Conectao Core is an intelligence machine the size of a book. It arrives with your trade's software already installed and an AI model that runs locally. You switch it on and work — and every booking, every contract, every guest conversation stays inside your company.
It is that in 2026 the bill stopped being predictable. Between March and August, all four major AI providers moved in the same direction — away from a fixed price per seat, towards a price that follows consumption. For a company that budgets once a year, that is a different kind of problem from a high price.
Anthropic removed bundled tokens from its Enterprise plan; Google moved Gemini Enterprise to pay-as-you-go with no base fee. What you spend next year depends on how much your team works — and on routing decisions you do not make.
Per user, per month, indefinitely, in dollars. The tenth colleague is another subscription, not a marginal cost.
The machine is yours, the software carries no per-user licence, the model runs locally. One capital cost, then an annual maintenance plan you know in advance. One more query does not cost one more query.
Public list prices in USD, read on 30/08/2026. Detail and sources in the article on the cloud bill.
Seven systems, each built for a trade. They run on the same machine and share the same archive — you start with one and add the others as the business needs them.
Your hotel's digital concierge: answers guests on WhatsApp in their own language, day and night, updates the system and hands over to a human what needs a human. The conversations stay in the hotel's records.
The office assistant: summarises minutes, prepares drafts, finds anything in the company archive. Works even with no internet.
Loads, trucks and drivers in a single view, with an AI dispatcher that proposes the assignments — and one empty run fewer.
The colleague who reads every contract twice: risk clauses flagged with the exact page, answers with citations, revision drafts ready.
The complete case-file office: guided opening, client verification, list screening, contract ready for signature.
Receipts and invoices read, extracted and posted as they arrive — with the export ready for your accountant. The shoebox of receipts stops existing.
Contacts, deals and reminders in a single view, on your machine. No customer forgotten, no list on somebody else's server.
Detailed product pages are currently published in Spanish: Calypso · Hermes · Atlas · Themis · Minerva · Plutus · Iris. The live demo is run in English.
Not an ordinary computer: an intelligence machine sized for your work, silent, the size of a book. What separates the configurations is memory — more memory, more capable models. We choose together after the demo.
| Configuration | Built for | What it does well |
|---|---|---|
| Core One | Reception and front desk | Answers, FAQ, one document at a time |
| Core Studio | The individual professional | Contract analysis, drafts, personal archive |
| Core Team | Teams of 3 to 10 | Full company archive, several users |
| Core Max | Hotels and companies of 10 to 30 | Top-tier open models, heavy loads |
| Core Enterprise | Large, multi-site companies | Dedicated server, several teams |
Machine, installation and plan depend on the configuration and the software: we put it in writing after the demo, line by line. We do not publish a price list, because every installation is sized on the business.
No black box. This is what makes the products work — and the reason a desktop machine is now enough where a datacentre used to be required.
Native, open source, written for one job: running large language models on an office machine. It exposes OpenAI- and Anthropic-compatible APIs — the software connects as if it were talking to the cloud, but nothing leaves the building. We use ds4, MIT licensed.
Processor and accelerator share a single pool of memory, up to 128 GB on the Core Max. That is what lets very large models fit on a desk, silently.
When memory runs out, the engine uses the NVMe as an extension: models larger than RAM, and contexts of up to one million tokens — hundreds of pages in a single question.
| Model | Where it runs | Intelligence index |
|---|---|---|
| Claude Opus 5 | cloud | 63.0 |
| GPT-5.6 | cloud | 58.9 |
| Gemini 3.7 | cloud | 56.0 |
| DeepSeek V4 Pro | Conectao Core Enterprise | 53.2 |
| GLM-5.2 | open weights — outside our range | 52.6 |
| DeepSeek V4 Flash | Conectao Core Max | 51.8 |
Eleven points separate the best of the cloud from the best you can install in an office. For reading contracts, preparing drafts and answering questions about an archive, that is a gap you do not feel — what you feel is where your data is. Source: Artificial Analysis Intelligence Index, BenchLM measurement, August 2026, read on 30/08/2026. Full reading in the article on the eleven points.
The GLM family has since moved on: GLM-5.3-Flash (weights published 25/08/2026, MIT licence) is the current line and the engine supports it. It has no independent measurement in this index yet, so it is not in the table — we do not lend it another model's number.
A page that only lists advantages is no use for deciding. These are the cases where we would tell you not to buy — and we say the same thing in the demo.
A handful of queries a month never pays back a machine. A cloud subscription is cheaper and starts today.
If your task genuinely lives inside those eleven points, the cloud is the correct answer. The way to find out is to test your real case, not to read a table.
If nobody decides who may see what, AI on top of it amplifies the disorder. The archive comes first — and if the process will change within six months, stabilise it before automating it.
Half an hour, at your premises or by video: we bring a machine that is switched on and test it against a case like yours. If it is not for you, we say so.
Conectao Core arrives with your trade's software installed, the models loaded and the archive ready to receive your data. Plug in, switch on, work.
Software and model updates, monitoring, support over WhatsApp. One person who knows your installation, not a call centre.
No. The AI model runs on the machine in your office: documents, conversations and drafts do not cross an external server. That is the reason Conectao Core exists.
Not to work: the intelligence is local. The connection is only used when we carry out maintenance or updates — and that happens with your permission, not behind your back.
Latest-generation open-weight models — the DeepSeek, GLM, Qwen and Llama families — chosen and updated by us according to the configuration. An open model, once downloaded, cannot be withdrawn, repriced or changed underneath you.
Because we install professional open-source software that is public and verifiable. You pay for the real work: the machine, the tailored installation and the maintenance. Not a monthly licence for every employee.
It depends on configuration and software. We do not publish a price list because every installation is sized on the business: after the demo you receive a written quotation, line by line. Any Conectao AI price figure that does not come from a signed quotation is not ours.
The maintenance plan covers monitoring and intervention. Your data sits on the machine's disk and in the backups defined at installation — the backup also stays where you decide.
No. Conectao Core works alongside your current tools: it reads the documents in your archive and produces ordinary files — PDF, Word — that remain yours.
No. It prepares, flags and proposes. Every message to a customer, every contract, every decision goes through a person. That is a rule written into the software, not a promise.
Half an hour at your premises or by video. We bring a machine that is switched on and test it against a case like yours: you see with your own eyes what it does — and what does not leave the building.