Griptape Agent Config: LM Studio Drivers

LM Studio is the friendliest local-LLM app on Windows and macOS - point-and-click model downloads, a ChatGPT-style chat box, and an OpenAI-compatible local server. This config node turns that running server into the prompt and embedding drivers for a Griptape agent. Everything stays on your machine: no API key, no cloud, no data leaving the box. If Ollama feels too terminal-y for you, this is the local path of least resistance.

How it works

The node talks to LM Studio's OpenAI-compatible server and builds a DriversConfig from two drivers: a chat prompt driver and an embedding driver. Both are free-text model names you type yourself - the changelog notes they disabled auto-loading of the model list, so you enter the exact name of the model you've loaded in LM Studio. The pack even sets a placeholder api_key of lm_studio by default, which is exactly what LM Studio expects: any non-empty string satisfies its auth check.

The two required-ish connection fields are base_url (default http://127.0.0.1) and port (default 1234) - LM Studio's default server port. The output is a single CONFIG that feeds any Agent.

The inputs that matter

• model - the prompt model name, matching what's loaded in LM Studio (e.g. whatever shows in its model dropdown).
• base_url + port - where LM Studio's server is listening. Defaults are right unless you changed the port in LM Studio's Developer tab.
• embedding_model - for agent memory / RAG. LM Studio serves embedding models too; leave it blank and memory won't work, so if you want retrieval, load an embedding model in LM Studio and type its name here.
• use_native_tools - defaults to off for LM Studio, which is the safe choice: smaller local models often fumble LLM-native tool calling, so Griptape's own tool engine is the more reliable default.

Plus the shared temperature, seed, max_tokens, min_p, response_format, max_attempts_on_fail.

Installing it

Install the pack once:

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

or ComfyUI Manager → search "Griptape". Then install LM Studio from https://lmstudio.ai, load a model, and start the local server from its Developer tab. No keys to configure - this is the rare Griptape config with zero account setup. Shared caveat: griptape[all] pulls a torch build that can clash with ComfyUI's; README troubleshooting covers the fix.

Common issues

• Connection refused - LM Studio's server isn't running, or the model isn't loaded. Start the server and load the model first.
• Model not found / API errors - the name in model doesn't match what LM Studio has loaded. Type it exactly as shown in LM Studio's UI.
• No memory/retrieval - embedding_model is empty. Load an embedding model in LM Studio and fill the field.
• Tool calls flaking out - flip use_native_tools to off (the default) and let Griptape handle tools.

Community signal backs the local-first appeal here: when people ask about ComfyUI LLM pipelines, the go-to answers are Ollama and LM Studio, and Griptape's nodes get named as among the most complete of the LLM custom-node options. This config is the LM Studio side of that story.