By name: vdb-query
description: Query the local raw-PDF vector database at ~/.wiki-chroma. Use when the user asks about indexed PDFs, the contents of a specific PDF, the vector DB / VDB / RAG / chunks, "what does .pdf say", or names an indexed PDF by filename or stem.
Raw-PDF Vector DB Query
Queries the Chroma collection pdf_chunks at ~/.wiki-chroma, fed by vdb-watcher from PDFs that have been ingested via the pdf-ingest skill.
When to use
- The user asks about content of an indexed PDF (e.g. "what does the handbook say about X").
- The user mentions the VDB, vector DB, RAG, or chunks.
- The user names a PDF by filename or stem (e.g.
fe-handbook-10-6).
CLI
$HOME/.local/bin/vdb-query status
$HOME/.local/bin/vdb-query sources
$HOME/.local/bin/vdb-query query "<question>" -k 8
When restricting to one source:
$HOME/.local/bin/vdb-query query "<question>" -k 8 --source "<pdf-stem>"
Answering discipline
- Query the VDB BEFORE answering, even if you think you know.
- If the first query returns weak hits, try up to 3 alternate keyword queries.
- Answer ONLY from retrieved snippets and any markdown files you explicitly read from the result's
source_path. - Cite claims as
[<source> chunk N/M]. - If retrieval is thin, say so: "the VDB does not contain enough evidence."
The query result includes a source_path field — that markdown file is the Docling extraction. Read it for surrounding context when a snippet is cut mid-thought.
Python env
Pinned to ~/.wiki-venv (Chroma + Docling + onnxruntime). Never use system python3 for this stack — the CLI shims hardcode the venv interpreter for a reason.
Companion
To add a PDF to the database, use the pdf-ingest skill (or vdb-watcher ingest <pdf> directly).