use cases
RAG over your local files — no vector service, no upload
Point dig at a directory and your agent can retrieve across it with hybrid search — PDFs, notes, and docs included, indexed locally, nothing uploaded to a cloud vector store.
Retrieval-augmented generation usually means shipping your files to a hosted vector database. dig does RAG over the directory as it sits on disk — locally.
How it works
- Index a directory.
dig init ~/library && dig scanbuilds a content-addressed store plus a search index over paths, labels, and content — PDFs and text included. - Retrieve.
dig find "<query>" --mode hybrid --jsonreturns ranked results combining full-text and semantic search. The--jsonsurface is built for agents. - Ground the answer. Your agent feeds those passages to the model. dig is the retriever; the agent and model do the generating.
dig init ~/library
dig scan
dig find "invoice acme 2024" --mode hybrid --json
Why dig for RAG
- No external vector DB. The index is a derived view over the store —
dig scanrebuilds it. Nothing to provision or keep in sync. - Nothing uploaded. With no model endpoint configured, dig makes zero network calls. Your documents stay on your machine.
- Whole corpus. Content-hash dedupe and real content indexing mean every file counts, not just Markdown.
- Any agent. Reach the same index from Claude Code, Codex, or the AI SDK over MCP.
Start
Install dig, point it at a folder, and your agent has a local retriever.