Griptape Util: Create Model from Modelfile

This is the second half of the Griptape pack's "bake your agent into an Ollama model" pipeline. Where "Create Agent Modelfile" generates the recipe, this node executes it: give it a Modelfile string and a name, and it writes the Modelfile to disk and runs ollama create to build a real, usable custom model on your local Ollama server. After this node runs, you've got a new model listed in Ollama that behaves like the agent you built - no ComfyUI required to use it afterward.

That distinction is the whole point of the two-node flow, and it's worth stating plainly because it changes how you think about the pair. The first node is pure text generation - it can't build anything. This node is the one that actually shells out to Ollama and produces a model. You can also use this node standalone: any valid Modelfile string works, not just one from the sibling node. Want a custom model with a different system prompt, different parameters, a different base? Craft the Modelfile yourself (or in a text node), feed it in, and out comes a model.

How it works

The node takes your Modelfile text, parses the FROM line to learn the base model, writes the file to disk as Modelfile, then runs two Ollama commands: ollama cp <base_model> <new_name> to copy the base, then ollama create <new_name> -f Modelfile to build the customized model on top. The resulting model is named {new_model_name}-{base_model} - so "new_model" over "llama3" becomes new_model-llama3. Note the code drops any tag after a colon in the base name, so llama3:latest becomes llama3.

The inputs that matter

• modelfile - required, and it's a force-input string, meaning it must be connected from another node (like the Create Agent Modelfile node) rather than typed in. It must contain a valid FROM line - that's how the node figures out the base.
• new_model_name - default new_model. The base for your custom model's name. Keep it short and descriptive; the final name gets the base model appended.

Two outputs: OUTPUT (STRING, the result of the ollama create call) and NEW_MODEL_NAME (STRING, the built model's full name - handy if you want to wire it back into a driver later).

Install

Same pack, same Ollama prerequisite:

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

or ComfyUI Manager → search "Griptape" → install → restart. And Ollama must be installed and running - this node literally invokes ollama commands, so it refuses to validate otherwise.

Where people get burned

The most common issue is connecting a Modelfile without a FROM line - the node returns nothing useful because it can't figure out the base. Make sure the first line of your Modelfile says FROM <model>. Second, Ollama has to be reachable; if you've moved it off localhost you'll need to make sure it's running, since this node uses Ollama's CLI directly. Finally, expect this node to do real work: ollama create on a big model isn't instant, and it consumes disk. If the OUTPUT comes back with an error, read it - it's Ollama's own message, and it'll tell you exactly what went wrong (missing base model, tag collision, etc.).