Can a prompt replace Lokalise?

Build a translation editor where every key carries a screenshot.

Keys, languages and translations in a database, imported from and exported to your project's JSON or .strings files.

The context system is the point. A screenshot is uploaded with a manifest listing which keys appear in it and at which pixel rectangles. In the editor, translating a key shows the screenshot with that rectangle highlighted and everything else dimmed. A translator who can see the string in place makes different and better decisions than one reading a list, and this single feature explains most of Lokalise's value.

Generate the manifest rather than typing it: add a debug build of your app that renders each localised string wrapped in a marker, screenshot the screens with Playwright in CI, and extract the marker positions from the DOM. Manual screenshot mapping decays within one release, so if it is not automated it is not worth building.

Store the rendered pixel width of each string region and warn when a translation is more than 30 percent longer than the source. German and Finnish routinely overflow layouts built against English, and catching it in the editor is far cheaper than in a bug report.

Add the checks that catch real breakage: placeholder mismatches between source and translation, unbalanced tags in strings containing markup, and missing keys per language before a release.

Export back to the project files with a byte-identical round-trip test, and make missing keys per language fail the build rather than fall back silently to English.

Out of scope: over-the-air mobile updates, branching, approval workflows and permissions.