LexLens — Anonymous Legal Notice Analysis

Understand any legal notice in 60 seconds. No account. Nothing saved.

LexLens converts legal notices (cheque-dishonour demand notices, debt-collection letters, payment demands) into a plain-language, citation-verified breakdown — sender, demands, deterministically calculated deadlines, severity, legal conditions and an action plan — in English and हिन्दी. Download the result as a structured PDF and finish. There is no history by design.

Built on Next.js 16 (App Router) with a clean white production UI (Poppins for the interface, Times New Roman serif for legal document text).

Anonymous session model (no accounts)

• No login, no signup, no profiles, no history. You open LexLens and analyze a notice.
• A cryptographically random session token (HttpOnly cookie) associates the current browser with temporary processing resources. The database stores only a SHA-256 hash of the token.
• Every temporary resource (uploaded document, OCR text, analysis result, evidence, draft, brief, PDF) is scoped to the session and verified server-side — one anonymous session can never access another session's data (foreign IDs return a privacy-preserving 404).
• Everything expires automatically (24-hour sliding TTL) via a cleanup sweep; uploaded files are removed together with their records.
• "Start New Analysis" on the report force-expires the current session and issues a fresh one — nothing carries over.

Quick start (local VS Code)

# 1. install dependencies (npm or bun both work)
npm install

# 2. create the environment file
cp .env.example .env        # Windows: copy .env.example .env

# 3. create / update the PostgreSQL database
npm run db:push

# 4. run the dev server
npm run dev

# 5. open http://localhost:3000

Environment variables (see .env.example):

Variable	Required	Purpose
DATABASE_URL	yes	PostgreSQL connection string for the application database
ZAI_BASE_URL	for OCR/LLM	Z AI API base URL, including /v1
ZAI_API_KEY	for OCR/LLM	Z AI API key
SDK credentials	optional	When the Z.ai SDK credentials are present in the environment the LLM engine is used; otherwise the app automatically falls back to the built-in offline engine and never hard-fails

For scanned PDF and image OCR, configure either ZAI_BASE_URL and ZAI_API_KEY in .env, or create .z-ai-config in the project root with {"baseUrl":".../v1","apiKey":"..."}. Keep both files out of source control. Without Z AI credentials, only PDFs with an embedded text layer can be processed.

Flow

Route	Purpose
/	Landing — Understand Any Legal Notice in 60 Seconds
/analyze	Upload a notice — paste text, upload PDF / PNG / JPG / WEBP, or launch fictional samples; optional jurisdiction hint
/processing	Working — animated pipeline while the analysis runs
/notices/[id]	Report — jurisdiction, classification, deadline, Action Center, timeline, key facts, missing information, evidence, legal basis, consequences, response draft, lawyer brief, Download Full Report PDF, Start New Analysis

Legal engine (safety-critical design)

• Jurisdiction detection (src/lib/lexlens/jurisdiction.ts): multi-signal detection → INDIA | USA | UNKNOWN. Conflicting or insufficient evidence is never guessed: UNKNOWN cases run no substantive rule pack until the user confirms the country.
• Jurisdiction firewall (src/lib/lexlens/validator.ts + corpus.ts): every legal claim carries a jurisdiction; claims from an incompatible legal system are REJECTED server-side before they can reach a report. India cases can never load FDCPA/US sources; US cases can never load the NI Act.
• Rule packs: INDIA:cheque_bounce (NI Act §138 — the recipient's 15-day payment window runs from RECEIPT of the demand notice; the 30-day period belongs to the payee and is never shown as a "dispute window") and USA:debt_collection (FDCPA §1692g 30-day validation window, loaded only when federal applicability is established from the facts). US state law is scoped to the detected state and never guessed.
• Deterministic deadline engine (deadline-engine.ts): the LLM never computes a deadline. calculateStatutoryDeadline({ ruleId, triggerDate, jurisdiction, statutoryPeriod }) returns MISSING_REQUIRED_FACT when the trigger date is unknown — nothing is invented. Date counting is calendar-day addition in UTC (addDaysISO), documented and covered by tests.
• Canonical case facts (server/notices.ts): one source of truth for dates/amounts consumed by the deadline engine, timeline, drafts, briefs and PDF. A receipt-date consistency gate refuses to render a contradictory report.
• Languages: English + Hindi only. Legal facts are language-neutral; only presentation changes. Unsupported notice languages are disclosed, never silently analyzed as if fully supported.

Tests

bun run scripts/test-engine.ts        # deadline engine, jurisdiction detection, firewall, classification (38)
bun run scripts/test-cleanup.ts       # TTL expiry sweep (7)
node scripts/test-privacy.mjs         # anonymous session isolation matrix (27)   [dev server running]
node scripts/test-regression.mjs      # India §138 / US NY / Hindi / missing-date regressions (49) [dev server running]

Disclaimer

LexLens provides legal information, not legal advice. Demo notices are fictional. No attorney-client relationship is created. For decisions with legal consequences, consult a qualified lawyer in your jurisdiction.