Ollama

Ollama is the most popular way to run local LLMs around, so of course this pack has a node for it. What's interesting is how the node is wired: it doesn't talk to Ollama directly. Look at the defaults and you can read the author's setup between the lines - the model is named ollama/mistral and the api_type is litellm. That's the LiteLLM convention, a proxy layer that sits in front of Ollama and exposes it as yet another OpenAI-compatible endpoint. So this node assumes you're running a LiteLLM proxy, not raw Ollama.

That's the trap, and it's a real one: the stock defaults point at http://0.0.0.0:8000, which is LiteLLM's port, not Ollama's (that's 11434). If you followed "install Ollama and go," nothing here works out of the box, and the failure looks like a mystery connection error.

The three fields

• model - default ollama/mistral. The ollama/ prefix is LiteLLM's routing syntax telling the proxy which backend to hit. Change the model name to whatever you've pulled in Ollama.
• api_type - default litellm. This is what tells the pack how to build the connection; leave it unless you know what you're doing.
• api_base - default http://0.0.0.0:8000. This is where the LiteLLM proxy lives. If you're running LiteLLM locally, localhost is friendlier than 0.0.0.0, and if you skipped the proxy entirely you'd point this at http://localhost:11434 and hope the OpenAI-compatible endpoint accepts the connection.

Output: LLM, into any agent node.

Getting it to actually work

# with Ollama installed and a model pulled:
pip install litellm
litellm --model ollama/mistral --port 8000

Then wire Ollama → Assistant → Chat and go. The proxy step is the friction, but it's also the insight: LiteLLM unifies a hundred model providers behind one API, so the same node could later point at a cloud model just by changing strings.

Expect the usual

The pack was unfinished ("not everything is functioning correctly yet") and its repo is now dead - xXAdonesXx/NodeGPT 404s as of this writing, author's account gone, so the README's install is historical:

cd ComfyUI/custom_nodes
git clone https://github.com/xXAdonesXx/NodeGPT

Restart to auto-install requirements. In 2026 you'll find far better Ollama nodes in maintained packs, many of which talk to port 11434 directly and save you the proxy entirely. But if you ever stare at an old workflow's Ollama node wondering why it won't connect, now you know: it's looking for LiteLLM on port 8000, not Ollama.