Private on-premise AI, on a machine in your office.
ZeroOne DOTS AI sets up private on-premise AI: an open-weight LLM on a machine in your office that answers from your documents and works without internet. For firms across India whose data must stay in-house; we work from Surat, Gujarat. From ₹50,000 one-time plus hardware you own. A 27B Qwen model passed 22 of 24 of our business tool-calling tests, running offline on one Mac.
We install the model on a Mac Studio-class machine or your own server, connect it to your documents and the tools your team already uses, and hand it over. In private mode nothing is sent to an outside AI service, and there are no per-use AI fees on what we build. You keep the model, the data and the machine.
Prefer a call back? Leave your number ↓
20+ client companies · 75+ live portals we host · 12 case studies
Updated · Reviewed by Meet Deshani, founder
Trusted AI implementation partner to
- Active Co
- Pioneer Engineering
- Parishram Architecture
20 companies run day to day on software we built and host. Read the case studies →
| Model in our test | A 27B Qwen open-weight model, 8-bit |
|---|---|
| Offline test result | 22 of 24 (91.7%) of our real business tool-calling tests, offline on one Apple M5 Max Mac with 128 GB memory |
| Runs on | A Mac Studio-class machine or your own server |
| Sent to outside AI services | 0 bytes in private mode |
| Price | From ₹50,000 one-time; hardware extra, and you own it; no per-use AI fees |
| Optional support | From about ₹5,000 a month |
| Private mode | On a machine in your office; works without internet; nothing sent to outside AI services |
| Hosted in India | On servers in India under your own web address; needs internet; your data exportable any time |
| Cloud AI tools | On the provider's servers; needs internet; monthly fees that rise with usage |
| Based in | Udhna, Surat, Gujarat; clients across India |
- Chemical, pharma and engineering firms whose formulas, drawings and price lists must stay inside the company
- CA firms, wealth advisors and lenders handling client financial data
- Owners and IT heads asked to bring in AI without sending company data to the cloud
- Offices that need AI to keep working when the internet is down
Staff paste price lists, drawings and client files into free AI chatbots
A private assistant on a machine in your office; in private mode 0 bytes go to outside AI services
AI fees rise with every user and every query
A one-time build from ₹50,000 that you own, with no per-use AI fees on what we build
Generic AI does not know your products, customers or formats
Answers come from your own documents, with the source shown
We map one task and the documents behind it, and tell you whether a machine you already have is enough before you buy anything.
The model goes onto your machine, connected to those documents. Your team asks real questions; answers are checked and corrected.
It goes live inside the tools your team already uses. You own the model, the data and the machine; support is optional, from about ₹5,000 a month.
- Company Brain can answer from a private model on a machine in your office, with search across documents, email and WhatsApp; the 91.7% offline result is published there. Read →
- Our own company runs on an AI operating system: customers, projects, tasks, payments and WhatsApp conversations, operated by AI agents through a controlled interface. Read →
- Aaamdhanee, a gig-worker lender: bank-statement analysis and credit scoring run inside the lender's own private cloud, with no customer data sent to an outside AI API. Read →
- WORO Media: private models, document-aware retrieval and an agent mesh on infrastructure WORO owns, running six brands. Read →
- Meet Deshani, who leads every build, won a government grant for on-device AI hardware. Read →
“The major blocker we had was how to integrate AI-driven bank statement verification and scoring without data leaving our infrastructure and passing all the banking audits. I can't believe Meet executed the whole flow in a single iteration.”
Custom build you own
From ₹50,000 one-time, up to ₹5 lakh and beyond for larger systems. Hardware is extra and you own it; in week one we tell you whether a machine you already have is enough. No per-use AI fees on what we build; optional support from about ₹5,000 a month.
Tell us what you run. We call back within one working day.
Four fields, no account, no newsletter. A person from the team reads it and replies on WhatsApp with a straight answer: what we would do first, roughly what it costs, or “not yet”.
- A reply within one working day, on WhatsApp or a call
- Your details go into our own CRM and to the team's phone, nowhere else
- No drip emails, no follow-up bots
Do we need to buy a GPU server?
Not always. The 27B model in our tests ran offline on one Mac; for your office we size either a Mac Studio-class machine or your own server. In week one we tell you whether a machine you already have is enough, before you buy anything. Hardware is extra, and you own it.
Does any of our data go to OpenAI, Google or another AI company?
Not in private mode: the model runs on your machine and 0 bytes go to outside AI services. Only if you choose Claude or ChatGPT for non-confidential work does anything go out, and then under the same per-person permissions. We never use your data to train tools for other companies.
How good is an offline model at real business work?
A 27B Qwen open-weight model at 8-bit, running offline on one Mac, passed 22 of our 24 business tool-calling tests. They check the calls an assistant makes to read and update records in our own business system, graded on its real schemas without touching live data: inspect before writing, look up IDs instead of inventing them, ask when a request is unclear. Before go-live it is checked on your own tasks and corrected where it misses.
What does private on-premise AI cost?
From ₹50,000 one-time, up to ₹5 lakh and beyond for larger systems, plus hardware you own. Rented AI tools charge monthly fees that rise with users and usage; what we build carries no per-use AI fees. Support is optional, from about ₹5,000 a month.
Will it keep working without internet?
Yes, in private mode, because the model and your documents sit on a machine in your office. A portal we host for you on servers in India, under your own web address, does need the internet.
How does private AI help with India's DPDP Act?
In private mode customer data stays on your machine, nothing is sent to an outside AI service, and each person sees only what they are allowed to. We build with the Digital Personal Data Protection (DPDP) Act in mind, and will sign an NDA before you share sensitive details.
Can the private model work in Gujarati?
Not proven yet. Our published 22-of-24 result was measured on English test prompts, and we have no Gujarati score to quote. In the 30-minute demo we run the model on a sample of your own documents, so you see how it handles Gujarati before you buy anything.
Can we still use ChatGPT or Claude for some work?
Yes. Confidential work goes to the private model on your machine; everything else can go to Claude or ChatGPT through the same permissions, so each person still sees only their slice.
Every service on one page: Services →