I care about privacy and running things myself

Leah exported patients_q2.csv (2,118 rows, dates of birth in column D) to “just get a summary” in a free personal chat on her phone in the cafeteria. The model was helpful. The problem was the room, the account, and the file. Legal asked later whether the vendor had a contract. The answer was a shrug and a screenshot. The file never needed to leave the laptop. A local model would have been slower and worse at prose. It would not have turned a lunch table into an incident.

This is Part 4 of Which AI product should I use? You care about privacy, or you want to run things yourself. That is not the same as “I heard open source is free.” Closed chat apps, hosted open-weight APIs, and a model on your machine are three lanes. What open means (OS1) names the license gap. This part is the chooser: whose laptop, whose account, would you email this to a vendor. If the honest answer is “I wanted a cheaper ChatGPT,” say that. Cheap and private are allowed to be different tickets. You can buy both. You cannot pretend they are the same SKU. Name which ticket you are buying.

What you’ll learn

Three privacy lanes: closed products, hosted open models, local runners
Why “open” does not mean “private,” and why “free” does not mean “yours”
A paste checklist you can run in ten seconds
When a $20 work plan beats a heroic local setup
What to try first if the real requirement is “this file stays here”

Three lanes

Three privacy lanes: closed chat apps, hosted open models, local on your machine. Would you email this file to that vendor.

Lane	What you get	Where the text goes	Pick it when
Closed chat (Claude, ChatGPT, Gemini, Grok)	Best everyday quality for most people	Vendor cloud, under that product’s policy	The content is allowed in that product, ideally a paid work plan
Hosted open model (Groq, Together, a cloud Llama, etc.)	Often cheaper APIs, downloadable cousins exist	Still someone else’s GPU	You want a model family, not a laptop project
Local (Ollama-style, llama.cpp, a GUI)	Control, offline, no cafeteria upload	Your disk, your RAM, your fan	The file must not leave, and you will maintain the stack

Leah’s file belonged in lane three or in a contracted work workspace, not in a personal free tab. Hosted Llama is still a vendor. Downloading weights later does not un-send Tuesday’s CSV. For the vocabulary of weights vs licenses, read OS1 after this page.

Before you paste

Before you paste a file: whose account, would you email this to a vendor, can you strip names, do you actually need local

Whose account? Personal free chat is not your hospital, school, or company.
Would you email this CSV to a random vendor? If no, do not paste.
Can you strip names and dates of birth first? A 20-row toy beats 2,118 real people.
Do you need local, or do you need a contract? Comfort is not a threat model.

Rule of thumb: Privacy is a location and a contract, not a brand personality. “It feels trustworthy” is not a control.

When local is the wrong trophy

Local models cost hardware, heat, and your time as the patcher. A 7-billion-parameter model on a laptop can summarize a meeting note. It will not reliably match a frontier chat plan on hard writing. If the real requirement is “good prose, allowed data,” a work Claude / ChatGPT / Gemini / Grok plan with admin controls may beat a weekend of Ollama. If the real requirement is “this CSV never leaves,” you accept the quality drop or you don’t use a model.

Practical AI on this site already covers what you should not paste for analysis. Same muscle, different door.

A 15-minute local smoke test (optional)

Only if you already decided the file must stay. This is not a full Ollama tutorial. That series comes later.

# Example shape only. Package names and flags move.
# 1) Install a friendly runner (Ollama-style) from its current site.
# 2) Pull a small model you can actually run (think 7B-class, not a 70B hope).
# 3) Chat with a toy paragraph, not patients_q2.csv.

# Fake check: if Activity Monitor shows the fan and the reply is slow,
# you learned the hardware tax. That is the point of the smoke test.

What that smoke test is for: feeling the fan before you promise legal a local stack. If you will not maintain updates, stop. Use a contracted work chat or do the summary yourself.

