Why we want you to see what your AI knows
Most AI memory is hidden from you. Ours is a folder on your Mac that you can open, read, and fix, and the people who push their agents hardest care about that the most.
There are two kinds of people who buy a Ferrari. The first kind wants the car, the way it looks and the way it feels to drive, and they trust that there's a great engine under the hood without ever needing to look at it, which is completely reasonable. The second kind wants to see the engine and understand what's actually powering the thing, because that's how they know how hard they can push it. They'll never fully trust a car at speed until they've looked inside it.
I think about AI memory the same way, because the memory is the engine. It's the reason an agent can draft the right email without you explaining who the person is, or prep you for a meeting in exactly the format you like, and as agents take on more of your work it becomes the most important part of the whole system. In most of the tools people use today, though, it's hidden. You use ChatGPT or Claude, and somewhere there's a memory of you, but I couldn't tell you what it remembers about me or whether the thing it's relying on is still true. For a lot of people that's the right design, because they don't want to think about any of it. The people who work with agents all day want to look inside, and right now they can't.
A folder you can open
In Turnstone, the Brain is a folder on your Mac. Underneath, it's just folders and markdown files that every one of your agents reads from and writes to, and Brainy keeps them up to date as your work changes. You can open it from the Brain tab and show it in Finder, and what you see there is the same thing your agents see. When something is wrong, you open the file and fix it. It's also your folder, sitting on your machine in plain files that any other tool or agent can read, so if you ever decide to leave Turnstone you take everything you've built with you.
I didn't fully appreciate how much this mattered until I watched it change someone's mind in real time. I was on a call with a user who had been trying the product, and the first thing he told me was that he'd connected his school email but not his work email, because there was too much sensitive stuff in his work account and he wasn't comfortable handing it to an AI app. I asked him to open the Brain tab and show it in Finder, and when he saw it was just a folder on his own laptop, his whole tone changed. He told me he was going to connect his work account right after the call, and all I had done was let him look at the engine.
Why power users need to see it
Black box memory makes power users nervous, and I think they're right to be. Once you hand an agent something like your investor updates or your customer follow ups, you need some way to know what it believes about your work, and "trust us" isn't a great answer. I mean think about it with a new hire. You'd never give someone your most important accounts without knowing what they'd been told about them. You'd want to see the onboarding doc, and if something in it was outdated you'd fix it before they started making calls.
The same thing happens in my own work constantly. I've been putting off my visa paperwork for months, and the other day I finally emailed my lawyer about it. The very next thing I wanted to know was whether the Brain picked that up, and where. Did the note about my visa get updated, or is some agent still going to remind me about a step I've already done? Opening the folder is how I find out. Once I can see what my agents know, I stop guessing and start handing them bigger things, and I end up pushing them a lot harder.
Seeing it also means seeing the mess
There's another side to this I want to be upfront about, because visibility cuts both ways. When you can see your AI's memory, you can also see when it's messy. Before Turnstone, our own company brain was a folder I let agents write into freely, and after a few months it had turned into slop, with hundreds of markdown files everywhere and old decisions sitting right next to the ones that replaced them. Some of our early users have opened their Brain in the first few days and found things it shouldn't have learned, like notes pulled from the wrong account or old pricing from an outdated email thread. Our Brain still doesn't always know what it should and shouldn't learn, and we're working hard on that.
Hidden memory has the same mess in it. You just can't see it, so you only find out when an agent confidently does the wrong thing. With a folder you can open, you can correct the mess in a minute, and every agent works from the corrected version the next time it runs. I'd much rather show you a Brain that's a little messy and let you fix it than hide one that's a little messy and ask you to trust it.
What local does and doesn't mean
Since a lot of this comes down to trust, I want to be precise about privacy, because it's easy to overclaim. Your Brain and your agents live on your Mac, as files you can open and read yourself. But local storage doesn't mean nothing ever leaves your computer. When an agent works on something, the request goes to the model provider you chose, whether that's Anthropic or OpenAI or anyone else you've connected, under that provider's terms, the same way it would if you were using their app directly. The apps you connect are also reached through their own services, under their own terms. So your memory is stored with you and stays yours to read and fix, and the thinking happens with whichever model you picked.
Some people will never open the folder, and the Brain works the same whether or not they look at it. We're building Turnstone for the people who want to see the engine, because they're the ones who end up trusting their agents with real work, and being able to see what's underneath is how they get there.
