Vertana¹ is an LLM-powered agentic translation library for TypeScript/JavaScript. It goes beyond simple LLM prompting by using autonomous agent workflows to gather rich contextual information, ensuring high-quality translations that preserve meaning, tone, and formatting.

• Agentic context gathering: Automatically invoke external sources before translation, or let the LLM query passive sources as needed via tools
• Smart chunking: Content-aware chunkers for plain text, Markdown, and HTML that respect structural boundaries while staying within token limits
• Glossary support: Static glossaries for consistent terminology, plus dynamic glossary accumulation across chunks
• Quality evaluation: Assess translations on accuracy, fluency, terminology, and style dimensions
• Iterative refinement: Re-translate low-scoring chunks with boundary evaluation until quality thresholds are met
• Best-of-N selection: Generate translations with multiple models and select the best result via parallel per-chunk evaluation
• Progress reporting: Track translation stages via callbacks
• Multi-runtime support: Works seamlessly with Deno, Node.js, and Bun

import { translate } from "@vertana/facade";
import { openai } from "@ai-sdk/openai";

const result = await translate(
  openai("gpt-4o"),
  "ko",
  "Hello, world!  Welcome to Vertana.",
);

console.log(result.text);

Vertana provides comprehensive documentation to help you get started quickly: https://vertana.org/.

API reference documentation for each package is available on JSR (see below).

Vertana is a monorepo which contains multiple packages. If you are looking for one package to start with, check out @vertana/facade. The following is a list of the available packages: