Lingo.dev

Lingo.dev is a developer-focused localization platform built around what the company calls a "localization engine": a stateful translation API that a team creates once and configures with a glossary, brand voice rules, locale-specific conventions, a ranked chain of model providers with automatic fallback, and AI reviewers that score output per quality dimension. Engines are called from backend code, a CLI, CI/CD pipelines, or an MCP server, and the persisted configuration is attached to every request so terminology remains consistent across locales and releases.

The platform's asynchronous pipeline runs six configurable steps with per-locale failure isolation: LLM source refinement (resolving ambiguous references, normalizing concatenated strings, flagging placeholder issues), context enrichment via cosine similarity over vector embeddings that matches source terms to glossary entries and injects brand voice and locale rules, LLM translation, optional human post-editing by a network of qualified translators managed behind the API, an independent LLM post-review producing boolean tag-preservation checks and percentage fluency scores, and delivery via HMAC-signed webhooks or WebSocket streaming. The company describes an approach it terms retrieval augmented localization (RAL), in which only changed content is re-localized while matching glossary context is retrieved per request; a controlled evaluation across multiple LLM providers and languages on terminology-dense content, scored with GEMBA-MQM against official human reference translations and tested with a paired Wilcoxon signed-rank test with Holm-Bonferroni correction, reported 17-45% fewer terminology errors over 42,000+ paired quality judgments.

Open-source tooling published under the Apache-2.0 license on GitHub includes a CLI that localizes JSON, YAML, markdown, CSV and PO files with a lockfile that tracks already-translated content, a GitHub Action plus GitLab CI/CD and Bitbucket Pipelines support for continuous localization, an MCP server that gives AI coding assistants framework-specific i18n knowledge for Next.js, React Router and TanStack Start, and an early-alpha build-time compiler for React (Next.js App Router and Vite + React) that generates localized variants without translation keys or t() functions. Users can default to a Lingo.dev engine or bring their own LLM from providers including OpenAI, Anthropic, Google, Mistral, OpenRouter and Ollama.

Founding story

Co-founders Veronica Prilutskaya and Max Prilutskiy met six years before the company's YC launch while working together. Prilutskaya has an AI, machine learning and data science background; Prilutskiy has a software engineering background and previously led a frontend team at a European unicorn. Both encountered localization problems firsthand at startups and enterprises and while bootstrapping a SaaS with customers in over 20 countries; they had earlier sold a SaaS startup, Notionlytics, to an undisclosed buyer. Their stated mission is for software to become multilingual from day one with minimal effort.

Business model

Lingo.dev sells localization infrastructure to software teams: customers create and configure localization engines on the platform and consume them through APIs, a CLI, CI/CD integrations and an MCP server. Open-source client tooling on GitHub connects to the commercial platform, and an optional human post-editing layer using the company's network of translators is offered behind the same API. Prospective users can create an engine self-serve or book a demo.

Paid use of the hosted localization engineering platform by software teams, with open-source tooling distributed free under Apache-2.0 as the entry point; an optional human post-editing service is offered as a configurable pipeline step. Specific pricing is not stated in the sources, and the CEO declined to share financial metrics.

Traction

The platform reports over 368 million words processed and states its infrastructure sits behind 200M+ translated words for customers including Mistral, Solana, SoSafe and Cal.com; at the time of its YC launch the figure was 3M+ words. The open-source repository has 5.4k stars and 815 forks across 1,922 commits, with documentation localized into more than 25 languages. Team size is listed as 7 by Y Combinator, while February 2025 coverage cited three employees.

Latest developments

The website shows a weekly shipping cadence, with a recent release enabling translation of Zendesk Help Centers, and reports over 368 million words processed on the platform. The company has published research on retrieval augmented localization and offers an early-alpha build-time compiler for React. Earlier stated roadmap items included using screenshots of application UIs to extract additional context, broader integrations with the modern development stack, more sophisticated brand voice controls, and applied AI research on gender-specific language.