rag-tool

name	rag-tool
description	Store and search text or markdown as semantic chunks in a reusable local vector store. Use when a user wants to ingest documents or retrieve relevant passages from previously stored material.

RAG Tool

Use one command for the requested operation:

cvstore --doc=DOC --kind=source|derived [--source=RELATIVEPATH] INPUT.txt
cvquery [--doc=DOC] [--kind=source|derived] [--page=N] [--label=TEXT] QUERY

Readiness

• Run cvstore --help and cvquery --help.
• If either command is missing or fails to start, read references/chunkvec-install.md, install the binary or missing runtime dependency, and retry.
• After installing the tools, tell the user how to configure DEEPINFRA_API_KEY or config.json before ingest or search.

Execution

• Request unrestricted network or escalated execution directly when running cvstore or cvquery. Do not run sandboxed network probes first.
• Do not inspect environment variables, shell profiles, or filesystem files to discover API keys. If authentication fails, ask the user to configure the key, then retry.

Mode Selection

• For ingest requests, read references/store-mode.md.
• For retrieval requests, read references/search-mode.md.
• Read both only when the request spans both operations.

Exit Codes

• 0: the requested work succeeded.
• 2: cvstore finished with permanent chunk failures; successful chunks were still stored. Report the partial failure.
• 3: startup or runtime failure. Report the error.

Stable Doc IDs

• Derive lowercase kebab-case doc ids from one stable document identity. Prefer the source file stem, then an explicit logical title.
• Keep the same doc for the original document and its derived notes, quiz, or flashcards. Express artifact type with kind, not a new doc.
• If pasted text has no stable file or title, ask for a short logical name before storing.