FEGenStringNBus

Some nodes are infrastructure for the community; some are infrastructure for their author. FEGenStringNBus is firmly the second kind. It's a prompt-generation node that talks to a service called "NBus" - the default api is http://127.0.0.1:24113/v1/chat/completions, the default model is sd_axl3plus_v2, and unless you happen to run that exact private service, this node has nothing to connect to. It's the most author-specific node in the pack, and the honest review starts and ends there.

If you squint at what it does, it's interesting: it sends a list of "pkw" (prompt keyword) values plus an optional background context to a custom OpenAI-format endpoint, and gets back a generated prompt and a parsed keyword structure. The response handling even extracts structured fields out of the streamed text. So it's designed for a workflow where you drive prompt generation from a keyword list rather than a free-text sentence - a particular philosophy of prompt construction that the author clearly built a service around.

The inputs

• pkws - *-typed, the keyword list that drives the request. This is the input that matters.
• api - the endpoint, defaulting to the NBus address from config.yaml.
• model - default sd_axl3plus_v2.
• cache_charge - boolean, default on. When on, the node caches results per model name in memory so repeat calls with the same model reuse the earlier answer; turn it off to always hit the server.
• background (optional) - extra context sent along with the keywords.

Outputs: prompt (STRING) and pkw (*) - the generated text plus the keyword structure.

For anyone who isn't running the author's NBus service, this node is dead weight: every run is a connection refused on port 24113 and a traceback in the console. Don't install the pack expecting this one to work - it's a red herring in the search results. The genuinely useful prompt generators in this pack are FEGenStringGPT (works with any OpenAI-compatible server you run) and, for the author's own Docker setup, the BC nodes. If you are building a keyword-driven prompt service and want a client node for it, this gives you a working template - the parsing of structured output out of an SSE stream is worth borrowing even if you rewrite the endpoint. But as shipped, for the average user, "skip this one" is the correct advice.