Local AI Setup puts working private AI on a machine you own, tests it on one task you actually recognize, and leaves you with a written handoff instead of a rabbit hole. This note walks through the whole service exactly as it runs: what happens before, during, and after, what each level costs, and where the boundaries are drawn. It is written for one owner deciding, not for an IT department.

Before anything is installed: preflight

Every engagement starts by checking the machine you already have, your comfort level, your actual use case, and what should not touch local AI yet. The point is to stop guessing from YouTube videos. Some computers are ready today. Some need a small hardware step. And some use cases simply should not run locally, and you hear that plainly.

The setup itself

On a machine that passes preflight, the work is practical and visible: install and configure a local AI app that fits your hardware, download two or three appropriate models for that machine, and connect one safe workflow using non-sensitive work. Not twenty experiments. One workflow you recognize from your own week, so the result is not a demo, it is yours.

After the session you get seven days of async support for install-related issues, so the setup survives contact with your real schedule.

What you walk away with

  • A working local AI app on your machine, configured, not just downloaded.
  • A short model list chosen for your hardware, so you are not chasing every release.
  • One workflow example tested on your own kind of work.
  • A written handoff: what runs locally, what still belongs in cloud tools, where to stop, and what to do next.

The levels, and what each costs

  • Local AI Setup Hour, $175. A 60-minute remote working session. Hardware and use-case check, then a written recommendation: local, cloud, hybrid, or stop. The cheapest honest answer available.
  • Local AI Starter Setup, $450 beta / $650 standard. Everything above, for one person on one existing machine.
  • Private AI Workbench, $1,250. The setup connected to real repeated work: two or three workflow templates for non-sensitive work, a model and use-case matrix, an update and troubleshooting note, and a written or recorded walkthrough.
  • Small Team Local AI Kit, $2,500 to $4,500. For 2 to 5 people who want a shared local AI surface. Requires a written scope and a named admin, and includes thirty days of setup support.

The boundaries, stated up front

Hardware is separate and stays yours. The starter service does not ingest private client, customer, medical, legal, financial, or regulated data; sensitive workflows need a separate written scope. No public server exposure without a technical scope. No magic model promises: local models are useful and measurable, and they do not always beat cloud tools. If you want the fuller reasoning, read private AI vs a ChatGPT subscription and what forward-deployed AI means for a small business.

How reliability is measured, not promised

The method behind this work is public. The delegation-bench benchmark asks which jobs can be safely handed to a local AI and walked away from, and answers with certified reliability floors over 495 trials, including a sabotage test the AI is supposed to refuse. The same lab publishes its measured results for a $1,400 mini-PC, from 93 percent HumanEval scores to a complete serving story with raw logs. When a PuenteWorks setup tells you a task is reliable enough to hand over, that claim has a method behind it you can check yourself.

FAQ

Do I need to buy a new computer first?

No. Start with the Setup Hour on the machine you have. If a hardware step makes sense, you hear it with numbers before spending anything.

Will my data stay private?

That is the point of the setup, and the starter service also draws a hard line: no private client, customer, or regulated data until there is a written scope designed for it.

Is this a one-time thing or ongoing?

The starter setup is one-time with seven days of support. Ongoing monthly support exists as a separate path when a business wants it.