K Kept

For Singapore law firms

Every lawyer in your firm, using AI the right way.

Kept builds AI systems for Singapore law firms and trains your lawyers to run them. First-pass document review and due diligence, checked by a lawyer, kept on your own tenant, and funded up to half by EDG.

One M&A data room

Manual first-pass review 40 h
With Kept, same sign-off 12 h
Recovered per data room 28 h

Representative arithmetic, set conservatively. Every flag is checked by the reviewing lawyer.

Built to MinLaw GenAI guide · Mar 2026Law Society AI advisoryEDG · up to 50% funded

The week you recognise

A corporate associate reads a few hundred contracts for one data room, clause by clause, against a deadline the deal team set. Change of control, assignment, termination. Most of it is pattern-matching billed at an associate's rate, and the client questions the bill.

Your associates already use ChatGPT. On personal logins, sometimes with a client's matter pasted in. The Law Society has told firms not to do this. It is a professional-conduct problem before it is an IT one.

In March 2026 the Ministry of Law published its guide for generative AI in legal work. Clients and opposing counsel now have a written standard to hold your firm to. Doing nothing stopped being the safe option.

A representative engagement

The due diligence review

A corporate associate reviews the material contracts in an M&A data room for a first-pass report: change of control, assignment, and termination across the target's agreements.

The margin math
Hours per data room, manual first-pass review
40 h
Hours with the review workflow, same lawyer sign-off
12 h
Recovered per data room
−28 h
Across 20 deals a year
560 h
Representative arithmetic, set conservatively, not a client claim. Your diagnostic replaces these numbers with your firm's own, measured on your matters.

Funded up to half

A S$40,000 build, up to half funded by EDG.

Custom AI implementation qualifies for the Enterprise Development Grant's AI path at up to 50% support for SMEs, subject to EnterpriseSG approval. You pay our invoice in full and EnterpriseSG reimburses you. We prepare the project scope with you before any work begins, as the grant requires. Law-sector schemes run through MinLaw and the Law Society as well, and we confirm what is open the week you apply.

The margin math
Build & Train engagement, fixed scope
S$40,000
EDG support, up to 50% for SMEs
−S$20,000
Net cost to the firm
S$20,000
Support level verified on enterprisesg.gov.sg, July 2026. EDG remains open until the new EDGE grant launches in 2H2026; applications filed now are processed under current terms. Approval takes 8 to 12 weeks, so scoping early matters. Eligibility sits with your firm: at least 30% local shareholding, S$100M turnover or 200 staff or fewer.

Privilege is the design, not an appendix

Two pages · for the managing partner Kept

Built for privilege, not just for speed

  • Runs on the firm's own tenant. Client matter never enters a public tool or a training corpus, an architectural constraint rather than a policy line.
  • A lawyer signs off on every AI-assisted output before it reaches a client or a court. The system drafts, it never advises.
  • Full audit trail: what the system flagged, what the reviewer changed, and who signed.
  • Access mirrors your matter-management roles and conflict walls.
  • A written AI-use policy your partners can adopt, mapped to the Ministry of Law guide and the Law Society advisory.

Prepared against the Ministry of Law's Guide for Using Generative AI in the Legal Sector (6 March 2026) and the Law Society's advisory on public AI tools, on the managing partner's desk before anything is built.

After the engagement

What your firm can do once we've left

The firm keeps the margin

Recovered hours go back to billable work, not to a discount on the fee. Realisation goes up, not down.

Associates do lawyer work

First-pass reading goes to the system. Your associates spend the hours on analysis and advice, the work that makes them partners.

Privilege stays intact

One reviewed workflow on the firm's own tenant replaces a dozen private ChatGPT logins. The confidentiality risk moves from unmanaged to documented.

Your people keep the system

Your associates run it after we leave. No per-seat licence, no vendor to call when the next matter lands.

How an engagement works

Diagnose. Build. Train. Leave.

  1. 01 90 minutes

    Working session

    You describe the problem. We map how we would approach it. Within 24 hours you receive a written assessment specific to your firm, written to be forwarded to your partners without editing. No second meeting implied.

  2. 02 2–3 weeks

    Diagnostic

    Staff interviews and process mapping. You receive the opportunity map, the roadmap with conservative arithmetic, and the grant-application scope. Self-contained: if AI is not the right answer for your firm, the document says so.

  3. 03 6–12 weeks

    Build & train

    One working system, on your tenant, built around your review standards, and your staff trained to run it. Fixed scope, fixed duration, documentation and a 30-day warranty included.

  4. 04 Ongoing

    You run it

    The capability stays with the firm. Most clients operate independently, which is the intended outcome. A maintenance plan exists for those who want one; it is optional and we say so.

The practice

An advisory practice. Not a software vendor.

We build and we teach

A build without teaching leaves you dependent on us. Teaching without a build leaves you with theory. Do both and the firm ends up owning something.

No retainer treadmill

Most legal-AI vendors rent you a seat and bill every month. We build the system once and hand it over. Do the work well and you stop needing us.

We do the grant paperwork

We scope the engagement to qualify for EDG support and prepare the application with you, before any work starts. The filing is part of the service, not a referral to someone else.

Common questions

Is AI-assisted review defensible to a client or the court?

Every point the system flags carries a documented lawyer review before it reaches a report or a filing. You keep the audit trail and the written AI-use policy as deliverables. The system reads at scale, your lawyer decides. That is the design constraint, and it is built to the Ministry of Law's March 2026 guidance.

Where does our client matter go?

Nowhere outside your control. The system runs on your own tenant, and nothing you put into it trains any model. You get the data-flow diagram in the diagnostic, before anything is built. This is the setup the Law Society advisory asks for.

We are a twelve-lawyer firm. Is this sized for us?

Yes. The model is built for the ten-to-a-hundred-lawyer practice. The diagnostic is priced for a managing partner to sign off alone, and EDG support applies at your size. The legal-sector schemes shift often, so we confirm what is open the week you apply.

Start here

A 90-minute working session. A written assessment within 24 hours.

You bring the problem. We bring the analysis. You leave with a document, not a pitch: no slide deck, no follow-up sequence unless you ask for one.