I, for one, welcome our new AI overlords

I never thought it would happen to us.

Ok, that’s a lie. Since the very beginning, I always knew it would happen to us.

All I can say is, I was never going to dive head first into “AI-powered” features for beanies just because it’s there. The beanies.family platform (or any software, for that matter) should have a reason to exist on its own merit, regardless of AI.

AI should never be the reason you use software. It should be the reason the software you use works a bit better.

The question for me was how I could leverage AI capabilities in a fun and engaging way, while staying true to the overall beanies philosophy. AI is not who we are, it is just something we do.

As many of you may know, if you’ve ever read a single word of any of my blogs, AI-assisted coding has helped make beanies the amazing software it is today. “Vibe coding” your app in a careful and methodical way is one thing, but using AI to read precious (and personal) user data is something else altogether. beanies is a private, secure, personal-data-protecting platform - I’m not forfeiting that principle simply to boost usability.

For AI to work for beanies, the stars need to align. I tend to think a lot (too much, some might say) about these decisions. For me, the stars aligning means: (a) the right use case, (b) the right AI/LLM privacy-preserving platform, (c) the right people running that platform, (d) a data protection agreement we can trust, and (e) probably more stuff that I’m not thinking of right now.

Every time data is sent to an LLM to be processed, that’s a risk, regardless of where that data is hosted. Every company has different terms, privacy policies, and (broadly speaking) their own set of ethics and morals. There is no standard in the wild west of the internet, and that’s even more true in the burgeoning frontier of artificial intelligence.

If you don’t read those agreements to the letter and negotiate for your principles, you may never truly know what will happen with your data.

wait, did i agree to this?

Some companies use your data to train their models, others use it for ad targeting, still others actively sell your data to advertisers, and who knows what else. There’s a million things your precious family data could be used for, but only one thing that I want it to be used for: to let beanies make your life easier.

Got it - so how can we take advantage of AI while still keeping our data safe?

When I (or my claude-bot, actually) discovered our privacy-preserving AI LLM provider, it was like a lightning bolt hit me in the head - a privacy-focused, secure-enclave, multi-model providing AI LLM platform. While they (perhaps understandably) charge more on a per-token basis than the big tech providers, they not only provide terms and a privacy policy that are far more amenable to privacy-concerned platforms, they actually agreed to work with a tiny startup like ourselves to negotiate a customized DPA (a data processing agreement, for those not immersed in the AI SaaS world), despite the fact that we’re delivering roughly the same level of traffic to their platform as a family of ants marching triumphantly across an interstate.

In this particular moment of AI, this is rare and refreshing to see in the world of data-harvesting LLMs. For those interested, some more information about Tinfoil, our AI data processor, is here.

Tinfoil guarantees, in writing, that your data will be kept private, processed in a secure enclave that isolates memory and processing known as a Trusted Execution Environment (TEE), and that no data will be retained. That means it cannot be used for training models, selling your info to advertisers, or any other purpose. The models are used for one thing only: extracting information from your precious family plans and sending that back to us to help make organizing your life easier.

I had a lovely conversation with both Sacha and Dani from the Tinfoil compliance team, who were both very accommodating of our requests. After reading more about beanies.family, Sacha was even kind enough to provide an awesome quote for this blog:

“We’re always excited to have companies like beanies.family using Tinfoil! It makes us very happy to see private and secure AI being used for cool use cases like family organization, especially when the companies building on our platform share our values and vision for AI safety.”

- Sacha Servan-Schreiber, Co-founder, Tinfoil.sh

We indeed have a kindred connection - building platforms that preserve privacy in this world of almost ubiquitous relinquishing of the same.

Depending on the model used, a token on Tinfoil costs roughly 5 - 10x the price of a token on a platform that harvests your data for advertising and/or training. For now, that’s the price that we (or, I, I should say) pay for your privacy. It’s worth it, at least to me - and hopefully to you too. Because, one thing I can be sure of is that I won’t be able to subsidize this thing forever :)

So, say it with me: We’re now beanies.family - powered by Tinfoil AI!

Ok - so how can I actually use this thing?

Staying true to our beanie selves, and because I also loved fairy tales as a kid, we’ve playfully named any AI-powered feature on the platform magic beans. So, just look for that.

the magic beans icon means AI we can trust. actually.

We started with two magic beans features:

(1) Activities

Take a picture or load an image straight into your activities, and all of the details are automatically created in your calendar (and pushed to your connected calendars, if you have that set up)!

(2) Travel Plans

Read in your travel plans - itineraries, screenshots, emails, whatever you have - just load it into beanies AI via the magic beans button, answer a couple questions (like who is actually travelling), and your travel plans will magically appear under a new (or existing) travel plan!

I’ve used both of these features now for a couple weeks, and take it from me - the work was well worth it. These features truly are amazing time-savers (you can even ask my wife - who HATES typing).

That’s amazing - what’s next?

Thanks, greg! I think so, too.

I’ll be following this up with blogs and stories about these (and more) exciting new features, including the ability to BYOK, starting with OpenAI - meaning you can plug in your own personal key, and your data goes directly to your provider, never passing through beanies (with more providers to follow).

In addition, I’m also working on improving privacy further for documents sent to our beanies AI provider, adding end-to-end encryption so that only the secure AI hardware can open them - meaning they could not even be read in the split second those documents pass through our own server.

For now, I just wanted to touch base with my precious beans about this new capability. I’m psyched about it, and I hope you are too!

Until next time, stay AI-safe, my beans

greg

By the way, if you think this is cool, let me know (or tell me what to build next!) on discord.