How VS Code Detects Programming Languages (github.com)

Microsoft published an open-source npm package, @vscode/vscode-languagedetection, that runs a small on-disk model to auto-detect programming languages from a code snippet. Installable via npm/yarn, it exposes ModelOperations.runModel(...) which returns a ranked list of languageId/confidence pairs (the sample call on a TypeScript snippet returned ts at ~0.48 confidence, then rs, js, c, etc.). By default the package runs in Node.js and loads model.json and group1-shard1of1.bin from the repo, enabling low-latency, local inference for editor features or automated pipelines. For non-standard environments you can override loading with modelJSONFunc and weightsFunc (both async), and tune detection with the options bag: minContentSize (default 20) and maxContentSize (default 100000). The repo includes standard build/test commands (npm install, npm run watch, npm test, npm run build, npm pack) and contributor guidance (CLA checks). This is significant for the AI/ML and developer-tooling communities because it provides a lightweight, reproducible on-device classifier useful for routing language-specific models, preprocessing corpora, labeling code datasets, or powering editor UX without cloud calls — while still being extensible for embedding in non-Node runtimes.