⌗ One install. One paste. Done.

Download LexVault.

Subscribe on this site, copy the license token we give you, paste it into the App on first launch — that's the entire setup. After that the App lives on your laptop, fetches its own updates, and never sends a single byte of your work back to us.

3,062,507legal chunks shipped
10jurisdictions live
0queries leaving your laptop

Don't buy if your hardware is below this line.

LexVault runs the model on your machine, not in the cloud. That's the whole point — your documents never leave your laptop. The tradeoff: a weak machine gives you a slow, frustrating product. We'd rather lose your subscription than have you blame us for that.

Buy if you have
  • Apple Silicon (M2 or newer) with 16 GB+ unified memory
  • NVIDIA RTX 3060 12GB or better, 16 GB+ system RAM
  • NVIDIA RTX 4060 / 4070 / 4080 / 4090, 16 GB+ system RAM
  • AMD Radeon RX 5700 XT or newer (8 GB+ VRAM), 16 GB+ system RAM
  • Intel Arc A750 / A770, 16 GB+ system RAM
  • A used RTX 3090 24GB (~$600) is the best value for power users
Workable but slower
  • 16 GB Windows/Linux laptop, no GPU — runs on CPU. ~6–10 tok/s. Functional, not fast.
  • RTX 3050/3060 8GB or RTX 4060 8GB — full GPU offload of 9B at 8K context (vs 16K on bigger cards).
Don't buy if
  • Less than 16 GB RAM
  • 8 GB Apple Silicon (M-series) — too tight for 9B + OS + browser
  • No GPU AND no AVX2-capable CPU (most pre-2013 hardware)
  • You expect ChatGPT-level latency on a 5-year-old work laptop. Honestly, you'll hate it.

On the recommended hardware, a research query returns in 5-15 seconds and feels like cloud chat. On marginal hardware, the same query takes 30-90 seconds. The product is the same — your machine is the bottleneck. Pick the tier you can live with before you subscribe.

macOS · Linux · Windows

LexVault App

The whole product in one binary. First launch: sign in, pick the jurisdictions you need, your data packs sync down. After that it works offline. Software and statute updates pull in automatically — Ed25519-verified, no user data sent the other way.

Research
Ask in any language, get answers grounded in real statute and case law. LLM Judge verifies every claim.
Doc analysis
Upload a contract — clauses, risks, missing standard provisions, key dates. Cited to the page.
M&A due diligence
8-stage pipeline across data-room documents. Risks, warranties, disclosure gaps, playbook positions.
6 languages
English, Deutsch, Slovenčina, Čeština, Română, Ελληνικά — UI and AI flip together with one toggle.
Download blocked or stalled? Try a direct link.

Some browsers (Safari especially) flag unsigned downloads. If the click does nothing, gives a "potentially harmful" warning, or stalls midway, paste this in Terminal — it bypasses browser security:

curl -L -o ~/Downloads/LexVault.zip https://lexvault.pinemarshlands.com/api/app/mac
# then double-click ~/Downloads/LexVault.zip to extract

For Intel Mac, swap mac for mac-x64. For Windows, the file is at /api/app/win.

How it works after you double-click

  1. Buy your jurisdictions on /buy. We give you a single license token (looks like lv_live_4f8a…). Copy it.
  2. Open the App → first-launch screen has one field: paste the token, click Activate. App detects your hardware (Apple Silicon / Intel / NVIDIA / CPU) and pulls the right LLM weights + your subscribed data packs.
  3. You're in. Disconnect from the internet — research, doc analysis, M&A due diligence all work locally.
  4. Software updates pull silently in the background using just your token (no account login, no password). Statute pack updates show up in Settings when ready.

First-run dialog on macOS

These builds aren't yet code-signed by Apple, so the first time you double-click you'll get an "unidentified developer" dialog. Two fixes:

  1. Right-click → Open the first time, confirm in the dialog. After that it opens normally.
  2. If macOS still refuses, open Terminal and run: xattr -dr com.apple.quarantine ~/Downloads/LexVault.app

Apple Developer ID signing is on the roadmap — once we ship signed binaries this dialog goes away.

⌗ Built for the laptop you already own.

The App picks the right model for your hardware.

Every paying customer runs the same model: Qwen 3.5 9B Q4_K_M at 16K context. No degraded variants. We picked one model that's strong enough for legal reasoning and engineered around it: mixed Q8/Q4 KV cache, GQA attention, full GPU offload — the whole 5.7 GB model + 16K context fits in 8 GB of VRAM with no spillage to system RAM. The App auto-picks the runtime path (Metal / CUDA / Vulkan / CPU) for your machine.

Apple Silicon Metal

MacBook Pro / Air / Studio

M2 or newer · 16 GB+ unified memory

  • Qwen3.5-9B Q4_K_M, full Metal offload
  • ~50–80 tokens/sec on M2 Pro
  • 16K context, mixed Q8/Q4 KV cache
NVIDIA CUDA

RTX 3060 12 GB and up

NVIDIA · 12 GB+ VRAM · 16 GB+ system RAM

  • Qwen3.5-9B Q4_K_M, full CUDA offload
  • ~60–120 tokens/sec depending on card
  • RTX 4090 is the speed king
AMD Vulkan

Radeon RX 5700 XT and up

AMD GPU · 8 GB+ VRAM · 16 GB+ system RAM

  • Qwen3.5-9B Q4_K_M via Vulkan
  • ~30–60 tokens/sec on RDNA 2/3
  • Full GPU offload — no CPU spillage
Intel Vulkan

Arc A750 / A770

Intel Arc · 8 GB+ VRAM · 16 GB+ system RAM

  • Qwen3.5-9B Q4_K_M via Vulkan
  • Same model, slower than CUDA/Metal
  • Workable for solo practice
CPU only AVX2

Modern laptop / desktop without GPU

Any CPU with AVX2 · 16 GB+ RAM

  • Qwen3.5-9B Q4_K_M on CPU
  • ~6–10 tokens/sec — slower but works
  • Below recommended; consider a GPU
Auto-detected on first launch and re-checked on every start. You can override your tier in Settings if you'd rather pull a smaller model on a stronger machine — useful for battery, fans, or just leaving headroom for other apps.

Updates flow in. Your work doesn't flow out.

The App reaches the network only to check for software updates and verify your license. Everything else — your queries, your documents, your conversations, your audit log — stays on your machine. We literally can't see them.

What goes out

  • Your license token (the only credential we have)
  • App version + platform (so we send the right update)
  • Hardware tier (so we send the right LLM model)
  • The jurisdiction codes you subscribe to (so we send the right packs)

What stays put

  • Your queries and conversations
  • Documents you upload
  • M&A data rooms and analyses
  • Vault contents and notes
  • Audit logs (yours, not ours)