Time Tool

LLMs don't actually know what time it is. Ask one directly and it'll either refuse, hedge, or confidently guess wrong based on whatever its training cutoff was - it has no built-in clock. time_tool is LLM Party's fix for that specific gap: it's a callable tool that returns the real current time for a given timezone, so an agent that needs to answer "what's today's date" or reason about time-sensitive things can just ask, instead of hallucinating an answer.

Like the rest of this pack's tool family, it doesn't do anything by itself when the graph runs - it packages a capability into a STRING and outputs it under the socket name tool. That output is meant to flow into tool_combine or tool_combine_plus, which merges it with whatever other tools you're giving the agent, and the combined bundle feeds into your LLM/agent node. The model decides at conversation time whether it actually needs to call it.

Inputs and outputs that matter

• timezone - defaults to Asia/Shanghai, a leftover of the pack's Chinese origins. Change this to your own IANA timezone string (America/New_York, Europe/London, and so on) or the tool will happily return the correct time - just in the wrong place for you.
• is_enable - the standard skip switch.

The single output, tool, is the packaged capability - not a live time value itself, since this node doesn't emit anything usable directly at graph-build time.

Installing it

Search comfyui_LLM_party in ComfyUI Manager and install, or clone by hand:

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

pip install -r requirements.txt inside your ComfyUI Python environment, then restart. No model or API key needed for this specific node, but it's bundled with the rest of the pack, which has a much heavier overall dependency list - LLM API clients, local model support, RAG, TTS, and a long tail of other tools. If you only need API-based LLM calling, the README's only_api branch skips the local-model weight.

Common issues

The timezone default is the one thing almost everyone using this outside China needs to change first - leave it on Asia/Shanghai and your agent will answer time questions correctly but unhelpfully, several hours off from wherever you actually are.

Beyond that, remember this node only offers the capability - it doesn't force the agent to use it. If your model keeps guessing the date instead of calling the tool, that's usually a prompting problem, not a wiring problem: make sure your system prompt actually tells the agent it has a time-lookup tool available and when to reach for it, rather than assuming the model will notice on its own that one's connected.