AI for law firms that keeps every client file inside the firm.
We build a private legal assistant on your matters, precedents and templates. Your lawyers ask in plain words and get a short answer with the exact clause or memo it comes from. The model runs on your servers or on dedicated hosting in Europe, and nothing you index is used to train anyone’s model.
- Precedent search
- Clause comparison
- Intake drafts

What AI for law firms does well today
AI for law firms is at its best on the firm’s own knowledge: finding the clause you negotiated three years ago, comparing a draft with your standard position, summarizing a data room, drafting an intake note. The assistant only writes from passages it retrieved in your files, cites each one, and leaves every legal judgment to the lawyer who signs.
Public chatbots have never read your matters, so they fill the gaps with plausible text. An assistant built on your document management system answers from what the firm actually wrote, within each lawyer’s access rights, and says so when the files are silent.
The first question partners ask us is where the files go. They go nowhere new: the index lives where you decide, the model is hosted in Europe or on your own machines, and every question and answer is logged for your own review.
What your associates tell you on Friday evening
- We negotiated exactly this cap last year, but nobody remembers on which matter.
- Nobody dares paste a client file into a public AI tool, so everyone searches by hand.
- The document management system returns a wall of files for one keyword.
- A data room review eats a junior’s whole week before the first real call.
- Each new client inquiry gets retyped into a conflict check and an intake memo.
Where a legal assistant earns its place first
Four uses we see first in firms of every size. Each one starts on your own documents, and each one ends with a lawyer’s review.
01
Search the firm’s precedents
Ask “what liability cap did we accept for a software distributor?” and get the three clauses closest to it, each linked to the matter file it comes from.
Knowledge assistant02
Compare a draft with your playbook
The assistant reads the counterparty’s draft, flags each clause that moves away from your standard position and quotes the position it is measured against.
Document AI03
Summarize a data room
Leases, change-of-control clauses, expiry dates: pulled from hundreds of documents into one table, with a link to each page so a lawyer can check before it goes into the report.
Document AI04
Prepare intake and conflict checks
Web form or email inquiries are read, the parties extracted and checked against your client list, and a draft intake note waits for the partner in charge.
AI agents
AI for law firms: what it reads in the firm, and what it returns
The assistant works only from sources the firm controls. Each source answers a different kind of question, and every answer points back to its page.
| Firm source | Typical question | What the answer shows | What the lawyer checks |
|---|---|---|---|
| Past matter files | “What liability cap did we accept for a software distributor?” | The closest clauses, each with its matter and its page | Whether the precedent fits this deal |
| Your playbook and standard positions | “Where does this draft move away from our position?” | Each departing clause next to the position it is measured against | Which departures to accept |
| A data room | “Which leases carry a change-of-control clause?” | A table of findings, with a link to each page | Every finding that goes into the report |
| Intake forms and emails | “Who are the parties, and do we already act for one of them?” | The parties checked against your client list, and a draft intake note | The conflict decision |
| Internal know-how notes | “What is our checklist for closing a share deal?” | The note and the passage the answer comes from | That the note is still current |
| Case law and statutes | Asked of your legal research databases, which stay the reference | Nothing quoted from memory: the assistant only cites documents it retrieved | Every citation, in its original source |
Illustrative scenario
Hélène needs last year’s indemnity clause before a 4 p.m. call
A business law firm, a counterparty draft on the table, and a partner who remembers that the firm already won a better cap on a similar deal. This scenario shows how the assistant fits into that afternoon. It describes no real client, and the live demo runs on a fictional company’s documents.
Try the same mechanism now on the live knowledge assistant: it answers from a fictional distributor’s procedures and cites its source for every claim.
- Hélène types her question in the firm’s intranet, in plain words.
- The assistant searches only the matters her access rights already open to her.
- It returns three indemnity clauses from past deals, each with the file and page it comes from.
- Hélène opens each source, keeps one, and adapts it herself.
- The question, the sources and the answer stay in the firm’s own log.
Confidentiality, competence and the lawyer who signs
- Client confidentiality comes first.The ABA’s Formal Opinion 512 (July 29, 2024) reminds lawyers using generative AI of Model Rule 1.6: information relating to the representation stays confidential unless the client gives informed consent. That is why we host the model in Europe or on your servers, under a contract that rules out training on your data.
- Competence includes the tool.The same opinion ties Model Rule 1.1 to understanding the benefits and risks of the technology used. We document how the assistant retrieves, what it is told to refuse, and where it can fail, in plain words for your partners.
- The lawyer reviews and signs.Every answer points to its passage so the review is fast. Nothing the assistant writes leaves the firm without a lawyer’s check, and the log shows who approved what.
- Your own regulator has the last word.State bars, the SRA in England and Wales, or your national bar each publish their own guidance. We map it with you during the assessment, before any file is indexed.
Law firms: your questions
Do our client files leave the firm?
They leave only to the hosting you choose: your own servers, or a dedicated server in Europe under a written contract. The model provider receives nothing for training, and each question and answer is logged where you can audit it.
Can it invent a case or a clause?
It is built to answer only from passages it found in your documents, and each claim points to its source. When nothing relevant is found, it says so. A lawyer still checks every source before anything is used, as your professional rules require.
Does it respect ethical walls and access rights?
Yes. The assistant applies the permissions already set in your document management system, so a lawyer only gets answers from matters they can open today. Ethical walls set on a matter apply to the assistant as well.
Which tools does it work with?
Your document management system, shared drives, email and intranet. When a tool has no connector, we build one. The assistant can live inside Teams, your intranet or a simple web page.
How long before lawyers use it?
A first assistant usually runs in four to eight weeks, on a scope and a price fixed in writing before we start. The free 30-minute assessment tells you whether your documents are ready.
Keep exploring
- AI contract review Drafts checked against your own playbook
- All industries What AI does in each sector we serve
- AI use cases Every use case, by function and industry
- Live demo: knowledge assistant Ask it a question, check its source
- Private LLM Models hosted in Europe or on your servers
- Security and data Hosting, logs and access control
- Accounting firms Client documents collected and pre-booked
- Insurance Claims files completed from the documents
Find last year’s clause before your 4 p.m. call.
Precedents, playbooks, data rooms, intake. Name one in the form and we reply within one business day to set up your free 30-minute assessment, with an honest view of what an assistant can do on your documents.