Yesterday's Top Launches: 2 Tools from November 10, 2026
The November 10 digest covers Sage2, a pair of APIs for extracting typed fields from text for developer workflows, and a $9.99 Mac utility that stores screenshots in the notch.
Two launches landed on November 10, and they share almost nothing except timing. One is a pair of APIs aimed at teams wiring language models into software that has to actually do something. The other is a $9.99 Mac utility that parks your screenshots in the notch. Neither has accumulated community rankings yet, which is expected this soon after release, so this digest leans on what the launch posts and product pages actually say.
Sage2 by Levanto Labs
Among the new developer tools shipping this week, Sage2 is the one worth a closer read if you have ever watched a model return a beautifully written paragraph when what your code needed was a number.
The pitch is narrow on purpose. Two API operations, separate but meant to be composed: extraction turns text into typed fields, and a streaming endpoint keeps a yes/no business decision current while a conversation unfolds. Your own code does the orchestration and the tool calls. Sage2 supplies the structure and the verdict.
Extraction works by declaration. You define fields with names, types, and definitions, then POST raw text to sage.levanto.ai/v1/extract with an API key. In the published example, a customer says “I want a midsize SUV for up to $450 a month.” The request asks for a vehicle (text), a monthly_limit (money), and a term (duration). Back comes the vehicle as “midsize SUV,” the limit as $450 USD, and the term as null, which is the useful part: null is a signal that your workflow should ask for a duration. When the customer later says “over 60 months,” the value arrives as P60M, an ISO 8601 duration you can feed straight into date math.
Every value comes back with the source text it was pulled from and a probability. That traceability matters more than it sounds. It means a support agent or a QA pass can see exactly which sentence produced which field, and a workflow can skip the guessing step entirely. Types cover text, numbers, money, dates, times, and durations, with either single-value or all-mentions extraction.
The decision half connects over WebSocket to wss://sage.levanto.ai/v1/stream. You declare questions up front (up to 32 per session, with 16,384 tokens of context), then append new text as it arrives. In the documented financing scenario, an offer alone is not acceptance, an objection holds the answer at “no,” and accepting a revised offer flips it to “yes.” Because the question is declared once and carried in the session, the decision stays a stable object rather than a fresh reading of the whole transcript each turn.
Reasonable caveats: this is infrastructure, not a framework. Everything downstream, from validation to quote generation, stays your job, and the quality of results tracks how carefully you write the schema. Reasoning, web grounding, and vision are all in beta. Pricing tiers are not published in the post either. You can start free through the Levanto platform sign-in, but you cannot budget against it yet. Best fit is teams already comfortable with prompt fragility who want the schema to live in configuration instead of inside a 400-line system message.
Archie
The second launch is smaller in scope and much easier to explain over coffee. Archie is a native macOS app that puts a shelf in your notch with three tabs: Screenshots, Image to Text, and File Tray. It requires macOS 13 Ventura or later, works with or without a physical notch via a display mode, and costs $9.99 once, with a three-day trial that skips the credit card.
The core idea is that screenshots should not pile up on your desktop, a default folder, or a menu bar dropdown you keep forgetting. New captures route to ~/Desktop/Screenshots instead, and hovering the notch gives you a small grouped list, sorted by Today, Yesterday, and Previous 7 days. The right-click menu handles Pin (⌘P), Copy (⌘C), Image to Text (⌘T), Show in Finder (⌘F), and Trash (⌘⌫). Pinned items survive auto-clean, which sweeps older captures into the Trash on a schedule you pick, from Never to 7 days.
The other two tabs earn their spot depending on how you work. Image to Text drops a screenshot and returns copyable text, whether you drag it in or hit ⌘T from the shelf. File Tray holds non-image working files, pasted with ⌘V, so a Figma file and a zip archive can sit next to a receipt instead of in Downloads. The editor is modest but covered: blur for private details, arrows and shapes for highlighting, crop, and numbered-step notes, all with copy-and-close.
Two honest notes. The notch is contested space, and a shelf there only survives if flicking files in and out feels faster than the alternative after a week of use; the trial exists for exactly that test. And auto-clean includes existing screenshots, so check the setting before you leave it running unattended. The OCR is single-image extraction with no batch or PDF workflow mentioned, which keeps expectations in line.
It suits anyone screenshot-first: designers referencing interfaces, developers capturing stack traces, support folks sharing blurred dashboards with arrows drawn on them.