KG CSV Toolkit Developer

This pack ships a matched pair for CSV-backed knowledge bases: KG_csv_toolkit_user for querying, and this one, KG_csv_toolkit_developer, sitting right next to it with an identical schema. Same file, same connection setup - the difference is in what capability gets handed to the LLM once it's wired up as a tool. Going by the naming convention this pack uses consistently across its knowledge-base tool pairs, "developer" reads as the authoring version: the LLM isn't just looking things up, it can add, edit, or restructure entries in the CSV itself.

That's worth sitting with before you wire this into anything: a "user" tool that returns wrong data is annoying. A "developer" tool that's supposed to only read but can actually write is a real risk to a knowledge base you care about. If you're not certain which capability you're granting, check what the tool's own description says once it's loaded into your agent - don't assume from the node's name alone, since that's an inference from this pack's naming pattern, not something spelled out in the schema.

The inputs and outputs that matter

• relative_path - a dropdown of files already inside the pack's own knowledge-base folder (test.csv is the sample). Drop your own CSV into that folder to make it selectable here.
• absolute_path - point this at any CSV file elsewhere on disk instead.
• is_enable - standard bypass toggle.

Output: tools (STRING) - the serialized tool definition, wired into the agent node's tool-list input.

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 - just a CSV file this node can reach, same as its "user" counterpart.

Common issues & troubleshooting

Before anything else: back up whatever CSV you point this at. If this node genuinely grants write access, the first time an LLM makes an edit you didn't expect (a reasonable model doing a reasonable-sounding thing with an unclear instruction), you'll want a copy to fall back to. This is standard practice for any tool that lets a model modify data, not something specific to a bug in this node.

Data changed and you're not sure why. If the LLM has write access through this tool, check the conversation that preceded the change - the model acted on some instruction or inference from the chat, even if the connection between "what the user asked" and "what got edited" isn't obvious at first glance.

The agent never edits anything. Same as any tool-calling setup - the model needs a reason to use the tool. If your agent's instructions don't mention it can modify the knowledge base, it may default to just answering from what it already has rather than reaching for this tool.

You actually only wanted read access. Use KG_csv_toolkit_user instead - no reason to grant write capability to an agent that only ever needs to look things up.