⚙️ Live Search Settings

⚙️ Live Search Settings is the least glamorous node in this pack and the one that decides how everything else behaves. The API Loader is the engine and the 🌐 Agent does the work, but this little node is the steering wheel - it emits a single search_settings wire that tells the Agent whether to search, how to search, and what language the answer should come back in. Change one toggle here and the same workflow becomes a plain chatbot instead of a research tool.

What each setting actually does

Only two toggles, one dropdown, one counter, and an optional proxy field:

• mode - T2T (text-only) or TI2T (text + image). This is the master switch: it tells the Agent which of the two models on the API Loader to call, and TI2T makes the Agent expect an image input. Flip it to TI2T without feeding an image and nothing sensible happens.
• enable_web_search - the big one. Turn it off and the whole pipeline skips DuckDuckGo entirely and just runs your prompt through the LLM. That's your "pure LLM chat" mode for free, and it's a genuinely handy way to reuse the pack as a plain chat node.
• num_results - 1–10 (default 3). How many search results get pulled and considered. Three is a sensible floor for a quick answer; go higher when you're researching a controversial or technical topic and want the Agent to weigh more sources.
• output_language - 中文 or English. The default is 中文. This is the one that'll surprise you: if you never touch it, every answer comes back in Chinese no matter what your prompt says. English speakers, set it once and save the workflow.
• optimize_query - on by default. Before searching, the LLM rewrites your prompt into better English search keywords (English indexes better on DuckDuckGo), then still answers in your chosen language. Leave it on unless you're debugging why a specific query got mangled.
• proxy - optional, and the placeholder (http://127.0.0.1:7890) is a tell that this pack was built by someone who needed a proxy to reach these services. If DuckDuckGo or the API calls time out in your region, this is where you fix it.

The one output

search_settings (type SEARCH_SETTINGS) feeds the Agent's search_settings input. That's it. If you have several Agents doing different jobs, you can branch one Settings node to all of them or give each Agent its own - the design encourages the latter when the tasks differ.

Installing it

Same pack as the rest: ComfyUI-Live-Search. ComfyUI Manager → search Live Search → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Zone-Roam/ComfyUI-Live-Search.git
cd ComfyUI-Live-Search
pip install -r requirements.txt

No models, no heavy deps - the requirements are just requests, beautifulsoup4, ddgs, python-dotenv, and geopy. This node itself is pure configuration; it doesn't even touch the network.

Where people get burned

• The Chinese default. You'll get 中文 answers until you flip output_language to English. It's the most common "why is it answering in Chinese" post in any workflow that uses this pack.
• Forgetting that enable_web_search exists. Turn it off and your "search" node is now just calling the LLM with no internet at all - which is fine if that's what you wanted, confusing if you expected fresh data.
• Proxy overreach. Only set proxy if something's actually timing out. A dead proxy address will break the search when no proxy at all would have worked.