Bilbs AI
[ Foundation · entry tier ] 5–10 lawyers

Foundation.

The smaller server. About the size of a desktop tower — it fits in a supply closet. We train it on your firm’s own files and it serves 5 to 10 lawyers. Built for Québec boutiques, notarial offices, and small firms that want a private AI inside the office without a server room. Nothing leaves the building.

See the other tiers
BILBS MINI FOUNDATION · 1U
from $29 / lawyer / month · one line on the invoice Hardware below · audited separately, yours or sourced by us
5 Concurrent users
7 Weeks from call to live
[ Who the Foundation fits ]

Built for boutique and pilot-stage firms.

Boutique practices

5 to 10 lawyers, often one or two practice groups. The kind of firm whose clients have started asking formal questions about AI — and whose associates are already using ChatGPT in browser tabs the firm can’t see.

Notarial & small accounting offices

A notarial office or a small accounting firm lives under the same professional-secrecy rules and the same Loi 25 exposure as a law firm. Often 8 to 25 clerks and professionals. The Foundation tier is sized exactly for them.

A pilot for a larger firm

A larger firm wants to try a private AI in one practice group — litigation, real estate, M&A — before the partnership signs off on a firm-wide rollout. We install the Foundation server for that group. If the partnership says yes, we trade up to the bigger server and credit 75% of what you already paid.

In-house legal departments

A corporate legal team of 10 to 20 in-house counsel handling contract triage, policy, and regulatory review — on a server that lives inside the parent company, never connected to anything outside the building.

[ Hardware specifications ]

For the firm’s IT director.

Form factor Tower with rack-ear kit (~1U)
GPU 1x NVIDIA RTX 5090
32 GB GDDR7
CPU AMD Ryzen 9950X
16 cores · 32 threads
RAM 128 GB DDR5
Primary storage 8 TB NVMe (RAID 1)
Model storage 2x 4 TB enterprise NVMe (mirrored)
Networking 2x 10 GbE SFP+
Power 1x 1000W 80-PLUS Platinum

Power & Environment

Idle power 150W
Inference load 260-380W
Peak / fine-tune 520W
Acoustic <45 dBA

Quiet enough for under-desk placement

Standard office outlet (110-240V / 15A). Recommended UPS: 1000 VA for 15 min runtime.

[ Model performance ]

What you can run on it.

Gemma 4 4B (FP16)

~110 tok/s
Fast intake · classification

Gemma 4 12B (FP16)

~55 tok/s
~2.2k tok/s prefill · default serving model

Gemma 4 12B (INT4)

~95 tok/s
Quantised for higher throughput

Gemma 4 Embedding

~9k tok/s
Retrieval over the DMS

Gemma 4 LoRA (firm)

~55 tok/s
Fine-tuned on the firm’s precedents

Simultaneous hosting

Gemma 4 12B serving model + Gemma 4 Embedding + a small reranker. Model swap takes ~90 seconds via admin UI.

[ Concurrency envelope ]

5 concurrent means more than you think.

"Concurrent" means actively generating tokens at the same wall-clock moment, not "users who have accounts". A 25-person team where 5 people are typically prompting at peak lunch-hour fits the Mini comfortably.

Chat (Gemma 4 12B, ~1k token prompts) 5 users
Single-step agents (tool-use) 3 users
Ask the firm’s files a question 4 users
Batch inference (overnight) Unlimited
Streaming transcription 2 streams
[ Pricing ]

One line. From $29 per lawyer.

Bilbs subscription · per lawyer, per month
from $29
/ lawyer / month · in Canadian dollars · goes up only if the firm adds optional services
Free audit · we scope the firm before you sign anythingIncluded
Deployment, indexing, on-site trainingIncluded
Updates, support, audit log, 24/7 pagerIncluded
Hardware (this Foundation spec)Yours, or we source it
Entry hardware spec
The server itself: one NVIDIA RTX 5090 graphics card (32 GB), AMD Ryzen 9950X processor, 128 GB of memory, 8 TB of fast NVMe storage in RAID 1
6-week deployment (fine-tune, indexing, on-site training)
Trained on every contract, memo, and matter your firm has handled
Admin UI, orchestration layer, eval harness
3-year OEM warranty + printed runbook
Hardware path A · you have one

Use your server

If the boutique already runs GPU-capable hardware in the back office, we deploy onto it. Nothing extra on the Bilbs line. The Foundation spec on this page is a reference for what works comfortably for 5–10 lawyers.

Hardware path B · you don’t

We source it

Most boutiques don’t have an IT director. After the audit we source the Foundation-spec server, install it on-site, configure the network, and own the runbook. Hardware sits on a separate, transparent quote at OEM list price — your call once you see the spec.

If the firm grows

No penalty

If the firm grows past 10 lawyers, we swap the chassis for the Practice spec and re-train on the new hardware over a weekend. No re-platforming charge. The per-lawyer line stays the same.

[ Timeline ]

Contract to live in ~7 weeks.

Step 1

Free 45-min call

Week 1
Step 2

It learns your firm

Week 2–4
Step 3

Server is built

Week 3–5
(in parallel)
Step 4

Server arrives

Week 5
Step 5

We install it

Week 6

Need it faster?

We keep two Foundation servers on the shelf at all times. With a faster two-week training, you can be live in three weeks instead of six — for an extra $1,999.

[ When to upgrade ]

Know when the firm has outgrown Foundation.

Protect a boutique firm’s files.

A 45-minute call within one business day. We’ll tell you whether the Foundation tier is the right size for your firm — and if it isn’t, we’ll tell you which one is. No pitch. No deck. No payment today.