GALIAIS-Nodes AI Provider Health Check

The first time you wire up an AI node in this pack and hit run, you'll get a wall of JSON and no idea whether the problem is your key, your URL, or the model name. This node is the answer to that specific pain: it takes the AI服务商 (AI provider) object from the GaliaisNodesAIProvider node and actually pokes the endpoint to find out what works.

Think of it as a ping before you commit. It checks whether the provider's /models endpoint responds, whether a real chat completion round-trips, and optionally whether streaming works - and it times each one. Out the other side you get 可用 (a boolean, is this thing alive), 延迟毫秒 (latency in milliseconds), a 健康状态JSON blob with per-check results, and 建议 (a plain-language recommendation string). That last output is the one worth reading; the author built it to hand you the next step, like "close streaming on the provider node" when the stream check fails.

How to use it

It's dead simple because it has almost no knobs of its own. Feed it the provider object, then set the check toggles:

• 执行检测 (run check) - this is the master switch, and it's False by default. This pack's pattern is "don't fire network calls until I say so": with it off, the node returns immediately with 可用 = False and a recommendation to turn it on. Flip it to True and re-run.
• 检测模型列表 (check model list, default on) - calls /models and reports whether your configured model is actually listed. This catches the classic "I typo'd the model name" failure.
• 检测Chat (check chat, default on) - sends your 测试提示词 (test prompt, default Reply with OK.) through a real completion. This is the test that burns a tiny bit of your API credit, and it's the one that proves the key works.
• 检测流式 (check streaming, default off) - verifies streaming responses, since some providers advertise SSE support and then don't deliver.

Wire the 可用 boolean somewhere visible, or just read the 建议 string in the node preview. In the reference workflow this node sits right after the AI Provider, which is the right call: you want to know the endpoint works before you build ten AI nodes on top of it.

What it's for, and what it isn't

This is a configuration-time tool, not a runtime guard. It does one probing round when you execute it; it doesn't monitor the provider continuously or fail a generation that's mid-flight. If you're a beginner, use it once when you first configure the AI Provider, then bypass it or leave it in the graph as a manual check. If you ever see a GALIAIS AI node erroring, come back here and re-run - the 建议 output will usually tell you whether it's the URL, the model name, or streaming that's lying to you.

Install and gotchas

Same as the whole pack - no separate setup:

git clone https://github.com/GALIAIS/GALIAIS-Nodes.git ComfyUI/custom_nodes/GALIAIS-Nodes

Restart ComfyUI, or search "GALIAIS-Nodes" in ComfyUI Manager. No pip dependencies, no model files. The one gotcha beyond the 执行检测 switch: this node makes live network calls and spends a fraction of a token per check, so if you leave it wired into a workflow you re-run a hundred times, you're paying a tiny tax per run. It's also a fresh, little-known pack - don't be surprised that there's no community writeup yet; the README and the health JSON are the docs.