vandelay knowledge¶

Manage the knowledge/RAG corpus.

Usage¶

vandelay knowledge <subcommand>

Subcommands¶

add¶

Add a file or directory to the knowledge base:

vandelay knowledge add ~/docs/report.pdf
vandelay knowledge add ~/my-notes/

Supported formats: .pdf, .txt, .md, .csv, .json, .docx, .doc

status¶

Check the current knowledge corpus status:

vandelay knowledge status

Shows: enabled state, embedder provider, document count, vector store path.

list¶

Show the current vector count and storage path:

vandelay knowledge list

refresh¶

Re-index the built-in corpus:

vandelay knowledge refresh
vandelay knowledge refresh --force  # Force full rebuild

Without --force, only re-indexes if source versions have changed.

clear¶

Clear all knowledge vectors:

vandelay knowledge clear --yes

How It Works¶

1. Documents are read from disk (file or directory)
2. Documents are chunked into passages
3. Each chunk is embedded via the configured embedder
4. Vectors are stored in ChromaDB at ~/.vandelay/data/knowledge_vectors/

The agent queries this vector store when knowledge.enabled is true and search_knowledge is active on the agent.

Embedder Resolution¶

See Knowledge concept for details on how the embedder is auto-resolved from your model provider.