What we build
We don’t sell seats or a chatbot. We build the layer that lets frontier AI do real legal work in your organization’s context — then teach your people to run it, and leave it with you.
How to start
We look at how your firm actually works. We teach your attorneys to get far more out of the AI tools they already have. Then we lay out what building further would take, with a fixed fee for each option.
Schedule of Fees
Ninety minutes by video with the founder, prepaid when you book. Bring your questions and your hardest recurring task. Before the call we send a short questionnaire; after it, a one-page written recap with prioritized next steps.
Credited in full toward an on-site engagement booked within 60 days.
Two and a half days (20 hours) at your office, for one fixed fee. There is no per-seat charge, so everyone who should be in the room can be. Travel is billed at cost, with no markup.
Knowledge layers, staged workflows, validation, private deployment, or ongoing stewardship. Each is scoped from the assessment and priced as a fixed fee before any work begins. The roadmap is yours either way.
Fees are for technology consulting and training; they are not fees for legal services. Availability is deliberately limited because the founder still practices law.
Day 1ReviewForms, processes, software, files
Day 2TrainYour attorneys, on your kind of work
Working sessions are prepaid at booking. Not sure which fits? Email first; there’s no charge for that.
The system, part by part
Each part answers a question a careful lawyer would ask of any AI output: where did this come from, what was it based on, who checked it, and will it be better next time?
Your documents, matters, and data, organized by matter, dated, typed, and traced to their source — with the access permissions they already carry.
Source connections · matter structure · provenance · indexing
Retrieval that gathers what a task actually needs — the matter’s evidence, controlling authority, the firm’s best prior work, the attorney’s preferences — and leaves everything else out.
Retrieval tuned to your tasks · relevance you can inspect
Your strongest briefs, agreements, and letters, designated as the standard. They teach structure and voice. They never supply the facts of a new matter; missing facts are marked as gaps, not invented.
Exemplar library · style guides · fact/style separation
Recurring work decomposed into narrow, checkable stages. Each stage uses the right tool — a model, deterministic code, a retrieval step, a validator, or a person — and hands a defined artifact to the next. Final documents are assembled from approved templates by code, so formatting doesn’t drift.
Workflows for your recurring matters · document assembly · run records
Checks that run against the record: factual assertions traced to a page, citations checked against their sources, and quality-critical steps that stop and say why instead of guessing.
Source checks · gap marking · fail-closed gates
Attorney review built into the workflow, not bolted on after. Edits and corrections are captured as preferences and examples, so the next draft starts closer.
Review points · correction capture · evaluation sets from real work
Frontier models through APIs your organization approves, private deployments, or open-weight models on hardware you own — chosen per task by evaluation, and swappable as models improve.
Model/role map · deployment · re-runnable evaluation
After the assessment
Every build is scoped from the assessment, priced as a fixed fee before work begins, tested on past matters before it touches live ones, and documented so your people can run it.
Weeks, not quarters
One high-value workflow built end to end on your own past matters — a discovery deficiency analysis, expert-deposition summaries, a chronology, a recurring motion — and measured before it goes live.
Practice group or department
The knowledge layer, context assembly, and a set of staged workflows for a practice group, litigation team, or legal department.
When it should stay in-house
Hardware sizing, open-weight model selection, and a tuned serving stack for workloads you decide should run on machines you control.
Which model, for which job
Evaluation of models and workflows on your tasks, with verified model identity, repeated runs, and results you can re-run as models change.
Independence
Ongoing refinement, model refreshes, new workflows, and follow-on training — or a clean hand-off.
Before any of it
Start with a working session. If you go on to the on-site engagement, the fee is credited.
What you keep
Questions
No. It changes what attorneys spend their time on. Every workflow we build ends in attorney review, and we won’t build one designed to skip it.
Those give you access to a model. We build what sits around the model: your knowledge, retrieval, staged workflows, validation, and review. Often the same model you already license can power it.
Not by default. Most of the value comes from retrieval, workflows, and validation, none of which require training a model. If fine-tuning ever makes sense for your organization, it is a separate, explicit decision with its own data governance.
Not necessarily. Many workloads run well through a frontier API your organization has vetted. Owned hardware makes sense for particular sensitivity or cost profiles; the assessment tells you honestly which applies to you.
No one honestly can. Architecture can reduce exposure — where data goes, who can retrieve it, what is logged — but your professional obligations remain yours to judge. We make the tradeoffs explicit so you can.
A 90-minute working session by video is $1,000, prepaid. The on-site assessment and attorney training is $15,000 for two and a half days (20 hours), plus travel at cost; the working-session fee is credited toward it if you book within 60 days. Builds are quoted as fixed fees from the assessment, before any work begins.
Yes. That is by design. Most attorneys use a fraction of what the tools they already pay for can do. The training day teaches deeper, verifiable use on your own kind of work, and that value stands on its own.
Nothing is reviewed until a mutual NDA is signed, and you decide what we see. Closed or redacted matters work well. We review in your office on your systems; nothing needs to leave your control for the assessment.
No. It is a technology consultancy. It does not provide legal services or legal advice, and engaging it does not create an attorney-client relationship.
Next step
Two paragraphs about your practice is enough for an honest first read on whether — and where — this fits. Or book a working session.
nick@proprietarylegal.com