How to Use Ollama in a Workflow Automation Tool (No Cloud Needed)

Every AI workflow tutorial ends the same way: “add your OpenAI API key here.” That’s fine until you automate something sensitive (client data, internal reports, confidential research) and realise every piece of text you process is sent to a third-party server.

Ollama changes that. It runs LLMs on your own machine, offline. Connect it to a node-based workflow tool and you get AI automations that process data without any of it leaving your device.

This article covers what Ollama is, how it plugs into workflow automation, and what that combination unlocks.

What is Ollama?

Ollama is an open-source tool to download and run large language models locally, on your own hardware. It serves the model through a small HTTP server on http://localhost:11434, with its own API and an OpenAI-compatible one, so many tools can use it in place of a cloud provider.

Once it’s installed, you use it from the command line:

ollama pull llama3.2
ollama run mistral
ollama run gemma3

Ollama handles the download, the quantised model files and the choice of GPU or CPU. On a recent laptop with 16 GB of RAM, models like Mistral 7B and Llama 3 8B run comfortably; with a good GPU, bigger models become practical.

Why local models matter for automation

When a workflow step summarises text, classifies data or writes something, it sends your data to a model. With cloud models, that’s OpenAI, Anthropic or Google. Often that’s fine. It’s a hard limit for:

Running the model with Ollama removes that limit. The model runs on your machine, the data stays there, and the automation works the same, minus the cloud dependency.

How Ollama connects to Agentic Workflow

Agentic Workflow is a node-based workflow tool that runs as a Chrome extension. It has an Ollama model node you plug into any AI step: Basic LLM Chain, Summarization, Text Classifier, Information Extractor, agents and more.

Running Ollama models also show up automatically in the assistant’s and agents’ model pickers, and you can add Ollama as a provider in Settings › Providers. If Ollama isn’t an option on a machine, the Web LLM node runs a smaller model inside the browser instead.

What you can build: local AI workflow examples

Confidential document summariser

Email triage for a sensitive inbox

1. Context Menu trigger while you’re viewing an email
2. Get Element Data reads the email body you picked once
3. Text Classifier with Ollama sorts it: urgent, action needed, FYI or spam
4. Highlight Elements or Show Notification shows the label

Metrics from internal reports

1. Schedule trigger
2. Tab Control opens the internal report
3. Get All Text reads it
4. Information Extractor with Ollama pulls the key metrics as structured fields
5. Google Sheets (or a local file) records them

Choosing a model

Not every model suits automation equally. Sizes are Ollama’s default 4-bit downloads:

For most automation tasks (summarising, classifying, extracting), Mistral 7B or Llama 3 8B are the sweet spot: answers in seconds and accurate enough for real use. More in best free LLMs for workflow automation, and test on your own inputs with do you need a big AI model?.

Try it

You get Agentic Workflow’s browser-native execution (page interaction, browser triggers, reading page content) with private AI reasoning on your own machine. No API key, no cloud, no data leaving your device.