@lingolint/ai

Optional AI linguistic review for LingoLint: semantic mistranslations, grammar, tone consistency, capitalization and unnatural wording, each with an explanation and a confidence score.

Everything here is additive. @lingolint/core and the lingolint CLI work fully without it.

Using AI review sends translation text to the configured provider. Bring your own API key; it is read from the environment and never logged or stored.

npm install @lingolint/ai @lingolint/core

import { analyzeTranslations } from '@lingolint/core';
import { createAnthropicProvider, withBatching, withCache, createMemoryCache } from '@lingolint/ai';

const ai = withCache(
  withBatching(createAnthropicProvider({ apiKey: process.env.ANTHROPIC_API_KEY }), {
    batchSize: 25,
  }),
  createMemoryCache(),
);

const report = await analyzeTranslations({
  sourceLocale: 'en',
  targetLocale: 'es',
  source,
  target,
  config: { ai: { enabled: true }, context: 'A note-taking app; informal register.' },
  ai,
});

Exports

| Export | Purpose | | ---------------------------------------------------------------------------------- | ------------------------------------------------------------------------ | | createAnthropicProvider | Adapter using the official Anthropic SDK with structured outputs. | | withBatching(provider, opts) | Split reviews into fixed-size concurrent requests. | | withCache(provider, cache) | Per-string result cache; unchanged strings are never re-sent. | | createMemoryCache, createFileCache | Cache implementations. | | createReviewerFromConfig | What the CLI uses: cache → batching → provider from ai config + env. | | registerAIProvider | Plug in another vendor under a name usable as ai.provider. | | buildSystemPrompt, buildUserMessage, reviewResponseSchema, parseReviewJson | Reuse the prompt and response contract in your own adapter. |

Adding a provider is documented in CONTRIBUTING.md.

MIT licensed.