Griptape Embedding Driver: LM Studio

Here's the cheapest embedding driver in the pack: it costs nothing per call because it runs against LM Studio, the desktop app that hosts local models, on your own machine. Griptape Embedding Driver: LM Studio points Griptape's embedding requests at LM Studio's OpenAI-compatible server on http://127.0.0.1:1234 - no API bill, no data leaving your computer, no Hugging Face queue.

If you're already using LM Studio for chat models, this is a no-brainer for experiments: install the app, load an embedding-capable model, and your RAG pipeline has local embeddings in minutes. The tradeoff is that local embedding models are typically smaller and faster than the managed giants - fine for personal projects, less so for production-grade retrieval.

How it works

Configuration node, pack-standard: it produces an EMBEDDING_DRIVER that downstream RAG/vector-store nodes use to convert text into vectors, with LM Studio doing the actual inference. Technically it's a subclass of the OpenAI-Compatible embedding driver, because LM Studio speaks that protocol.

The inputs:

• embedding_model - which model LM Studio serves embeddings from. Load it in the app first.
• base_url - defaults to http://127.0.0.1; the local server.
• port - defaults to 1234, LM Studio's default server port. If you changed it in the app, change it here.
• use_native_tools - a boolean with a slightly confusing name for an embedding driver. It's inherited from the OpenAI-compatible base; leave it off unless you have a reason.
• api_key - defaults to the string lm_studio. LM Studio's local server doesn't actually check keys, so any value works; the default is just there to satisfy the API shape.

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.

You'll also need LM Studio itself (lmstudio.ai): install it, load an embedding model, and enable the local server. 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 breaks ComfyUI's build).

Gotchas

"Connection refused" is the #1 issue, and it's almost always the same cause: the LM Studio server isn't running, or it's on a different port than the node's 1234 default. Start the server in the app before running the workflow. Second: LM Studio must have an embedding model loaded - a chat model doesn't serve embeddings, and requests will fail or return garbage. And because everything runs on your machine, embedding a big corpus takes local CPU/GPU time; don't compare its speed to a managed API.