Chat with Ollama

Runs locally

Needs Ollama with one model pulled. No key, and what you paste stays on your machine.

Paste a blunt work message and get it back clear and kind, ready to send. A chat box, a prompt with your instructions, a local model and a view of the conversation. Everything runs on your machine, so you can paste a message about a customer or a colleague without sending it to anyone.

Download the graph and save it in user/graphs/ inside your Boltjar folder; it shows up in your saved graphs. Or build it yourself below. The graph is MIT-0: copy it and license what you build from it however you like.

1. Pull a model

ollama pull gemma4:12b

The figures on this page use Gemma 4 12b, about 8 GB on disk. Any Ollama chat model you pull shows up in the LLM’s picker; the LLM node’s page lists the ones that ship with a manifest.

The downloaded graph has no model picked, because a model can cost money and Boltjar never picks one for you. Until you pick one, the LLM answers with a mock reply.

2. The nodes

Node	Settings
Chat Input, named Message	placeholder Paste a message to rewrite...
Template, named Prompt	the text below
LLM	model: pick Gemma 4 12b (Ollama), or the model you pulled
Chat	none

The Prompt’s text:

You rewrite short work messages so they read clear and kind. Reply with the rewrite only.

Message: {message}

“Reply with the rewrite only” keeps the model from adding a greeting or an explanation, so the reply is ready to copy.

3. The wires

From	To
Message trigger	Prompt trigger
Message text	Prompt message
Prompt trigger	LLM trigger
Prompt out	LLM prompt
Message trigger	Chat user_trigger
Message text	Chat user
LLM trigger	Chat reply_trigger
LLM response	Chat reply

A wire dropped on the Prompt’s + tag socket makes a tag named after the node it comes from, so Message text becomes {message}. Every trigger input needs a wire, the Prompt’s included, or the graph stays Off. The trigger runs through the Prompt to the LLM: the message fires the Prompt, which fills in its text once and fires the LLM, so one message runs the model once.

1. The message socket the wire from Message text made.
2. The trigger runs through the Prompt to the LLM, so one message runs the model once.

The Chat sits right of the LLM. It takes the Message’s two wires on user_trigger and user, and the LLM’s two:

1. The reply: LLM response into Chat reply, with its trigger into reply_trigger next to it. The Chat shows the message and the rewrite, here with Gemma 4 12b picked.

4. Turn it On

Press On and paste a message into the Message node, for example Your invoice is overdue. Pay now. The Prompt fires on the message and fills in your text, the LLM fires next and reads that prompt, and the rewrite lands in the Chat node.

If Ollama is not running, the LLM turns red, and its error output fires with the reason. If Ollama runs but the model is not pulled, the graph stays Off, and its problems list says to pull the model in Settings › AI Providers.

Keep the conversation

This graph sends one message at a time: the model does not see earlier turns. The chat example that opens on the first start keeps the history in a SQLite table: a Database node, DB nodes that insert each message and read the last ones, and a Format List that turns the rows into lines for the prompt. Open it next to this one to see the pattern.

Change it

• A cloud model. Add an Anthropic, OpenAI or xAI key in Settings › AI Providers and pick one of that provider’s models on the LLM node. Without the key, the node answers with a mock reply.
• The date in the prompt. Add a Time node named Date, wire its out onto the Prompt’s + tag (the tag is {date}), and write Today is {date}. in the Prompt.
• Reading images. A model that reads images (Gemma 4, Qwen3.5, Claude, GPT, Grok) grows an image input on the LLM. Wire an Image or a Screen Capture into it.