Skip to content
Caseagent

AI Legal Assistant for Lawyers That Works Inside Your Case Files

An AI legal assistant is software that does a lawyer's non-billable casework: reading the file, spotting issues, drafting documents, and tracking deadlines. Caseagent runs that assistant inside case management, on a five-stage loop, with a licensed attorney reviewing every output.

Run the live demo
Exhibit A Direct answer

What is the agent loop?

The agent loop is the cycle Caseagent runs on every active matter: ingest, analyze, draft, calendar, notify. Each pass reads what changed in the file, works it, and hands the results to you for review. The five stages below are broken down one by one, with real inputs and outputs.

01
Ingest. The agent reads every document in the matter: pleadings, records, correspondence, intake notes.
INGEST
02
Analyze. It builds the chronology, spots issues, and researches the case law those issues raise.
ANALYZE
03
Draft. It writes the first pass: demand letters, motions, discovery, research memos, client updates.
DRAFT
04
Calendar. Every date in the file lands on one deadline timeline with escalating reminders.
CALENDAR
05
Notify. You get a queue of drafts to review and a weekly brief per matter. Nothing moves without your sign-off.
NOTIFY
Exhibit B Stage 1 of 5

Ingest: the agent reads every page you give it

Connect the matter and the agent reads all of it, not a summary, not the first ten pages. It turns a pile of PDFs into a structured case file: parties, claims, dates, facts, and open questions. This is the AI paralegal work that usually eats the first billable-free afternoon.

IN
A 38-page intake packet on Rodriguez v. Hale Trucking (fictional sample matter): police report, two medical records releases, an ER discharge summary, and four client emails.
SAMPLE INPUT
OUT
A dated chronology of the collision and treatment, a party-and-insurer roster, a list of the three records still missing, and a flag that the client's second email mentions a prior back injury the intake form did not.
AGENT OUTPUT
Exhibit C Stage 2 of 5

Analyze: issues spotted, authorities pulled

With the file structured, the agent works like a first-year who never gets tired: it spots the legal issues the facts raise, then runs the research those issues need. Authorities come back as a memo with every citation flagged for your verification, the same output our legal research software page walks through in detail.

IN
The Rodriguez chronology plus one question you type: "Comparative negligence exposure if the client braked late?"
SAMPLE INPUT
OUT
A research memo outline: the governing comparative-fault standard, the leading authorities to verify, how the late-braking fact pattern has been treated, and a candid weaknesses section for the file.
AGENT OUTPUT
Exhibit D Stage 3 of 5

Draft: first-pass documents from the actual record

The agent drafts from the file it just read, not from a generic template. Names, dates, damages, and exhibits come from the record, so your edit starts at 80 percent instead of zero. See what the legal drafting software covers, document by document.

IN
One instruction: "Demand letter to Hale Trucking's carrier." The agent already has the chronology, the specials, and the liability facts.
SAMPLE INPUT
OUT
A structured demand letter first draft: facts, liability, injuries and treatment, itemized specials, and a demand figure left bracketed for you to set. Marked "DRAFT FOR ATTORNEY REVIEW" until you approve it.
AGENT OUTPUT
Exhibit E Stage 4 of 5

Calendar: every deadline on one timeline

Every date the agent finds while reading, statutes of limitation, discovery cutoffs, response windows, hearing dates, lands on the matter's deadline timeline. Reminders escalate until a human confirms; a deadline is never marked handled by software alone. The full mechanics live on the deadline tracking page.

IN
Defendant's discovery requests arrive by mail scan on March 3. Nobody types anything.
SAMPLE INPUT
OUT
A computed response deadline on the timeline, a reminder set for you and your paralegal at 14, 7, and 2 days out, and a note that the deadline stays red until an attorney confirms service of the responses.
AGENT OUTPUT
Exhibit F Stage 5 of 5

Notify: the file reports to you, not the other way around

The loop closes with communication. You get a review queue of everything the agent produced; clients get status-update drafts you approve before anything sends. Intake summaries and client-facing drafts are covered on the legal intake software page.

IN
A week of agent activity on Rodriguez: records received, memo drafted, deadline computed, demand letter in the queue.
SAMPLE INPUT
OUT
A Monday brief for the matter, a client status-update draft in plain English awaiting your approval, and a review queue ordered by deadline proximity.
AGENT OUTPUT
Exhibit G The teardown

Anatomy of the agent

Under the hood, Caseagent is not one chatbot. It is a set of specialized parts wired around your case file, each with a narrow job and a hard limit.

G-1

The reader

Parses every document into the matter's structured record: parties, dates, facts, claims. Everything downstream works from this record, not from a summary.

G-2

The chronology

A living timeline of the matter. New documents update it; contradictions between documents get flagged instead of silently averaged.

G-3

The researcher

Turns spotted issues into memos with authorities. It never invents case names, and every citation ships flagged for attorney verification.

