AI Legal Intake · LegalOS

AI Intake for Law Firms. Nothing Leaves Without Approval.

Every inquiry in the firm's mailbox read, sorted and summarised on one page, facts kept apart from statements, deadlines flagged, and first replies drafted for a person to approve. This page shows how we designed it for an immigration law firm in Germany.

Case summary: family reunification, visa refused

Urgent: statutory deadline in 21 daysCategory: standard
  • DocumentVisa refused, notice dated 3 March
  • Client statementSpouse has lived in Germany for three years
  • Authority, reportedEmbassy said a language certificate was missing

Missing documents

Refusal notice, page 2; marriage certificate

Illustrative example with invented data, not a real case.

What is AI legal intake?

AI legal intake is a system that reads every inquiry a law firm receives, opens one file per matter, records each fact with its source, flags statutory deadlines and prepares a one-page summary and a draft reply. Lawyers and staff decide; the system prepares and never sends or declines on its own.

See AI agents for law firms
Project Design

What did an immigration law firm in Germany ask us to design?

A system for the firm's inquiry mailbox: read every email and attachment, separate what is documented from what the client says, catch deadlines, sort inquiries by the firm's own rules and draft the first reply and a short legal opinion for approval. We wrote the full design, every step, screen and acceptance criterion, as the basis for the project. It is a design, not a system in production yet.

Visa refusal with a deadline for appeal

Naturalisation, with or without a running procedure

EU Blue Card with a fixed job start

Family reunification

Permanent residence

Inquiries outside the firm's area, summarised only

Kinds of case from the firm's own examples. Names of the firm, its lawyers and its clients are not published.

See our custom AI development services

The Flow

How does an inquiry move through the system?

In five parts, each with fixed, numbered steps. Parts 1 to 3 take an inquiry from the mailbox to an approved first reply and short legal opinion; parts 4 and 5 take the mandate to the firm's software and the first letter to the authority.

Milestone 1

A new inquiry comes in

One file per matter. Email and attachments read, documents named, facts stored with their source, urgency checked twice and the inquiry sorted by the firm's rules.

Milestone 1

First reply, approved by staff

A one-page summary for every file. For the most frequent kinds of case, a draft reply with a document list, questions and the offer from the firm's table.

Milestone 1

Payment and short legal opinion

Every new email is attached to its file or set aside. Once a person ticks "paid", a short legal opinion is drafted with gaps marked, for the lawyer to approve.

Milestone 2

Mandate and handover

Conflict check as a name search the lawyer confirms, roles and master data, power of attorney and fee agreement from templates, then the file created in the firm's software.

Milestone 2

Letter to the authority

Notice of representation from templates, approved by the lawyer, a follow-up date, and the authority's reply analysed for contradictions and deadlines.

From the client's email to the firm's file
  1. Prospective client
    writes to the firm
  2. Inquiry mailbox
    Microsoft 365
  3. AI system
    reads, sorts, summarises, drafts
  4. Staff and lawyer
    approve or send back
  5. Firm's file
    mandate in the firm's software
The prospective client writes to the inquiry mailbox, the AI reads and prepares the file, staff and the lawyer approve, and the mandate goes to the firm's software.

See the Outlook and Microsoft integrations

Facts With a Source

How does the system keep facts apart from what the client says?

Every statement is stored with its source and one of the firm's categories, and the categories are never mixed. The lawyer's own assessment is kept separately and is never recorded as a fact.

  • Documented fact, from a document
  • Statement of the client
  • Reported statement of an authority or third party
  • Assumption or fear of the client
  • Unresolved allegation

Source: the client's first email, line 4

We applied for the visa in June.

I think the embassy did not receive my language certificate.

The refusal notice is attached.

Statement of the clientThe client is unsure himself, so it is not recorded as a documented fact.
Illustrative example with invented data, not a real case.

Read how documents are searched and cited

A Person Decides

Who approves what before it leaves the firm?

The firm sets the rules, and the system enforces them technically. Drafts wait on one approval page until a person decides.

  • Nothing leaves the firm without a person's approval, and every approval is recorded with version and time.
  • Staff approve the first reply; the lawyer approves legal opinions and every letter to an authority or court.
  • The system never declines an inquiry on its own; a decline letter is a template a person approves.
  • Products, prices and fees come only from the firm's own table, never from the AI.
  • A file with a statutory deadline or a negative decision turns red and is seen by a person first.

