Product Updates
By Emirhan Karahasan
Loqalit Is Now Agentic Localization QA. Here’s What Changed.
Loqalit now runs as a crew of localization QA agents — Scout, Editor, and Verifier — that scan your live pages on a schedule, propose fixes with proof, and route them into Crowdin or Phrase.
Loqalit no longer waits for you to press Scan. Set your pages, languages, and a schedule once, and a crew of localization QA agents takes it from there.
This is the biggest change we have shipped: Loqalit is now an agentic localization QA platform, not a tool you have to remember to run.
What actually changed
The old workflow asked you to remember: install the extension, sign in, open the page, run the scan.
The new workflow asks you to decide once: which pages, which languages, and how often.
Everything after that runs unattended. The agents open your real pages in a real browser, on the schedule you chose: daily, weekly, quarterly, or a single date in the future, with nobody logged in and nobody watching a progress bar.
Here is what is new, concretely:
Scheduled scans instead of manual ones, set once and left to run
A crew of agents, Scout, Editor, and Verifier, instead of a single black-box AI
Every finding now carries a screenshot, a reason, and an MQM severity, not just a flag
Accepted fixes go straight into Crowdin, Phrase, or your favorite TMS, exported in whatever format you need
Loqalit now fingerprints every page it sees, so it knows what your last release actually broke
None of this required you to change how you translate. It changed when and how the checking happens.
Meet the crew
Three agents do the work, and each one has a narrow job: Scout, Editor, and Verifier.
Scout crawls from a page you give it, reads the sitemap, and proposes which pages are worth watching. Editor reads the exact string a scan flags and drafts a correction that respects your glossary, brand voice, and style guide. Verifier resizes the browser, measures the element, photographs it, and checks the glossary again before anything is marked as a real finding.
None of them can ship a change on their own. Every fix the crew proposes lands as a proposal card you accept or reject. Nothing moves until a person presses Confirm.
Every finding now ships with proof
A finding used to be a flagged string. Now it is a screenshot of the exact element where the text sits, a plain-language reason it is wrong, an MQM severity, and a suggested fix, all in the same card.
You do not have to reconstruct context from a spreadsheet row. You look at the element the way your users will, decide if the crew got it right, and move on.
The screenshots are captured at 2x, so a truncated word or an overflowing button is visible even on a small crop.
What happens when you accept a fix
Accepting a finding is not the end of the workflow, it is the start of delivery. The correction is routed server-side into Crowdin, Phrase, or your favorite TMS, the same place your translators already work, and exported in whatever format you need, so nobody has to copy text between systems.
Accept, and it ships. That is the whole loop: a finding arrives with proof, you accept it, and the fix lands in your TMS.
Loqalit also watches for what changes after you ship. Every page it scans gets fingerprinted, so the next time that page is republished, Loqalit can tell what actually moved, not just that the page loaded.
A release breaks something, you get an email. When a fingerprint changes in a way that introduces a regression, an alert goes out with per-person email preferences, so the right person hears about it without watching a dashboard.
The result is a QA layer that runs on your release cadence instead of your calendar reminders. You set it once, and it keeps working after you have stopped thinking about it.
See the whole platform
This update touches more than the scanning loop. Glossary import, a brand voice and style guide per project, a configurable pass threshold, signed webhooks, a REST API, and a Chrome extension for scanning a page on the spot are all part of the same release.
We grouped everything into one page so you can see how the pieces fit together: Detection, automation, the agents, your standards, delivery, reporting, and team.
If you have not looked at Loqalit in a while, this is the version worth a second look.
