Featured

langshift

Translates an entire repository between 13 programming languages, then runs the result and fixes what breaks.

Core stack
  • Python
  • FastAPI
  • React
  • Playwright

Screens

Screen 1 of 4
01Translation form: source repository, source and target language, and provider selection
Chapter 01

Overview

Translating one file with an LLM is a demo. Translating a repository is an engineering problem: dependency manifests have to map across ecosystems, files too large for a context window have to be split without cutting a function in half, rate limits have to be survived rather than crashed on, and a run that dies at file 200 should not start over. langshift handles each of those explicitly. Its most useful design decision is the offline provider — a rule-based transformer with no LLM at all — which lets the entire test suite and CI run deterministically with no API key and no network.

Chapter 02

Highlights

  • 13 languages, with dependency manifests translated across ecosystems (package.json to requirements.txt and back)
  • Oversized files split at blank-line boundaries so no chunk cuts a function body in half
  • Failures feed the error back as context and retry, bounded per provider
  • Checkpoint and resume: a run that stops at file 200 restarts from file 200
  • A rule-based offline provider keeps CI deterministic with no API key, behind a 92% coverage gate