FoundrexFoundrex
Case study · Legal technology

AI that does the legal work — and cites every line of it.

Mindcrate is an agent platform for law firms. It researches across your matter files and live case law, drafts client-ready memos, and flags contract risk — grounding every claim in a verifiable source, with a human approving each step.

Client
Mindcrate
Industry
Legal technology
Year
2024
Role
Product design · UI · Front-end
sources grounded
1,240
every claim, verifiable
Cited
Mindcrate dashboard showing active agent runs, sources grounded, time saved, and items awaiting human review
Overview

Most legal AI gives you an answer. Mindcrate gives you the answer and the receipts.

Lawyers can’t act on a confident paragraph from a black box. Compliance with the record, the statute, and the deadline is the job — and a single stale citation can sink a memo.

So we built the product around proof. Every agent works only from your connected matter files and authorities, every sentence links back to the exact statute, case, or prior memo it relied on, and a human checkpoint sits between the agent and the client. The interface had to make that trust legible at a glance.

Grounded in your sources · verifiable in one click
1,240
Sources grounded
6.2h
Saved per matter
18+
Cited sources / memo
5
Specialised agents
Research agent live run listing each completed step and the legal sources found
The agent works in the open.Steps

Every step is visible and every source is listed as the agent runs — then it pauses at a human checkpoint before drafting.

Cited research brief with legal findings on the left and a numbered, verifiable source list on the right
Output you can defend.18 sources

Findings come back as a structured brief — each claim carries its sources, one click from verification, with conflicts surfaced and resolved.

Where it was breaking down

Four quiet failures compounding every week.

01
Research never scaled

Every memo started from a blank page, even when the firm had answered the question before.

02
Citations were “trust me”

No fast way to confirm a claim traced to a real, current source.

03
Knowledge was scattered

Drive, SharePoint, uploads, and case law all lived apart; nothing was connected.

04
No human gate

AI drafts either went out unverified or never got used at all.

How we built it

Three months. Four moves. One platform live.

01
Weeks 1–2
Map the workflow

Turned the firm’s research-to-sign-off process into discrete agent jobs.

02
Weeks 3–4
Ground everything

A permission-aware knowledge base; citations made a first-class object, not an afterthought.

03
Weeks 5–10
Ship in sprints

Real matter data from day one, not a single mocked screen.

04
Weeks 11–12
Put a human in the loop

A checkpoint between every agent run and the client.

The system

Five agents. One grounded workflow.

Research Agent

Runs deep, multi-source legal research and returns a cited memo you can drop straight into a brief.

Memo Drafter

Turns research into a structured client memo: question presented, analysis, recommendation.

Contract Reviewer

Reads contracts and clauses, flags risky terms, suggests safer language with citations.

Plain-Language Editor

Rewrites legal drafts into clear, client-ready language, grounded in your firm style guide.

Intake → Matter brief

Converts an intake call transcript into a structured matter brief automatically.

Custom agents

Define instructions, tools, and knowledge to build a specialised agent for any recurring task.

What we built

One grounded layer across every step.

Mindcrate runs research, drafting, and review on the same connected knowledge base — citing as it goes and pausing for human approval before anything leaves the firm.

Inside the product

The interface, screen by screen.

Screen 01 · Agents

Pick an agent and go.

A workspace of specialised agents, each with its status, usage, and a one-click run — plus templates to spin up a new one in seconds.

Start hereTemplatesRecently used
My Agents workspace showing five specialised legal agents as cards with status and usage
Screen 02 · Knowledge base

Grounded in your sources.

Connect Drive, SharePoint, uploads, and live case law. Agents only ever see what you choose to share — grounding that’s powerful and permission-aware.

DriveSharePointCase lawPermission-aware
Knowledge Base screen showing connected sources and indexed collections shared with agents
Screen 03 · Drafting

Drafting that stays cited.

The agent drafts a section beside the document, shows exactly what it’s grounded in, and lets you insert it with one click — citations attached.

Grounded inInsertQuick actions
Grounded drafting editor with an AI agent panel drafting a cited section beside the document
Screen 04 · Review

Nothing ships unverified.

AI reviews the draft against your knowledge base, proposes tracked changes with sources, and routes everything through a human approval step before it leaves the firm.

Tracked changesNeeds sourceApprove
AI review and approval screen showing tracked changes, cited suggestions, and a draft-to-approval workflow
Tech stack

Engineered to stay grounded.

Next.js
Next.js
TypeScript
TypeScript
Python
Python
PostgreSQL + pgvector
PostgreSQL + pgvector
Vercel
Vercel
The impact

From a research marathon to a morning’s work.

Metric
Before
With Mindcrate
Research per memo
6–8 hours
~35 min
Verifying each citation
by hand
one click
Memos per associate / week
3
12
Confidence at sign-off
“hope it’s current”
cited & dated
1,240
Sources grounded
6.2h
Saved per matter
18+
Cited sources / memo
Scope of work

What we delivered.

01
Product strategy & flows
02
UX for agent workflows
03
UI design system
04
Front-end build
05
Knowledge-base integration
06
Citation & verification system