Hello, we are the artificial intelligence
for law firms.
And we never send a single document outside your office. AG Core arrives with your trade's software pre-installed and an AI model that works locally: you switch it on, you work — and every contract, every file, every question stays within your walls.
The bill no one has shown you yet
Cloud AI software is paid for three times over: in francs, every month, for every member of staff; in data, because every uploaded document travels to someone else's servers; and in dependence, because when the provider changes its prices or terms, all you can do is accept. For a firm bound by professional secrecy and the Swiss FADP (nLPD), the second item is not a cost: it is a risk.
Documents travel
Every contract uploaded to an American AI service passes through servers you do not control, under jurisdictions you did not choose.
The cost grows with the firm
USD 300 to 500 per professional per month is the going rate for specialised AI platforms. Ten people, another salary a year.
The logic is reversed
The machine is yours, the software carries no per-user licences, the AI works locally. You pay once for the machine, plus an annual fee for maintenance and updates.
The products
Eight systems, each built for a trade. Every page includes a demonstration you can try right away, with fictitious data: your real documents will never be seen even by us — they stay on your machine.
Fiduciaries · Notaries
Law firms
Every office
Every person
Logistics
Hospitality
Bookkeeping
Sales
AG Core — the machine
Not just any computer: an intelligence machine sized for your work, silent, no bigger than a book. What sets the configurations apart is memory — more memory, more capable models. We make that choice together, after the demo.
| Configuration | Designed for | What it does well |
|---|---|---|
| Core One | Reception and front desk | Answers for the public, FAQ, one document at a time |
| Core Studio | The individual professional | Contract analysis, drafts, personal archive |
| Core Team | Firms of 3 to 10 people | Full firm archive, several users at once |
| Core Max | Firms of 10 to 30 people | Higher-tier models, intensive workloads |
| Core Enterprise | Companies and multi-site | Dedicated server, multiple working groups |
Machine price, installation and annual fee depend on the configuration and software: we put them in writing after the demo, item by item, tailored to your firm.
The technical sheets
Power figures are the processor's rated TDP (average consumption in the office is lower); the final configuration comes in the written offer.
Core One — the silent front desk
Processor Intel N100 · TDP 6 W
Memory 16–32 GB DDR5
Storage 512 GB NVMe
Cooling passive (fanless): zero noise at reception
Models 7–13 billion parameters
Typical use answers for the public, FAQ, one document at a time — Eco
Core Studio — the individual professional
Processor AMD Ryzen 7 8745HS or Intel Core Ultra 7 · TDP 28–54 W
Memory 32–64 GB LPDDR5
Storage 1 TB NVMe Gen4
Cooling low-noise active, desktop format
Models 13–27 billion parameters
Typical use contract analysis, drafts, personal archive — Eco, Hermes, Themis, Plutus, Iris
Core Team — the firm of 3 to 10 people
Processor AMD Ryzen AI 9 HX 370 · configurable TDP 15–54 W
Memory 64 GB unified LPDDR5X (shared CPU/GPU)
Storage 2 TB NVMe Gen4
Cooling active, designed to sit in an office, not a server room
Models 32–70 billion parameters
Typical use full firm archive, several users — the whole suite
Core Max — the intelligence workstation
Processor AMD Ryzen AI Max+ 395 ("Strix Halo") · configurable TDP 45–120 W
Memory 128 GB unified LPDDR5X · 256 GB/s bandwidth
Storage 2–4 TB NVMe Gen4 (the engine uses the disk as a memory extension)
Cooling active, sized for continuous workloads
Models large MoE models, quantised (Q2) · ~11 tokens/s measured on a 235bn MoE
Typical use near-frontier quality: DeepSeek V4 Flash, GLM-5.2 — intensive workloads, several users
Core Enterprise — beyond the desk
Architecture two Core Max units in distributed inference (high-speed link),
or a workstation with a dedicated GPU
Models beyond the capacity of a single machine, distributed inference
Typical use companies of 30+, multi-site, multiple working groups
Note fully bespoke configuration: sized on actual use, not from a catalogue
The technology inside AG Core
No black box: this is what makes the products run — and the reason a desktop machine is now enough where a datacenter used to be required.
Native, open source, written for a single job: running large language models on an office machine. It exposes APIs compatible with industry standards — the vertical software connects as if it were talking to the cloud, but everything stays in-house.
Processor and accelerator share a single memory pool, up to 128 GB on the Core Max: it is what lets enormous models sit on a desk, in silence.
