How long does it take to set up private AI in an office? Installing the software is quick. Deciding who may see what is the slow part.
Short answer. A working chat on a local model can run as soon as the model has downloaded: Ollama installs with one command, and a 17 GB model takes about 23 minutes at a steady 100 Mbps. Making it ready for an office takes longer, because the work is mostly decisions rather than software: who gets accounts and roles, which documents may be loaded, what is logged and for how long, how it is backed up, and what the policy says. A shared front end such as Open WebUI adds accounts, groups and local document search. ANVE has no delivery date and no installed units yet, so there is no setup time to quote; a paid pilot’s scope and fee are agreed in writing before installation.
01
Three levels
Level
What you get
What it involves
What drives the time
1. One person, one model
A local chat for one technical user
Install Ollama (one command on macOS and Linux, a PowerShell command on Windows) and run a model with ollama run
The download: 3.3 GB for a 4B model is about 4 minutes at 100 Mbps, and 17 GB for a 27B model about 23 minutes
2. A shared chat for staff
A web chat with accounts, groups and document search
Run Open WebUI (its README gives a Docker command), connect it to Ollama, create users and groups, add LDAP or single sign-on if needed
Docker skills, hardware that can serve several people at once, and deciding roles
3. Office-ready
Accounts, a record of use, backups, a written policy and trained users
Everything in level 2, plus logging, backups, network placement, loading and indexing documents, a policy and training
Decisions and approvals, tidying the document set, and testing on real tasks
02
The tasks behind “set up”
Choose where the device sits and how it joins the network.
Install the runtime and the models, and test them on your own tasks.
Create accounts and roles, and decide who may see which documents.
Load and index documents, and check that permissions carry over.
Decide what is logged, who may read the log and how long it is kept.
Set up backups, and test a restore.
Agree how updates are applied and who applies them.
Write the policy: approved tools, forbidden data, what to do after a mistake.
Train staff with real examples.
Run a pilot with real work, then review it.
03
Download time, worked out
Download time is file size in gigabits divided by line speed. A 17 GB model is 136 gigabits: about 1,360 seconds, or 23 minutes, at a steady 100 Mbps, and about 6,800 seconds, or 1.9 hours, at a steady 20 Mbps. Real connections fluctuate, so treat these as minimums. A model can also be copied from another machine on a drive.
04
Where schedules slip
Waiting for approvals on what may be loaded.
A document set with unclear ownership and permissions.
No one owns the system after the pilot.
A model chosen without testing it on the office’s own tasks.
No backup or restore plan until something fails.
05
Where ANVE fits
ANVE is meant to remove the build project, but it is pre-release: no sealed unit exists, the software runs today on a development machine, there are no installed units and there is no delivery date. A demo call is free. A pilot is paid, with the fee agreed in writing before installation. ANVEAI installs updates on site today, signed update packages applied offline by USB are not on ANVE yet, and ANVE does not update itself over the internet. The reservation is ₹999 in India or US$12 elsewhere, refundable in full until you confirm an order in writing.
06
Questions people ask
How long does Ollama take to install?
Ollama’s README gives one install command for macOS and Linux and a PowerShell command for Windows. After that, the time is mostly the model download.
Do staff need to install anything to use a shared server?
With a web front end such as Open WebUI they use a browser. Open WebUI runs as a container and supports LDAP, Active Directory and single sign-on.
Can we pilot before buying anything?
Yes. A laptop with Ollama, or a trial of a shared front end, is a cheap pilot of the idea. Test it on real tasks from your own files before you decide anything bigger.
Who updates it afterwards?
With a self-built stack, your IT person: Ollama and the front end need regular updates, and Ollama has had security fixes, such as the one for CVE-2024-37032 in version 0.1.34.
07
What is not real yet
No sealed ANVE unit exists; every image of the appliance on this site is a labelled render. The software runs today on a development machine. No customers yet, no certifications, no delivery date, and the final price is not announced.
Reservations are ₹999 in India or US$12 elsewhere, refundable in full until you confirm an order in writing (terms). Offices that cannot use cloud AI can apply to be one of ten design partners.
