Pool is a new screenshots app that promises to sort a user's camera roll with AI. The real question is whether the category has finally matured, and what the user has to trade to find out.
Your phone is full of screenshots you will never look at again: a boarding pass from a trip already taken, a recipe you saved and forgot, a 2FA code you used once, a meme you meant to send three months ago, a doctor's note you cannot quite find. Every so often, a new app promises to fix the mess. The latest is Pool, a screenshots app from Random Access Memories Co. that David Pierce flagged as out of beta this week in The Verge Installer, and is now being pitched, in its own words, as a way to save anything with a screenshot.
Pool is built around the idea that a camera roll is actually a filing cabinet in disguise, and that the right AI can finally make it usable. It auto-categorizes screenshots, links them into shareable collections called pools, and lets a user search across them like a personal database. The pitch is ambitious. The category is old.
The lead source for Pool's debut is The Verge Installer #132, a weekly newsletter roundup where editor David Pierce flagged the app as a highlight of the week. That framing matters. Installer is a links-and-context column from a respected consumer-tech writer, but it is tester-positive by design and short on adversarial analysis. Pool itself presents the product on its official site as a screenshot-first experience, and the company has a /research section that includes manifestos on why the camera roll is the next trillion-dollar AI surface and how WHOOP, Spotify, and Plaid supposedly made the same data bet that screenshots are about to make.
The framing is deliberately grandiose. The track record of this category is mixed. For years, a parade of apps and platform features has tried to do what Pool is now selling: Google Photos' visual search, Samsung Gallery's on-device smart suggestions, Apple's screenshot OCR and visual lookup in iOS, Notion's web clipper, Readwise Reader's saved-image inbox, and a long tail of smaller tools. Each promised to turn the screenshot pile into something a person could actually use. Most delivered partial wins. Most users still default to scrolling and squinting.
The reason that history matters is the trade-off Pool asks for. To sort your screenshots, Pool has to read them. A boarding pass, a 2FA code, a prescription label, a screenshot of a bank balance, a private message: all of it becomes input. Pool's own privacy policy makes the architecture explicit: screenshots are sent to third-party AI providers — the policy names Google, OpenAI, and Anthropic as examples — to generate embeddings, extract text, and deliver the indexing and search features. This is not an on-device operation. The company's /research essays and the Verge's prior screenshot-AI piece frame screenshots as a fundamentally great interface for using computers. That is a real product thesis. It is also the thesis that decides what Pool's cloud AI partners are allowed to see.
Pool's site carries dated /press, /investors, /careers, and /club pages, which signals a company positioning itself as more than a hobby tool. The privacy policy also notes that embeddings and derived signals are retained alongside original content until deletion, and that screenshots may be retained in backups or caches even after removal from the main interface. The out-of-beta framing in the lead source reflects one editor's characterization rather than an explicit confirmation on the company's homepage.
The competitive backdrop also cuts against the pitch. Apple has been quietly building on-device screenshot search and visual lookup into iOS for years, and Google has done the same in Photos and the system gallery on Android. Both platforms already see the local image and can index it without sending it off the device. Pool's answer to that structural problem is to run the workload in the cloud with explicit third-party AI partners — a different privacy posture than the OS-level alternative, and one that means your most sensitive image surface is being processed by external AI providers rather than staying on the device.
Pool is the most polished entry in a category that has been trying to mature for years. It is also a concrete example of where that maturity question actually lives: not in the model's accuracy, but in what the user is handing over and who is doing the reading. The privacy policy names the providers. The question is what they see, what they retain, and what happens when the next model update lands.