Open Web tool

The agent-callable sibling of this pack's Open Web node. Where that one has a path_or_url field you set yourself and runs every time deterministically, this node has nothing to configure at all beyond enabling it - because the URL or path isn't decided by you, it's decided by the LLM, at call time, based on the conversation. There's no other field for it to come from in the schema, so whatever this tool opens, the model is supplying that argument itself as part of the tool call.

How it works

Wire the tool output into an LLM/agent node's tool-list input. During a conversation, if the model decides opening a link or file is the right move - following up on something it was told about, or acting on a user's request - it calls this tool with a URL or path of its own choosing.

The inputs and outputs that matter

• is_enable - the only input, the pack's standard bypass toggle.

One output: tool (STRING) - wire it into your agent's tool-list input, the same pattern as every other _tool-suffixed node in this pack.

How to install it

• ComfyUI Manager: search "comfyui_LLM_party", install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/heshengtao/comfyui_LLM_party, then pip install -r requirements.txt from inside the pack folder using ComfyUI's own Python, then restart.

No API keys of its own - it needs whatever LLM node you've already configured, since the model is what decides to use it.

Common issues & troubleshooting

Nothing visibly happens even though the agent says it opened something. Same underlying issue as the deterministic Open Web node: if your agent is running on a headless server or a cloud GPU worker with no display attached, there's simply nowhere for a browser to appear. Worth knowing before you build a demo around an agent that "opens a link" - on most cloud ComfyUI execution, it won't be visible to anyone.

The LLM never calls the tool. Whether the agent reaches for this depends entirely on its instructions establishing that it can - if your persona or system prompt doesn't mention the capability, the model may just describe a link in text instead of actually calling the tool.

The call fails or the model hallucinates a result. Confirm the LLM node this tool is wired into actually supports tool/function calling - not every model or API configuration does, and one that doesn't will either skip the tool silently or make up a plausible-sounding "result" instead of really opening anything.