G-4

The drafter

Assembles first drafts from the record and your firm's prior work product. Output is always labeled a draft until an attorney approves it.

G-5

The docketing engine

Extracts and computes dates onto the deadline timeline, escalates reminders, and refuses to close a deadline without human confirmation.

G-6

The review gate

The hard limit around everything above: no document is filed, sent, or promised to anyone until a licensed attorney signs off. The agent proposes; you dispose.

All of it runs on infrastructure designed for privileged material: encrypted in transit and at rest, isolated per firm, never used to train models. The full stance is on the legal AI security page.

Exhibit H The rhythm

What you get every week

The loop runs continuously, but your week has a shape. Here is what a typical week looks like once the agent is on your caseload.

MON
The Monday brief. One page per active matter: what arrived, what the agent did, what needs you, and every deadline inside 30 days.
BRIEF
TUE
A review queue, not a to-do list. Drafts ordered by deadline proximity. You edit and approve instead of starting from a blank page.
QUEUE
WED
Deadline escalations. Anything unconfirmed inside its reminder window pings you and your paralegal until a human closes it.
WATCH
THU
Client updates, drafted. Plain-English status drafts for every matter with activity, waiting for your approval before they send.
CLIENT
FRI
The file, current. Chronologies updated, new documents read and filed, missing-records list refreshed. Monday starts from a clean record.
CLEAN
Exhibit I On the record

You stay the lawyer. That is the design.

Two things are true on every page of this site, and they are true here too.

An attorney reviews everything. Every draft, memo, and client message the agent produces is first-pass work product for a licensed attorney to review, edit, and approve. Nothing is filed or sent without your sign-off, and deadlines close only on human confirmation.

This is not legal advice. Caseagent is a tool for legal professionals. It does not replace an attorney, and using it creates no attorney-client relationship. It exists to give your judgment more hours, not to substitute for it.

The fastest way to check whether that promise holds is to watch the agent work: run the live demo on a fictional sample file.

Exhibit J Asked and answered

AI legal assistants, common questions

The questions US lawyers search for most before putting an agent anywhere near a live matter, answered directly.

What is an AI legal assistant?

An AI legal assistant is software that performs the non-billable casework a paralegal or junior associate would otherwise do: reading the file, building a chronology, identifying issues, running research, drafting documents and tracking deadlines. It produces first-pass work product, not finished work. A licensed attorney reviews, edits and approves everything before it is filed or sent.

How does an AI legal assistant work?

It runs a loop on each active matter. The agent ingests every document in the file and structures it, analyzes the issues the facts raise and researches them, drafts from that record rather than from a template, calendars every date it finds, then hands the output to the lawyer in a review queue. Each pass reads what changed since the last one.

Can an AI legal assistant replace a paralegal?

No. It absorbs the mechanical parts of the work, such as reading a 900 page production, building a chronology, or finding every deadline buried in a scheduling order. It does not exercise judgment, manage clients, or take responsibility for the file. Firms that use one well tend to move paralegal time toward case strategy and client contact rather than cut headcount.

Is it ethical for a lawyer to use an AI legal assistant?

Yes, with duties attached. There is no AI exception to professional responsibility. ABA guidance and state bar opinions consistently require competence in the tool's limitations, verification of every output before you rely on it, protection of client confidences under Model Rule 1.6, and billing only for time actually spent. Rules vary by state, so read your own bar's opinion rather than a general guide.

Do I have to verify what an AI legal assistant produces?

Always, and courts have sanctioned lawyers who did not. Every citation must be read and confirmed to exist and to stand for what it is cited for. Good tools make that easier by flagging each authority for verification and linking to the source rather than presenting research as finished. The professional responsibility never transfers to the software.

Does using an AI legal assistant waive attorney-client privilege?

Not if the tool is set up properly. The risk is disclosure to a third party, so what matters is whether the vendor trains models on your matter data, who can access it, and whether firm data is isolated. Consumer chatbots that retain and train on inputs are the real hazard. A legal platform that isolates data per firm and never trains on it does not create that exposure.

How much does an AI legal assistant cost?

The market splits sharply. AI added inside a practice platform runs roughly $40 per user per month. Standalone legal AI drafting tools sit near $179 per user per month. Enterprise research assistants range from $225 to $500 per user per month, and often require a research database subscription on top, which can double the real spend. Ask what the total is, not the tier price.

What is the difference between an AI legal assistant and case management software?

Case management software is the system of record: it stores matters, contacts, calendars and invoices, and waits for you to put information into it. An AI legal assistant does work on the contents of that record. The two are complements, and most firms run an agent alongside a platform. Our legal practice management software guide compares the platform layer, and legal AI vs practice management software works through the overlap.

EARLY ACCESS · 2026

Put this loop on your caseload

Caseagent is in early access. Join the list and we'll email you the moment your spot opens; early users lock in launch pricing.

Run the demo first