When memory runs out, the engine uses the NVMe drive as an extension: models larger than RAM and contexts of up to one million tokens — hundreds of pages in a single question.
How intelligent they are: the cloud versus your own office
Intelligence index — higher is better
Eleven points separate the best of the cloud from the best you can install in your office. For reading contracts, preparing drafts and querying an archive, it is a difference you will not notice — the one you will notice is where your data lives. Source: Artificial Analysis Intelligence Index, as recorded by BenchLM, August 2026. Local inference with ds4, an open-source engine (MIT).
On the GLM line. GLM-5.2 (June 2026, MIT licence) remains supported and is the model installed on a Core Max today. The current line of the family is GLM-5.3-Flash, weights published on 25/08/2026, also MIT. It has no independent measurement in this index yet: that is why it is not in the table, and we do not lend it another model's number. When the measurement exists, it will be added with its date.
What model size each Core can load
Billions of parameters loadable per configuration
Core Enterprise goes further, distributing the model across several machines.
| Configuration | Loadable models | In practice |
|---|---|---|
| Core One | 7–13 billion parameters | Answers for the public, one document at a time |
| Core Studio | 13–27 billion | Contract analysis, drafts, personal archive |
| Core Team | 32–70 billion | Full firm archive, several users |
| Core Max | large MoE models, quantised | DeepSeek V4-Flash or GLM-5.2 in Q2 |
| Core Enterprise | beyond, across several machines | Cluster, multiple working groups |
Parameters are the size of the model: more parameters, more reasoning capacity. We load open, verifiable families — DeepSeek, GLM, Qwen, Llama — and when a better generation is released, the upgrade arrives with the maintenance. Without touching your data.
«MoE» means mixture of experts: the model has hundreds of billions of parameters — 304 billion for DeepSeek V4-Flash, 753 billion for GLM-5.2 — but activates only a part of them for each answer. That, together with quantisation (the weights file is reduced to the Q2 format), is the technical reason a model of that size fits on a desktop machine today and not in a datacentre. Precisely: the full uncompressed file does not fit in 128 GB — what gets installed is the quantised version. Parameter counts read at the source (HuggingFace, cards deepseek-ai/DeepSeek-V4-Flash and zai-org/GLM-5.2) on 02/09/2026.
How you get to switch it on
Live demo, on your case
Half an hour, at your office or on video: we bring a running machine and try it on a case similar to yours. If it is not right for you, we will be the ones to say so.
Delivered pre-configured
AG Core arrives with your trade's software installed, your models loaded and the archive ready to receive documents. Plug it in, switch it on, get to work.
Swiss maintenance
Software and model updates, monitoring, support in Italian and German. A contact who knows your installation, not a call centre.
What the rules ask of anyone using AI on client data — and why a machine in your office simplifies every answer.
The questions everyone asks us
Do my documents leave my office?
No. The AI model runs on the machine inside your firm: documents, questions and drafts never pass through any external server. It is the reason AG Core exists.
Do I need an internet connection?
Not to work: the intelligence runs locally. The connection is only used when we perform maintenance or updates — and that happens at your invitation, never behind your back.
What artificial intelligence is inside?
Latest-generation open-source models, selected and updated by us according to the configuration. They are the same family of models the labs publish openly: you can have them audited by anyone you choose.
And the software: why don't I pay licences?
Because we install professional open-source software, public and verifiable. You pay for the real work: the machine, the tailored installation and the maintenance. Not a monthly licence for every member of staff.
How much does it cost?
It depends on the configuration and the software. We publish no price list because every installation is sized to the firm: after the demo you receive a written offer, item by item, with no surprises.
What happens if the machine breaks down?
The maintenance fee covers monitoring and intervention. Your data sits on the machine's disk and in the backups defined together at installation — the backup, too, stays wherever you decide.
Do I have to replace the programs I use today?
No. AG Core works alongside your current tools: it reads documents from your archive and produces ordinary files — PDF, Word — that remain yours.
Does the AI decide anything on its own?
No. It prepares, flags, proposes. Every deed, every email, every decision goes through a person. It is a rule written into the software, not a promise.
A live demo, on your case
Half an hour at your office or on video. We bring a running machine and try it on a case similar to yours: you see with your own eyes what it does — and what never leaves the room.
Prefer to talk right away?
Write or call: you will reach a person who knows the machines, not a switchboard.
WhatsApp +41 79 863 37 88 Tel. +41 91 814 86 00