File-Augmented Retrieval: Making Every File Readable to Coding Agents via Persistent .meta Sidecars

AI coding agents—Claude Code, Codex, GitHub Copilot—cannot read binary files.
In a typical enterprise repository, 30–40% of files are non-text, yet they contain critical context.

We propose File-Augmented Retrieval (FAR): a file-time augmentation paradigm that adds a persistent .meta sidecar to every file, containing extracted content as Markdown with minimal YAML provenance.

Unlike RAG, FAR requires no runtime infrastructure—agents simply read .meta files directly from the filesystem.

On a 10,000-file heterogeneous corpus, FAR achieves 82.6% file-discovery accuracy compared to 58.7% for RAG, with zero infrastructure and 6.3% storage overhead.

RAG performs retrieval at query time. FAR performs augmentation at file time.