Griptape Embedding Driver: Ollama

If you want to try Griptape's RAG features without signing up for any API, Griptape Embedding Driver: Ollama is the on-ramp. It produces embeddings locally through Ollama - the same tool you probably already installed to run Llama 3 or Mistral - so embedding is free, private, and runs on whatever you already have. It's the driver the README's install guide is built around.

The whole setup is: install Ollama, ollama pull an embedding model, and this node talks to it at http://127.0.0.1:11434. That's the entire cost of entry.

How it works

Configuration node, pack-standard: it builds an EMBEDDING_DRIVER that downstream RAG and vector-store nodes use to turn text into vectors. Ollama does the inference on your machine; nothing leaves it.

The inputs:

• base_url - http://127.0.0.1 by default, the Ollama server address. It can read a configured ollama_base_url from the pack's settings if you've set one.
• port - 11434, Ollama's default port. Both base_url and port are required inputs here (most pack drivers make everything optional), so if Ollama runs elsewhere or on another port, this is where you say so.
• embedding_model - which model to embed with, left blank by default. You need a real embedding model pulled locally - the chat models you run for chat won't serve embeddings.

What's notably not here: no API key, no env var, no cost. The node has zero credential inputs because there's nothing to authenticate against.

Installing

Ships in the ComfyUI Griptape Nodes pack:

• ComfyUI Manager: search "Griptape" → install ComfyUI-Griptape.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/griptape-ai/ComfyUI-Griptape, then restart.

Plus Ollama itself: download from ollama.com, then pull an embedding model:

ollama run llama3        # chat model, for the prompt driver side
ollama pull nomic-embed-text   # an actual embedding model

Pack dependencies are griptape[all], openai, python-dotenv, plus git-hosted extensions. The torch caveat applies pack-wide (reinstall torch with the CUDA index if Griptape's install conflicts with ComfyUI's build).

Gotchas

The most common failure is a mismatch between what the node lists and what Ollama actually has: if ollama list doesn't show the embedding model you typed, calls fail - pull it first. Don't put a chat model in the embedding_model field and expect vectors; Ollama serves chat and embedding from the same API, but the model itself decides which it does. And if you've changed Ollama's bind address or port, both fields on the node must match - the classic "connection refused" cause.