Approval: first reply, version 2

Thank you for your inquiry. Based on the refusal notice, an appeal is possible within the deadline. To assess your case, please send us the documents listed below. We can offer a short legal opinion or representation as described in the attached offer.
ApproveEdit and approveSend back with commentBring in the lawyer

Recorded: who approved which version and when. The reply is sent by a person from Outlook.

Illustrative example with invented data, not a real case.

See how Tridan AI handles access and approvals

Acceptance

How is the system tested before it touches real inquiries?

Against the firm's own cases, with targets agreed before work starts. The same test cases run again before every go-live and every change to rules or model.

Swipe sideways to see the full table

Acceptance criteria in the LegalOS project design
CriterionTargetMeasured on
Category as the lawyer would assign itabove 80%cases classified by the firm
Questions as the firm would ask themabove 80%the firm's past replies
Drafts usable with minor editsabove 80%20 to 30 drafts rated by the lawyer
Live operationtwo weeks without a critical errorreal inquiries in the mailbox

A draft counts as usable with minor edits only if no legal correction was needed, however short the edit.

Critical errors: a single one blocks acceptance

  • A missed statutory deadline
  • A message or document on the wrong file
  • An invented fact that matters for the decision
  • A wrong price or an invented product
  • A file that continues without the required lawyer's decision

Read how AI agents are trained and tested

In Phases

What comes first, and what comes later?

Each milestone is a separate project with its own acceptance, so the firm sees results before committing to the next.

Milestone 1: intake and first reply

Weeks 1 to 5 on the firm's anonymised past cases, then weeks 6 and 7 on the live inquiry mailbox: files, summaries, first replies and short legal opinions with approval.

Milestone 2: mandate and authority

Conflict check, roles, power of attorney and fee agreement, the file in the firm's software, the first letter to the authority and the waiting state with a follow-up date.

Later stages

Substantive letters, automatic processing, court proceedings, further legal areas and use by other firms, each scoped and agreed separately when the firm is ready.

See the self-hosted deployment option

Professional Secrecy

How is client data protected?

The design treats client data as privileged from the first email. These are the measures it specifies; they describe practices, not certifications.

Server in Germany and an EU AI provider under a data processing agreement (Art. 28 GDPR)

Names, email addresses and dates of birth replaced by placeholders before any call to an AI model

Every service provider with access to client data bound to confidentiality in writing (section 43e BRAO)

Access only to the inquiry mailbox, with read and draft rights and no right to send

Login with the firm's Microsoft 365 accounts; roles assigned by the firm

A record of every AI call and every approval; all accounts in the firm's name

Read about security and data processing

FAQ

Common Questions from Law Firms

Does the AI send emails to clients on its own?

No. In the LegalOS design the AI legal intake system only prepares: it reads, sorts, summarises and drafts. Every reply, legal opinion and letter waits on an approval page until a staff member or the lawyer approves it, and the reply is sent by a person from Outlook. Access to the mailbox has read and draft rights only, with no right to send.

Can the system decline an inquiry automatically?

No. The AI legal intake system sorts inquiries by the firm's own rules, including a category for matters that are not economically worthwhile, but it never declines anything on its own. A decline is a short letter from the firm's template, with no legal assessment, that a person approves before the file is closed.

How is client data protected under GDPR and professional secrecy?

The design runs on a server in Germany with an EU AI provider under a data processing agreement (Art. 28 GDPR). Names, email addresses and dates of birth are replaced by placeholders before any AI call, every provider with access is bound to confidentiality in writing (section 43e BRAO), and every AI call and approval is recorded.

Does it work with our existing software?

The LegalOS design connects to the firm's Microsoft 365 inquiry mailbox and login, and hands each mandate to the firm's practice management software through its interface: addresses, the file, documents with metadata and statutory deadlines. From then on the firm's software leads the file. The connection is scoped for your software in the analysis.

Is this only for immigration law?

No. The LegalOS design was written for an immigration law firm in Germany, but the approach applies to any practice that receives many similar inquiries: intake, facts with sources, deadlines, summaries and drafts with approval. The legal area is configuration, and each new area gets its own categories, templates and test cases before it goes live.

How much does AI legal intake cost?

AI legal intake is priced on request, per phase, because the scope depends on your inquiry volume, your kinds of case, your software and which milestones you need. It starts with an analysis of your intake and approvals, followed by a proposal with a fixed scope and acceptance criteria for each milestone before any work begins.

Browse the platform FAQ

Last updated · Reviewed by Nemanja Milivojevic, Founder

Start With an Analysis of Your Firm

We map your intake, your rules and your approvals, then send a phased proposal. Pricing is on request, per phase.

AI agents for law firms: calls and chat