LLM Loader (LM Studio) v1.0 - Akki

The LLM Loader (LM Studio) v1.0 is the easiest on-ramp into the whole AkkiNodes suite, because it means you don't have to fight llama-cpp-python wheels at all. Instead of loading a GGUF inside ComfyUI, it connects to a running LM Studio server and hands the pack an llm_model proxy object that every Akki agent node accepts. If you already use LM Studio for chat, you're one checkbox away from a working pipeline.

The name is a slight lie, by the way - it doesn't download anything and needs no API key. LM Studio exposes an OpenAI-compatible local server, and this node just points at it. All it does is create a client against your server address and return a proxy that routes every completion call to LM Studio. The heavy lifting - model loading, GPU offloading, context management - happens in LM Studio, where you've already got it configured.

Setting it up

In LM Studio: load a model, go to the Developer / Local Server tab, and start the server (default port is 1234). Then in ComfyUI, the node's single required input is server_address, defaulting to http://localhost:1234/v1. If your LM Studio is on another machine or port, change it. That's the whole config.

The proxy is implemented via the openai Python package with an OpenAI-compatible client, and it's been hardened across versions - the current one builds the request payload manually so the stop token handling doesn't break. It reports token usage to the console on every call, which is a nice touch when you're wondering where your VRAM went.

The one input, and how to use the output

• server_address - the only input. Default http://localhost:1234/v1.
• llm_model - the output. It's the same LLM_MODEL type the GGUF loader produces, so it plugs into every agent node in the pack: Cinematographer, ScriptCrafters, Lookdev nodes, Scene Choreographer, all of them.

Swap this loader in wherever you'd use the pack's GGUF-based LLM Loader, and the rest of your workflow doesn't change.

Install and gotchas

Pack install: ComfyUI Manager search "AkkiNodes LLM Suite", or git clone https://github.com/routhakash/AkkiNodes-LLM-Suite-for-ComfyUI into custom_nodes, run install.bat as admin on Windows (it installs the openai package this node needs along with llama-cpp-python), restart.

The gotchas are all about the server. If LM Studio isn't running, every call fails - and here's the trap: the proxy catches the error and returns a string that literally starts with ERROR: API call to LM Studio failed. Is the server running? instead of raising. Your pipeline keeps running on garbage text. So if your screenplay suddenly reads like a tech-support error message, check that LM Studio is up before debugging anything else. Also, LM Studio can unload the model if it's out of memory, which manifests as the same failure. One more: this is the local path - no API fees, but the response speed is whatever your GPU gives you. If you're on a 12GB card, a big model in LM Studio will be slow, same as any local LLM.