eloqnt/studiov0.5
Navigation›

Changelog

@eloqnt/sdk

Historic reference of all changes to @eloqnt/sdk.

0.5.1

  • Improve prompt caching.

0.5.0

  • eloqnt lint now distinguishes errors from warnings. Missing translations are warnings and no longer fail the run, so lint can gate CI without requiring translations on every commit. Pass --strict to treat warnings as errors.

0.4.0

  • Added eloqnt review source: Reviews source strings for typos and grammatical errors and fixes them in place.

0.3.1

  • Support path templates for messages.path to support non-next-intl consumers (e.g. i18next)

0.3.0

  • Drop internal exports from @eloqnt/sdk's public entry. The main entry now surfaces only what's intended for external use (lint, defineConfig, defineCodec, translate, and their return types); everything else moved to @eloqnt/sdk/_internal for workspace-only consumption.
  • Drop the engine API's backward-compat fields for pre-split CLIs. The older clients weren't actually picking them up in practice, so the fallback wasn't doing anything anyway.
  • Merge @eloqnt/inference into @eloqnt/sdk.
  • Add support for messages extracted by experimental useExtracted in next-intl@<4.13.
  • When config discovery fails, check for common misspellings of .eloqnt (.eloquent, .eloqunt, .eloqent) and point the user at the rename.
  • Load .eloqnt/config.{ts,mts,js,mjs} and custom codec files through jiti instead of Node's native loader. The host project no longer needs "type": "module" to use TypeScript, and TypeScript files now work on older Node versions.

0.2.0

  • Upgrade to AI SDK 7.

0.1.0

  • Split plans into "Hosted model" and "Bring your own model"

0.0.8

  • Run static post-translation checks (ICU placeholder parity) on every received translation, with up to two automatic re-translations on failure.

0.0.7

  • Consume AST context for .json formats, output file reference for undefined-key rule

0.0.6

  • Speed up linting on large projects by scanning source files for message usage concurrently.

0.0.5

  • Drop a redundant prompt fetch in engine-mode translation. A run with nothing to translate now makes no request, so it no longer fails up front when signed out or unsubscribed.

0.0.4

  • Remove toJSONString from the custom codec interface. Catalogs are loaded via decode, so codecs no longer define it.
  • Handle messages split across several messages.path folders correctly, including a new duplicate-id lint rule for ids repeated across folders.

0.0.3

  • The inconsistent-args lint rule now catches rich tags (<b>…</b>) a translation drops or invents, not just {argument} placeholders. Placeholders are compared by their literal token, so a <name> tag never stands in for a {name} argument, and the reporter underlines the stray tag in the value.

0.0.2

  • Make CLI errors friendlier.

0.0.1

  • Exercise the automated release pipeline. No functional changes since the initial 0.0.0 publish.
@eloqnt/engine