The one-person case
A single engineer with a Mac Studio or a Strix Halo mini-PC and Ollama gets a private model in an afternoon. Nothing we make beats that for that person. We say so on the pitch and we will keep saying so.
ANVE: the private AI office, on the record. See what that means
Engineering note 02 · 22 September 2026
The most common question after our first note was "why not a Mac Studio, or a DGX Spark, with Ollama?" The honest answer: for one technical person, that is what you should buy, today.
ANVE is for the office around that person. It is the project of turning that computer into something an office can use, and this note is about the difference.
A single engineer with a Mac Studio or a Strix Halo mini-PC and Ollama gets a private model in an afternoon. Nothing we make beats that for that person. We say so on the pitch and we will keep saying so.
Ten to two hundred people, none of whom will open a terminal. Named users through a browser with role-based access, so the CA firm's article clerk does not see the partner's files. An audit trail the institution keeps, because DPDP and departmental rules ask who accessed what. Meetings that become minutes with owners and dates. Search and answers over the office's own records with the source shown. A device that sits next to the router and that nobody administers. Updates that do not require a technical person to be present.
That list is the product. The computer is the part that ships in a box.
Turning a workstation into that list is a multi-month software project: identity, permissions, audit, ingestion, retrieval, a meeting pipeline, an update mechanism, an offline entitlement path for sites with no internet. Most offices that buy "an AI server" end up with one engineer's toy. We have watched it happen.
Because the software carries the value, the entry unit can be a Ryzen AI 300 class part with a 50 TOPS NPU and a far-field microphone array, running a 4B-class model for transcription, summaries, action items and document Q&A. Published target ₹95,000; final price not announced. Bigger tiers move to AMD unified memory (Strix Halo) and RTX PRO for 27B-class models and more concurrent users.
Update, 7 October 2026. We have retired the ₹95,000 entry target: our platform has never been tested on a Ryzen AI 300 part, and at that price the hardware margin was about 13%. The lineup is now ANVE One, Office and Office Pro, with indicative launch prices from ₹2,99,000 excluding GST; final prices are confirmed after our own engineering units are validated. See the lineup.
What a 4B model does not do: deep reasoning, long research, strong code generation, agents. We say that on the box.
The same list, side by side. Where the ANVE column describes something we have not built or cannot yet stand behind, it is marked as a target.
| Item | Computer + Ollama | ANVE |
|---|---|---|
| Setup | An afternoon, by an engineer | 60 secondstarget |
| Users | One, or a shared login | Many named users, with roles |
| Audit trail | None | Kept on the device |
| Meetings to minutes | Do it yourself | Built in, with owners and dates |
| Search over office records | Do it yourself | Built in, sources shown |
| Updates | Manual | Managedtarget |
| Offline entitlement | Not applicable | Not built yet (Track G)target |
| Who fixes it | The engineer | Us, under warrantytarget |
Reservations are ₹999 in India or US$12 elsewhere, refundable in full until you confirm an order in writing. The terms are two screens long and say so. If you run an office like the one in section 02, we are taking ten design partners.
Send this to whoever says "we can't use AI here."