It Reads Your Screen

On Monday I told you to go find out what your connected tools can reach, and warned you off anything holding broad standing access to your work.

Today I am reviewing a tool that watches your entire screen.

I am aware. Let us deal with that directly rather than pretend it is not the first thing you thought when you read the headline, because the tension is real and it is the most interesting thing about this product.

Littlebird is a macOS assistant that runs continuously in the background and builds a persistent memory of your working day. It came out of San Francisco earlier this year with an eleven million dollar seed round behind it, founded by Alexander Green, Alap Shah, and Naman Shah.

The premise is a single frustration everyone reading this has felt. You sit down to ask a question and first you have to explain the situation. What the project is, who the client is, what was decided on the call, what you already tried. The assistant is capable but amnesiac, and you are the one paying the cost of that in typing.

Littlebird's answer is to already know, because it was there.

The technical detail that separates it from the last generation of tools in this space is important. It does not screenshot. It reads structured text out of the applications you have open. Documents, browser tabs, messages, whatever is on screen as text, plus meeting transcription across a good range of languages.

That distinction is doing a lot of work. The screenshot approach, which is what Microsoft Recall and the earlier Rewind generation used, means a rolling archive of pictures of everything you have ever looked at, sitting somewhere, expensive to store and catastrophic to leak. Text only capture is lighter, immediately searchable, and considerably less alarming when you think through the worst case. It is a genuinely better architectural decision and it is the reason this tool is worth a serious look rather than a flat no.

There is a real free tier, which is unusual in this category and not a trick. It is limited on chat volume, but it works and you can form an actual opinion on it.

Above that, Plus sits somewhere around seventeen to twenty dollars a month depending on whether you pay annually, Power runs about forty two, and Pro starts around a hundred. Paid plans come with a fourteen day trial. Check the current numbers when you sign up, because this is a young company and this category has been repricing every few months.

The free tier will push most active users to Plus within about a week. That is by design and it is not unreasonable, but budget for it rather than being annoyed by it.

Three things landed properly.

The end of day recall is the actual product. The pitch is context aware assistance, but what I used constantly was the ability to ask what happened today and get a real answer. Not a calendar summary. An account of the four decisions that got made across three applications, including the one in a message thread I had already forgotten. If you finish most days unable to account for where they went despite having clearly worked the entire time, this alone is the case for the tool.

Meeting notes without a bot in the room. It transcribes locally from what is happening on your machine rather than joining the call as a participant. No visible attendee, no awkward moment where somebody asks what the recording thing is. Worth noting that this convenience is exactly the consent question that has been getting notetaker vendors into trouble this year, and the fact that nobody in the meeting can see it is a feature for you and a problem for them. If you need something everyone in the room can see and agree to, Fathom remains the more honest fit and I would not talk you out of it.

Writing in your voice, without you doing the work. Because it has read a great deal of what you have actually written, output comes back closer to your register than anything you get from a cold session. This is the same result you get from building a standing brief by hand, arrived at from the opposite direction: instead of describing yourself, you let it observe you. Both work. This one requires no discipline, which for a lot of people is the deciding factor.

Now the part that matters, and I am not going to soften it.

This tool sees your banking portal. It sees the client contract with the number in it. It sees the message where somebody told you something in confidence. It sees the admin panel of every system you log into, because that is text on your screen and reading text on your screen is the entire product.

The company's posture on this is better than most. Encryption at rest, no image storage, local processing where it can, SOC 2 in progress rather than complete. Processing for the actual intelligence happens in the cloud, which is the honest limitation and the thing you have to make peace with. That is a real answer, not a marketing answer, and I will take a real answer with a caveat over a vague reassurance any day.

But you should still ask three questions before you install it, and you should ask them about any tool in this category, not just this one.

Am I the only party to what is on my screen? If you handle other people's medical, legal, or financial information, the answer is no, and their consent was not yours to give. That is not a preference question, that is a regulatory one, and it ends the conversation for a lot of professionals.

What is my exclusion list? Before you turn it on, decide what it never sees. Banking, password manager, HR system, anything under a client confidentiality clause. Configure the exclusions first, then use the tool. Doing this in the other order means the capture already happened.

What happens on the day I leave? Find the delete path before you need it. How you export, how you purge, how long deletion takes, what survives. If you cannot find a clear answer in ten minutes of looking, that itself is your answer.

If you decide to go ahead, this is the setup that makes it defensible rather than reckless.

Start on the free tier for a full week before you pay anything. Configure exclusions on day zero, not day four. Keep it off any machine that touches customer records you do not own. Put a monthly calendar reminder to open the memory and look at what it has actually collected, because the failure mode with tools like this is never a dramatic breach, it is quiet scope creep you stopped noticing in week three.

And add it to the grant list from Monday. It belongs in the same spreadsheet as everything else, with a scope column that reads broadly and honestly, so that future you making the quarterly pass sees it and reconsiders on purpose instead of by accident.

One practical note on the first hour, because it is the part where people bounce. Littlebird is close to useless on day one and it is supposed to be. It has no memory yet, so everything you ask returns the same generic answer you would have got from any assistant, and the natural conclusion is that the product does not work. Give it three full working days before you form an opinion. The whole value proposition is accumulated context, and on Monday morning there is none.

The moment it clicks, for almost everyone I have spoken to about it, is the first time you ask a question you would normally have had to go and look up. What did we agree on the pricing for that account. Where did I see that number. Who was it that raised the concern about the timeline. Getting a correct answer to one of those, sourced from a conversation you half remember, is worth more than any feature list, and it does not happen until the tool has watched you for a while.

Anyone not on a Mac. It is macOS only. There is no workaround and pretending otherwise wastes your afternoon.

Teams looking for shared knowledge. This is a personal assistant. It builds your memory, not the team's. If the problem you are solving is that nobody knows what anybody else decided, this is the wrong shape of tool entirely and you want a documentation system.

Anyone in a regulated seat. Covered above. If your screen routinely contains data belonging to somebody who has not consented to this, the interesting product conversation is irrelevant.

I like this tool more than I expected to and I trust the team's judgment more than I expected to. The text only decision is not a marketing line, it is a real engineering choice with real privacy consequences, made by people who clearly thought about the failure mode before they thought about the demo.

But the honest summary is that the value and the risk come from the same property. It is useful because it sees everything. It is a concern because it sees everything. There is no version of this product where you get the first without the second, and any review telling you otherwise is selling.

So the decision is not really about the software. It is about whether the contents of your screen are yours to share. For a solo operator running their own business on their own machine, I think the trade is good and I would take it. For anyone holding somebody else's confidential information, I think it is a clear no, and no configuration setting fixes that.

Free tier, one week, exclusions configured first. That is a cheap way to find out which one you are.

Jordan

The AI Newsroom is written for people who run something. Forward it to the one person you know who needs it.