use cases

Retrieve and recall across your research notes

dig indexes a folder of research notes for hybrid search and hands your agent a token-budgeted, provenance-tagged recall pack — so it answers from your sources, locally.

Research notes pile up faster than you can re-read them. dig makes the whole pile retrievable, and gives an agent exactly the slice it needs.

How it works

  1. Index the notes. dig init ~/research && dig scan builds a content-addressed store plus a search index over paths, labels, and note content.
  2. Find across them. dig find "<query>" --mode hybrid --json ranks results by fusing full-text and semantic search — the --json surface is built for an agent to consume.
  3. Recall for the agent. dig recall "<query>" --budget 2000 returns the query-relevant window of each hit, capped to a token budget and tagged with provenance — no window overflow, every claim traceable to a note.
dig init ~/research
dig scan
dig find "transformer attention ablations" --mode hybrid --json
dig recall "what did we conclude about scaling laws" --budget 2000

Why dig

  • Grounded, not guessed. Recall serves passages from your notes; the agent and model do the answering. dig is the retriever.
  • Local. With no model endpoint configured, retrieval makes zero network calls — your research stays on your machine.
  • Any agent. Reach the same index from Claude Code, Cursor, or the AI SDK over MCP.

RAG over local files · Agent memory