DEMONSTRATION DAY
$0
One day on-site, on your own documents. Nothing leaves your building.
PROOF & PRICING
Start with your documents, not our sales pitch.
Proof happens in two stages — a day in your office at no cost, and then a measured evaluation for firms that decide to proceed. The numbers are below both.
← Back to overviewSTAGE ONE
One day, on-site, on your own files.
Nothing is anonymised, nothing is emailed, and nothing leaves the building. Your files move over your own network, get indexed on the machine we bring, and the index is destroyed in front of you before we pack up.
Book a demonstration day →STAGE TWO · EVALUATIONMeasured on your corpus
RetrievalRecall@K · clause discovery · negative queries
AnswersCorrectness · completeness · unsupported claims
CitationsSource correctness · coverage · deep-link provenance
SecurityUnauthorized retrieval · matter isolation · injection tests
Performancep50 / p95 latency · concurrency · throughput
IncludedWithin production implementation
STAGE ONE · THE DEMONSTRATION DAYNo cost. Fixed scope, agreed in writing beforehand.
Included
- One day on-site at your office, under a mutual NDA
- A representative sample of your real documents — no anonymising, no preparation
- Files moved over your own network and indexed on our machine, in your building
- Your questions answered live, each with the document, page and section to verify
- Architecture walkthrough, and a first read on where local AI fits your workflow
- The index destroyed in front of you before we pack up
What that means, and what waits
- Nothing transmitted off your premises, and no copy taken away
- No hardware purchase — the demonstration runs on equipment we bring
- Full-corpus evaluation and scorecard, which follow at implementation
- Production DMS integration, identity wiring, or HA and disaster recovery
- Model fine-tuning or custom training
No cost, no hardware purchase, and no obligation to proceed. Deeper workflow mapping — the sessions where we work out what should connect to what — is quoted as advisory days before any of it begins.
COMMERCIAL MODEL
Clear economics. Scoped where complexity is real.
No per-query or token charges for local inference. The firm owns its production hardware, and the managed service is priced as a platform rather than per seat — closer to a managed-IT relationship than a software licence.
WORKFLOW ADVISORY
$2,000/day
On-site sessions mapping how the practice runs, where the gaps are, and what should connect to what. Useful whether or not you deploy anything.
PRODUCTION IMPLEMENTATION
from$12,500
Includes the full-corpus evaluation and scorecard. Final scope depends on document-system integration, identity, matter-access controls, and selected security mode.
MANAGED PRIVATE AI
$1,250/mo
- Up to 5 users$1,250
- Up to 10 users$1,750
- Up to 20 users$2,500
A platform fee, not a per-seat licence. Adding people costs nothing until the next tier, so the cost per lawyer falls as the firm grows.
WHAT CHANGES IMPLEMENTATION SCOPE
Standard file repository ↔ iManage / NetDocuments
Simple SSO ↔ complex ethical walls
Managed Sovereign ↔ Fully Isolated
Single node ↔ HA / DR requirements
HARDWARE
The firm buys the hardware, owns it, and keeps it.
Production inference runs on equipment the firm purchases directly. LexDeploy does not resell it or take a margin on it. Nothing needs buying to see the system work — the demonstration day runs on equipment we bring. These envelopes exist so you can size the decision early; the binding number comes out of the evaluation.
A small practice does not need datacentre hardware. Modern open-weight models activate only a fraction of their parameters for any given token, which is why a five-person firm can run this properly on equipment costing less than its copier — and why over-specifying is now the more expensive mistake.
Small practice
5 – 10 users
- ComputeTwo workstation-class GPUs
- MemorySized to model and context length
- StorageEncrypted NVMe with a backup target
- SitingTower unit; needs a ventilated cupboard, not a server room
Under $5,000Viable because the models we deploy activate a fraction of their parameters per token.
Quiet office
5 – 15 users
- ComputeLarge unified-memory workstation
- MemoryShared across model and context
- StorageEncrypted internal, with backup target
- SitingSilent, low power, sits under a desk
≈ $10,000Warrantied, near-silent, and unaffected by the GPU memory shortage.
Resilient
Where HA or DR is required
- ComputeDatacentre accelerators, multiple nodes
- MemoryScaled for higher concurrency
- StorageRedundant, with off-node encrypted backup
- SitingRack-mount with failover
$20,000+Topology drives most of the cost, so this is scoped separately.
WHAT DETERMINES THE SIZE
Selected model and quantizationContext length and answer latency targetsNumber of concurrent users at peakDocument volume and ingestion backlogWhether HA or DR is required
The asset stays with the firm. We specify and validate the configuration; the firm buys the equipment and keeps it whether or not the engagement continues. Suitable existing hardware can be used instead of a new purchase.Envelopes verified August 2026. Component pricing is volatile; figures are indicative, not a quote.
NEXT
Then judge us against everyone else.
Opinion 24-1 requires you to investigate an AI vendor before trusting it with client information. We published those questions as a working checklist. Use it on us.
Open the vendor checklist →