🖥️Save Embeddings Database

Retrieval-augmented generation needs two halves: something that builds a searchable index out of your documents, and something that queries it later. Save Embeddings Database is the write half, inside comfyui_LLM_party's knowledge-base tooling - feed it text, and it chunks, embeds, and writes a local vector store to disk that another part of your workflow can search against later.

How it works

The node takes text, splits it into overlapping chunks (so context isn't lost at chunk boundaries), runs each chunk through a local embedding model, and writes the resulting vectors - plus enough metadata to map them back to their original text - to a file at the path you specify. It has no outputs on the graph; the entire point is the side effect of writing that file.

The inputs and outputs that matter

• file_content - the text to index. Typically the output of one of the pack's loader nodes (Load File, Load Wikipedia, a web-scraping tool) rather than something you type by hand.
• model_path - a local path to an embedding model. This node consumes a model, it doesn't fetch one for you - the pack has its own embedding-model loader nodes elsewhere for downloading or locating one; point this at whatever path that setup gives you.
• save_path - where the resulting vector database gets written.
• chunk_size / chunk_overlap (200 / 50) - how the text gets split before embedding. Smaller chunks give more precise retrieval later at the cost of losing broader context; the overlap keeps ideas that span a chunk boundary from getting cut in half.
• device (auto / cuda / mps / cpu) - where the embedding model runs.
• is_enable - the standard bypass toggle.

No outputs - this is a terminal node (is_output_node: true with an empty output list). Its job is done once the file is written to save_path.

How to install it

Search comfyui_LLM_party in ComfyUI Manager and install, then restart. Manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/heshengtao/comfyui_LLM_party.git

Run pip install -r requirements.txt from the pack's own folder with ComfyUI's Python, then restart. Embedding locally needs a real model on disk at model_path before this node can do anything - that's a separate download/setup step from installing the node pack itself, handled by the pack's embedding-model loader nodes rather than this one.

Common issues & troubleshooting

It errors immediately, before touching your text. Check model_path first - this node doesn't download or validate a model for you, it just tries to load whatever's at the path you gave it. An empty or wrong path fails at that step.

Retrieval later feels off - irrelevant chunks coming back. That usually traces back to chunk_size/chunk_overlap, not the retrieval step itself. Chunks that are too large blur distinct topics together; chunks too small lose the context needed to match a query meaningfully. There's no universal right value - it depends on how dense and how long your source documents are.

GPU embedding fails. Same story as the pack's other device: auto/cuda/mps/cpu nodes - if auto/cuda can't find a working GPU setup, force cpu explicitly rather than debugging driver issues mid-workflow.