Worked example: 2,118 rows in a cafeteria

Choice	What leaves the laptop	Quality of the summary	Incident risk
Personal free chat	The CSV	High	High
Work workspace, allowed data	The CSV, under a contract	High	Lower if policy matches
Toy 20-row strip, then any chat	Fake names	Good enough to learn	Low
Local small model on the real file	Nothing	Uneven	Low on network, still a stolen-laptop problem

Leah’s version was row 1. The try-this-first for a regulated file is row 3 or 4, never row 1 on public Wi-Fi.

Work plans are a privacy tool too

Teams sometimes treat “we refuse all vendors” as the only moral position, then someone pastes anyway on a phone. A paid work workspace with a data processing agreement, SSO, and an admin who can turn off training-on-chats (when the vendor offers that control) is a real control. It is not as strong as air-gapped local. It is much stronger than Leah’s cafeteria tab. Read the current admin docs. Do not quote a Twitter thread from 2024 as the policy.

Personal projects are allowed to be sloppy in a different way. Your recipe blog is not 2,118 dates of birth. The chooser still asks you to pick a lane on purpose. If you want local for learning, great. Feel the fan. Then decide. The later local-LLM series will spend pages on Ollama, llama.cpp, and GUIs. This page only has to stop the unexamined paste.

Hosted open models, without the halo

A DeepSeek or Llama endpoint at Groq, Together, Fireworks, or a cloud you already pay for can be cheap and fast. You still sent the text off-device. The license on the weights does not travel to the host’s logs. If the reason you wanted “open” was privacy, a hosted open model missed the point. If the reason was cost or a model family you like, it may be the right lane. OS1 is where the license vocabulary lives. Keep the lanes honest here: location of the bytes, then license of the file.

Common mistakes

Calling a free ChatGPT tab “private” because you deleted the thread. Deletion is not a legal theory you should invent.
Assuming Llama on a host is local. The GPU is still in someone else’s building.
Promising legal a local stack you will not patch.
Pasting dates of birth to “make the summary more personal.”
Skipping OS1 and calling every downloadable model “open source.”

A copy-paste policy you can actually use

Teams stall because they wait for a 40-page AI policy. You can ship a 12-line one this week and let legal thicken it.

Paste policy (v0)
Allowed in personal free chats: public text, toy data, your own shopping list.
Allowed in work workspace only: internal drafts without regulated fields.
Never: dates of birth, patient/student IDs, payroll, unpublished financials,
       other people's inboxes, secrets, API keys.
If you cannot strip it, do not paste it.
Local models: only on company laptops if IT says so; still encrypted disk.
When unsure: ask [name], do not test in the cafeteria.

What that policy does: it names lanes without pretending a Llama icon is a control. Put it in the same place as the 10-line chooser note from Part 6. Review quarterly because vendors change training-on-chat toggles. Leah’s 2,118-row file fails every line after “Never.” That is the point of writing it down before the export.

Travel and coffee shops make people sloppy. If you must work on a train, work on the toy file, not the real one. A privacy screen helps shoulder surfers. It does not help the POST request that already left the phone. Location of the human and location of the bytes are different maps. This part is about the bytes. Stolen-laptop risk still exists for local models: encrypt the disk, use a firmware password if you know how, and do not leave patients_q2.csv on an unlocked cafe table next to the “private” llama. Local is not magic. It is a smaller room with a lock you have to remember to use.

How to practice this week

Take one file you almost pasted. Strip it to a toy. If you cannot strip it, you cannot paste it. Next: Part 5 for images and live answers, or OS1 if you want the license vocabulary. Decision tree: Part 6.

Quick recap

Closed, hosted-open, and local are different places for your text.
Would you email this to a vendor. If no, do not paste.
Local is control plus homework. It is not a free lunch.

Sources

Vendor privacy / terms pages for Claude, ChatGPT, Gemini, Grok (re-check; they change)
Open Source Initiative: Open Weights
OSI: Open Source AI Definition
AMS: Practical AI

Keep going

Same lessons in your feed

Short diagrams and hooks on Instagram, X, and Facebook.

@analyticsmadesimple Follow @AnalyticsSimply Follow Analytics Made Simple Follow