KG CSV Toolkit User

Not every knowledge base needs a graph database behind it. Sometimes what you actually have is a spreadsheet - a flat table of products, FAQs, reference facts - and a full Neo4j setup would be overkill. KG_csv_toolkit_user is the lightweight version of the same idea this pack's Neo4j toolkit covers: it gives an LLM agent the ability to query a local CSV file as its knowledge source, no external database required.

Like every "tool"-output node in this pack, it doesn't do anything when the graph runs - it registers a capability. Wire its tools output into the agent/LLM node, and the model decides during the conversation whether to actually look something up in the CSV.

The inputs and outputs that matter

• relative_path - a dropdown populated from files already sitting in the pack's own knowledge-base folder (test.csv is the built-in sample). To point this at your own data, drop your CSV into that folder first and it becomes selectable.
• absolute_path - an alternative to relative_path: point this at any CSV file elsewhere on disk instead of using the pack's own folder.
• is_enable - standard bypass toggle.

Output: tools (STRING) - the serialized tool definition, meant for the agent node's tool-list input, not for reading directly.

"User" vs. "Developer": this pack ships both a KG_csv_toolkit_user and a KG_csv_toolkit_developer. Going by the naming pattern (and matching how this pack's other user/developer tool pairs are structured), the reasonable read is that "user" grants read/query access to the CSV - safe to hand to an agent talking to end users - while "developer" grants broader authoring access, letting the LLM modify the data itself. That split isn't spelled out anywhere in this node's own schema, so if the exact boundary matters for something you're building, verify what each tool's description actually says once it's loaded into your agent, rather than assuming from the name alone.

How to install it

Via ComfyUI Manager: search comfyui_LLM_party, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/heshengtao/comfyui_LLM_party.git

Then pip install -r requirements.txt from inside the pack folder using ComfyUI's own Python, and restart. No external database to stand up - that's the whole appeal versus the Neo4j toolkit - just a CSV file sitting where this node can find it.

Common issues & troubleshooting

The dropdown doesn't show your file. relative_path only lists files already inside the pack's knowledge-base folder - if your CSV isn't showing up, either move it there first, or skip the dropdown entirely and use absolute_path to point straight at it.

The agent never queries the CSV. Standard tool-calling behavior: the LLM only reaches for a tool if it has reason to think it's relevant. Make sure your agent's instructions actually mention it has access to this knowledge base.

Queries come back empty even though the file loads fine. That points at the CSV's actual contents, not the node - if the data relevant to the question isn't in the file, there's nothing for the tool to retrieve. This node is the connection layer to your CSV, not a source of knowledge on its own.