(Deprecated) Tara LLM Daisy Chain Node (Deprecated)

"(Deprecated) Tara LLM Daisy Chain Node (Deprecated)" - the display name says it twice, just to be sure you got it. TaraDaisyChainNode was the pack's original text-composition node, and its whole trick was making one LLM call's output reusable as the next stage's input, so you could build multi-step chains like checklist → verify → execute → refine inside a single ComfyUI graph. It's marked deprecated because the author replaced it with TaraAdvancedComposition and explicitly renamed the concept - "Composition is a better and more apt name than Daisy Chain." If you're loading an old workflow that still uses it, it works. Just know you're on the legacy path.

How it works

It's a single chat completion, and the clever bit is in how inputs become messages. Your guidance becomes the system prompt; prompt becomes a user message; and if you also provide positive or negative (both optional), each becomes its own labeled user message. The model replies with freeform text - no JSON, no formatting - and that text is your output_text. Because nothing scrubs it, you can feed it straight into the next node's prompt field and chain forever.

The inputs worth knowing:

• api_key and model - the legacy raw-input pair. Dropdown has the same 2024-era list (groq/mixtral-8x7b-32768 being the README's favorite, and genuinely a good free option back then). Only OpenAI and Groq work here; the endpoints are hardcoded.
• temperature (0.4), max_tokens (1000), top_p (1), frequency_penalty / presence_penalty (0) - the sampling controls, all exposed on the node itself because this predates the config system.
• guidance, prompt, positive, negative - the text material, all multiline.

Why it got deprecated

Everything that made it fiddly is the legacy design: raw api_key input instead of the key loader, a frozen hardcoded model list, no seed, hardcoded providers. Two inherited annoyances carry over: the code prints your API key to the ComfyUI console on every run (a leftover debug line), and if you wire a key from a text node, that key can end up embedded in a shared workflow file. The modern replacement, TaraAdvancedComposition, takes a llm_config object instead - so it gets seed support, configurable sampling, and any OpenAI-compatible backend including local Ollama.

Should you use it?

Only if you're maintaining an old workflow. There's no functionality here that the composition node doesn't do better, and the display name is the author's own roadmap. New work should go straight to TaraAdvancedComposition + TaraLLMConfig/TaraPresetLLMConfig.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/ronniebasak/ComfyUI-Tara-LLM-Integration

Restart, or Manager → Install via Git URL. Deps are just orjson and openai - no models to download. The pack is a small alpha from ronniebasak that's been quiet since 2024, so expect deprecated nodes to keep running but never improve. It's a fine way to see the original Tara idea, and a good reason to appreciate what the config-based rewrite fixed.