can run on a machine isolated from the outside.
- Roles separate who may only read from who may upload or delete, and the filter applies before passages reach the model.
- The engine's source is public under AGPL-3.0: it can be verified rather than taken on trust.
Where it pays off most
- Internal precedent: recovering the argument already used on a comparable question.
- Due diligence and contract review: quickly locating the clauses governing a specific case inside very large document sets.
- Scanned files: PDFs with no selectable text made searchable through OCR in Italian and English.
- Multilingual matters: an Italian question finding the passage inside a contract drafted in English.
- Knowledge transfer: a new joiner querying the archive instead of consuming a senior colleague's time.
The system does not produce legal advice and must not be used to. It returns passages with their provenance: the legal judgement remains entirely the professional's.
Frequently asked questions
- Do client documents remain covered by professional privilege?
- The software does not change professional obligations, but it changes the underlying facts: with documents, index and model on a server in the firm there is no transfer to an external vendor to assess, disclose or justify. Ordinary security measures on that server still apply, exactly as they do to the file server you already run.
- Does it work on scanned files?
- Yes. When a page yields too little text it is recognised as a scan, rasterised and passed through OCR in Italian and English. That is the typical case for documents received as copies or captured on filing.
- Does a mid-sized firm need serious infrastructure for this?
- No. The engine is a single executable that starts its own components: you need a server with an NVIDIA GPU for fast answers, or a machine without one if longer waits are acceptable. It does not need an IT department, just the person who already administers the firm's server.
- Can the system get a citation wrong?
- Every answer arrives with the passages it was drawn from, so an error is immediately checkable. The assumption of use is that the professional verifies the source before relying on it, exactly as with research done by an associate.
Ready to run RAG on your own infrastructure?
Start with the open-source Community edition, or talk to us about Pro with structured extraction, SSO, audit log and SLA.