MNeMoNiCuZ-ComfyUI-mnemic-nodes/README/llm_request.md

✨ LLM Request

One node for every language model: ChatGPT, Claude, Gemini, Grok, Groq, OpenRouter, Mistral, DeepSeek, local servers like Ollama and LM Studio (on your own PC or elsewhere on your network), and your Claude Code or Codex subscription through their CLIs. It shares its prompt presets with the Groq nodes.

Features

• Built-in endpoints for all of the above, and any number of your own in nodes/llm/UserEndpoints.json.
• Local Claude Code / Codex subscriptions: runs the CLI on your PC with your subscription, no API key needed.
• Custom Endpoint - WARNING: enter an address and key right on the node. They're stored on your machine, not in the workflow; read the risks in the node's help panel before using it.
• Nothing secret in your workflows. Keys and private addresses live in .env. A shared workflow or image only contains the endpoint name and the model name.
• Status panel on the node showing where the endpoint runs (🖥 this PC, 🏠 network, ☁ cloud), whether its key is set, and what is missing if not.
• 🔍 Model browser: a searchable list of the models the endpoint offers, with context size, parameter count and quantization, and which Ollama models are already in memory.
• ⚡ Connection test and 📜 preset viewer.
• Live streaming preview of the reply, with a collapsible 💭 thinking section, token counts and speed.
• Vision: connect images and every image in the batch is sent.
• Reasoning control: one setting maps to reasoning_effort (OpenAI-style), think (Ollama) or Claude's extended thinking.
• Self-correcting requests: parameters a model refuses (e.g. temperature on OpenAI reasoning models) are dropped and the call retried automatically.
• Retries with backoff on rate limits and server errors, and Cancel stops a streaming request mid-reply.
• VRAM friendly for local models: unload the Ollama model after replying, or free ComfyUI's models before the call.

Setup

1. Start ComfyUI once. A .env file is created in this pack's folder from .env.example.
2. Fill in the keys and addresses you use, e.g.:

   OPENAI_API_KEY=sk-...
   ANTHROPIC_API_KEY=sk-ant-...
   OLLAMA_NETWORK_URL=http://192.168.1.50:11434
   

3. Pick the endpoint on the node. Its status dot turns green once its key and address are set; ⚡ Test checks that the server actually answers.

Ollama and LM Studio running on the same PC need no setup.

The full reference, covering every input, adding endpoints and every config field, is in the node's help panel (the ? on the node), or here.