Griptape RAG Retrieve: Vector Store Module

This is the quieter sibling of the Griptape RAG Retrieve: Text Loader Module. Where that node loads a document and retrieves from it in one go, this one assumes the vector store is already populated - you just tell it which store, which namespace, and how many hits you want back.

That split matters once your knowledge base gets too big to re-ingest every run. You index your docs once (via a text loader module or the pack's upsert tasks), then every later workflow just queries. No re-chunking, no re-embedding, no wasted API spend. If you're building something you'll run more than once, this is the node you'll reach for most of the time.

How it works

It constructs a VectorStoreRetrievalRagModule from the Griptape framework - the retrieval stage of the RAG engine with no loader attached. On a query it hits the connected vector store driver, pulls the nearest matches under the given namespace, and returns them as a MODULE output (MODULE_LIST type) for the Griptape Rag Engine node's retrieval_stage_modules input.

A word of caution: if you leave vector_store_driver disconnected, the pack silently drops in a local vector store backed by a dummy embedding driver. It "works" in the sense that nothing crashes, but you're getting crude matches instead of semantic ones. Connect the real driver from your driver config.

The inputs that matter

Honestly, this node is almost insultingly simple:

• vector_store_driver - the store you populated earlier. The one input you should basically never leave empty.
• namespace - which section of the store to search. Default "default"; keep it consistent with whatever namespace you wrote your chunks into.
• count - how many results to return. Default 5. -1 skips the count entirely, letting the driver fall back to its own default.

That's it. One MODULE output.

Installing

Same pack as the rest of Griptape's nodes. ComfyUI Manager → search "Griptape" → Install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/griptape-ai/ComfyUI-Griptape

Then restart ComfyUI. The heavy dependency is griptape[all], which drags in torch - fine on a machine that already runs ComfyUI, but worth knowing that the first install is not quick.

Common gotchas

The two that bite people across every node in this pack: the torch version conflict on Nvidia (reinstall with --extra-index-url https://download.pytorch.org/whl/cu121 if import errors appear) and the stale-griptape ImportError that shows up as cannot import name 'OllamaPromptDriver' - fixed with python -m pip install griptape -U.

The RAG-specific trap is the namespace. If your ingest step wrote into a custom namespace and this node is still on "default", you'll get empty results and blame the model. Check the namespace first; it's almost always the answer.