Knowledge Vault

The knowledge vault stores documents that RAG (retrieval-augmented generation) pipelines can search through. All uploaded documents are embedded into a vector store (LanceDB) and retrieved automatically when a pipeline uses a memory-read node.

Supported file types

• PDF
• DOCX
• TXT
• MD
• CSV

Uploading documents

1. Go to Memory > Knowledge Vault.
2. Click Upload File.
3. Select a file from your computer.
4. GridWork embeds it into the vector store.

The document is searchable within 30 seconds.

Reindexing

If you update a document or add new files, click Reindex to re-embed everything. GridWork re-processes all vault documents and updates the vector store.

Reindexing takes a few minutes depending on file size.

Using the vault in pipelines

In the React Flow editor, add a Memory Read node and select Knowledge Vault. The node will retrieve relevant documents based on a query and pass them to downstream LLM nodes.

Deleting documents

1. Go to Memory > Knowledge Vault.
2. Click the delete icon next to a file.
3. Click Reindex to update the vector store.