Griptape Agent Config: LM Studio [DEPRECATED]

The deprecated LM Studio config is what the pack's local-LLM support looked like before the "Drivers" family arrived. It still runs - old workflows depend on it - but it only wires up a prompt driver against LM Studio's local server, and it predates the cleaner modular configs. If a downloaded workflow throws this node at you, here's what it is and how to handle it.

What it is

A single config node that points a Griptape agent's prompt driver at an LM Studio model served on localhost. Three inputs are required: model (the LM Studio model name), base_url (default http://127.0.0.1), and port (default 1234). On the optional side you get the old-style set: temperature, seed, max_attempts_on_fail, env, use_native_tools (defaults to off here, a sensible choice for local models), and max_tokens. It outputs a single CONFIG.

The mechanism is the same as its modern sibling: it constructs an OpenAI-style chat driver pointed at your LM Studio server (http://127.0.0.1:1234 by default) and bundles it into a config.

Why it's deprecated

The replacement, "Griptape Agent Config: LM Studio Drivers," does everything this node does and adds an embedding driver plus a placeholder lm_studio API key - so agents get memory and RAG support out of the box. This old node is prompt-only, which means a memory-equipped agent required bolting on separate embedding and vector store nodes. The deprecated version also lacks the newer connection handling and the model list niceties. It's kept alive so legacy workflows don't break, nothing more.

Installing it

It ships in the one pack, no separate install:

cd ComfyUI/custom_nodes
git clone https://github.com/griptape-ai/ComfyUI-Griptape

or ComfyUI Manager → search "Griptape". Restart, and you need LM Studio itself running with a model loaded and its local server started. No API key needed - that's the appeal of the local path.

Should you use it?

Only for maintaining an existing workflow. The migration to the Drivers version is trivial: same CONFIG socket, same base_url/port fields, and you gain an embedding driver slot (and the lm_studio placeholder key that makes auth just work). Building anything new, there's no reason to reach for the deprecated node.

Common issues

• Connection refused - LM Studio's server isn't running or the model isn't loaded. Start the server, load the model, retry.
• Model name mismatch - model is free text and unvalidated here; it must match what LM Studio has loaded exactly.
• No agent memory - expected with the old design; add embedding/vector store nodes manually, or switch to LM Studio Drivers.