🔑 Live Search API Loader

The name says "loader," which usually means a model file is about to land in your models folder. This one doesn't, and that's the first thing to get straight. 🔑 Live Search API Loader is a configuration block: you pick an LLM provider and a model, drop in an API key, and it bundles all of that into a single model_config wire that the other two nodes in this pack consume. Nothing downstream runs until it's configured, so every Live Search workflow starts here.

It's the first step in the pack's three-node chain - API Loader → ⚙️ Live Search Settings → 🌐 Live Search Agent. The modular split is borrowed from comfyui_LLM_party, and it's a good one: a single API Loader can feed multiple Agents, and you can flip search settings without touching your API config at all. That's the whole point of keeping it separate.

What you actually set

Of the eight inputs, three matter on day one.

• provider - 11 choices including OpenAI, DeepSeek (Official), Gemini (OpenAI-Format), Anthropic (Claude), Grok, Volcengine (Doubao), Qwen (Aliyun), SiliconFlow, Ollama (Local), and Custom. The default is DeepSeek (Official), so if you're here for a specific provider, change it first.
• t2t_model - the text-only LLM used in T2T mode. 129 choices. Defaults to deepseek-chat, which is exactly right if you keep DeepSeek and confusing if you switch providers and forget to update it.
• ti2t_model - the vision model used in TI2T mode (33 choices). It shows "No VLM models available" for providers without a vision lineup, which is fine - you only need this if you plan to feed the Agent an image.

The rest you mostly leave alone: temperature (0–2, default 0.7), max_tokens (default 2048), timeout (default 120s). api_key and base_url can stay empty if you use .env or api_config.json - empty base_url falls back to the provider's default endpoint.

The one output

model_config (type MODEL_CONFIG) plugs straight into the 🌐 Live Search Agent's model_config input. There's nothing else to do with it.

How your key actually gets in

The node resolves keys in a strict order: widget input > .env file > api_config.json. The README's recommended setup is a .env file - copy .env.example to .env and fill in keys like OPENAI_API_KEY=sk-... or DEEPSEEK_OFFICIAL_API_KEY=.... Keys never get written to disk by the node itself, which is the sane default for a shared GPU box.

Installing it

It ships as part of ComfyUI-Live-Search. Easiest path is ComfyUI Manager → Install Custom Nodes → search Live Search → install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/Zone-Roam/ComfyUI-Live-Search.git
cd ComfyUI-Live-Search
pip install -r requirements.txt   # or ..\..\..\python_embeded\python.exe -m pip install -r requirements.txt on Portable

There are no model downloads - this pack is pure API calls. Its real dependencies are light: requests, beautifulsoup4, ddgs, python-dotenv, geopy. None of them touch your torch environment, so this is one of the easier packs to install.

Where people get burned

• Provider/model mismatch. Defaults assume DeepSeek. Pick OpenAI and keep deepseek-chat and you'll get a confusing 404 from a model that doesn't exist there.
• GPT-5.x returns 403/404 even with a valid key. The README notes those models require a plan with GPT-5 privileges (Pro/Team/Enterprise) - it's an account permission problem, not a node problem.
• Ollama (Local) ignores most of this: no key needed, but Ollama itself has to be running and serving the model you name.
• Security. This is the node holding your API key, and the ecosystem has a documented history of malicious custom nodes (the LLMVISION malware was distributed exactly this way). Install from the official repo, and don't expose ComfyUI to the internet without auth - every node in your graph runs arbitrary code with your user account.