🔒 Privacy in the age of AI is very expensive.

I believe we should be aware of the privacy settings in everything we touch with AI.

And even though I have my local agentic AI setup, which I call Argus and which handles my notes, drafts and daily briefs, it is still not 100% private. Because I still use a chat to share my thoughts and instructions, etc., which goes through the cloud. This is not relevant to writing code, but rather to the “personal assistant” that Argus is to me.

So today I moved from the free messaging that is included within my paid subscription to an API per-token payment.The API use runs under Anthropic’s commercial terms, so it isn’t used for training by default, which makes it more private.

But I can see, just after a few hours of testing, that this approach would cost me more than the price of the regular Claude Code subscription I pay on top of it (90ish a month).

The question here is where the line lies between spending too much for your privacy and sharing too much private information.

There are a couple of ways to total privacy that I can see.

🖥️ One is to update the hardware, which would cost me thousands of pounds. Because running local LLMs with reasoning at a relatively equal level to the online versions is expensive. But I don’t have a spare 10k lying around to spend on a nice new setup.

🤐The second option is that I would not share anything personal with Argus. But then I would limit the scope of context it can operate with. Which sort of beats the point of using Argus in the first place.

When I look at the current setup and what it does for me, most of it wouldn’t bother me if it got “leaked”. But more and more I lean towards more useful areas where I would need to share something private like health, finances, etc. And that’s where I stop and ponder where this goes and where I should draw the line.