Languine

Translation files rot. Someone adds an English string, the other locales lag, and the gap surfaces when a user meets an untranslated label. Languine puts an AI translation service on infrastructure you own and wires it into the two places strings actually change: a developer's terminal and continuous integration.

The deploy button forks the repository into your GitHub account, creates a Vercel project rooted at the web app, provisions serverless Postgres through the Vercel Marketplace, and runs Drizzle migrations during the build. Background work runs on Vercel Workflows, which are durable and resumable. Translation calls go through the Vercel AI Gateway, authenticated in production by the project's OIDC token, so there is no separate model provider key to manage.

A handful of settings shape a deployment, and one of them is not optional.

• Deployment Protection: Vercel authentication must be switched on, or the dashboard and the token page are public.
• API key: a single random string shared by the dashboard, the CLI, and the GitHub Action.
• Model choice: an AI Gateway model slug, defaulting to GPT-4.1 and swappable for Claude or a smaller model.
• Database URL: injected automatically by the Marketplace integration you picked at deploy time.
• Project id: written into a config file by the init command, which creates the project over tRPC.

Day to day use is three commands: login against your deployment, init, then translate. For continuous integration a base URL and key as environment variables cover non-interactive runs, and the published Action can open a pull request with the results.