Skip to content
Lucerna Law

Private AI for law firms

Privileged by design. AI for your client files.

Ask questions across every document, email, and text message in a matter. Get answers cited to the source. Your data is never used to train any model.

See how your data is handled
Okafor v. Brightwell LogisticsSample matter

Question

When did Brightwell’s HR first learn of Dana Okafor’s complaint?

Answer

On March 3, 2025, eleven days before the termination. Reyes testified that the written complaint reached her that morning1. She texted the site manager about it within the hour2, and he confirmed that evening that he had read it3.

  1. 1

    Reyes Dep. 142:7–12

    Q. When did you receive it?
    A. That Monday morning. March third. It was on my desk when I got in.

  2. 2

    SMS · Mar 3, 2025 · 9:42 AMBW-004417

    L. Reyes (HR): Dana just filed a written complaint about the schedule change. Forwarding it to you now.

  3. 3

    SMS · Mar 3, 2025 · 6:05 PMBW-004431

    T. Marsh (Site manager): Read it. Let's talk before anyone responds to her.

Illustration using a fictional matter. Select a citation to see its source.
  • No training on your data

    Not by us, and not by any model provider. It is a contract term, not a setting.

  • Encrypted with your firm's own keys

    Each firm's files are encrypted under a separate key, in transit and at rest.

  • Every action audit-logged

    Uploads, questions, answers, exports, and permission changes, exportable per matter.

  • SOC 2 Type I audit planned

    We will not claim a certification until the auditor's report exists.

4,000 texts.
600 emails.
A 900-page production.
One deadline.

Most of a matter is not in the pleadings. It is in the message threads, the inboxes, and the production nobody has finished reading. Lucerna Law reads all of it and gives you a dated timeline in minutes, with every entry linked to the message or page it came from.

What it does

Four kinds of work, each one cited.

How your data is handled

From upload to answer, in five steps you can audit.

  1. 1

    Upload

    PDFs, Word files, email archives, and text-message exports go straight into the matter.

  2. 2

    Encrypted in your firm's vault

    Files are stored under your firm's own key, isolated from every other firm and matter.

  3. 3

    Processed under zero-retention terms

    Only your question and the relevant excerpts reach the model. The provider keeps nothing.

  4. 4

    Answer with citations

    Every statement links to its page, paragraph, or message. No source, no answer.

  5. 5

    Logged

    The question, the answer, and who asked are written to an audit log you can export.

The only data that leaves your firm’s boundary is the question and the excerpts needed to answer it. Ethical walls and logging are enforced in the application and the database, not only on screen.

For the ethics partner

Built for your ethical duties.

The ABA Model Rules apply to AI the way they apply to any other tool or assistant. Each control below exists because a rule requires it.

Rule 1.1Competence
Every answer is cited to its source, so the lawyer can verify it before relying on it.
Rule 1.6Confidentiality
No training on client data, per-firm encryption keys, matter-level isolation, and a published list of who processes data.
Rule 5.3Supervision
Role-based access, ethical walls between matters, and an audit log of what each user asked and exported.
Opinion 512Generative AI
A one-page mapping covers competence, confidentiality, communication, supervision, and fees.

Design partners

We are working with our first firms now.

We are opening a small number of 90-day pilots for litigation firms of 5 to 150 lawyers. Pilot firms use Lucerna Law at no charge and tell us where it falls short. We will publish their names and words here only with written permission.

How it compares

As easy to start as a chatbot. As defensible as an incumbent.

Lucerna LawConsumer AI chatbotsEnterprise legal AI
Training on your dataNeverDepends on plan and settingsTypically never
Matter isolationEvery matter, with ethical wallsNoneYes
Citations to sourceEvery answerInconsistentYes
Audit logExportable per matterLimited or noneYes
Time to startThis weekMinutesWeeks, through a sales process
PricingPer lawyer, monthlyPer user, monthlyAnnual contract

Categories are described generally. Individual products vary; check each vendor’s current terms.

Pricing

Plans from $99 per lawyer per month.
No per-gigabyte fees.

Unlimited matters on every plan. Evaluate on a preloaded sample matter, so no client data is needed to decide.

Talk to us about pricing

Questions

What partners and IT ask first.

Do you train on our data?

No. Your files, questions, and answers are never used to train any model, ours or a provider's. This is written into our contracts with model providers and into our agreement with you.

Which AI models do you use?

Claude models from Anthropic, run on AWS Bedrock. AWS does not store your questions or answers, does not use them to train any model, and does not share them with Anthropic. We name every provider that touches client data in our subprocessor list.

Where is data stored?

In the United States by default. Files are stored encrypted under your firm's own key. The database holds text and search indexes, never the original files.

Can we delete everything?

Yes. Your firm sets retention per matter and can delete a matter or the whole account. We confirm deletion in writing with a deletion certificate.

Does this waive privilege?

Lucerna Law is built to keep client information confidential: it is isolated per matter, encrypted, and never shared or used for training. Whether a given use affects privilege is a legal question that depends on your jurisdiction and facts. We recommend your firm's counsel review our data promise and terms before you upload client files.

What about hallucinations?

Answers are drawn from the documents in the matter and each statement links to its source. When the file does not support an answer, Lucerna Law says so rather than guessing. The lawyer's duty to verify remains, and the citations are there to make that fast.

See it on a real-looking matter in 20 minutes.