Local Indexing

Local Indexing is NELA's retrieval layer. It lets the app ground responses in your own files without sending your source content to an external inference service.

How it works

• You add files or folders from the workspace UI.
• NELA parses files into chunks and stores them in a local workspace index.
• Embedding models convert chunks and queries into vectors for similarity search.
• Classifier and grader models can improve retrieval quality and ranking.
• Retrieved chunks are passed into generation prompts for grounded responses.

Supported ingestion spans common docs, text, code, and selected audio formats used in transcription-assisted flows.

Architecture snapshot