Run it locally with Ollama

The private path for the reasoning engine: a model running on your own machine, no API key, nothing sent out.

What Ollama is

Ollama is software that runs a language model directly on your own computer or server. Instead of sending your investigation to a company over the internet, the model runs locally, using your machine's own memory and processor. PathWise talks to it over a local connection that never leaves your machine.

Two things follow from that. First, you need no API key, since there is no outside provider to authenticate with. Second, the evidence PathWise gathers, including the network details in your investigation, does not go to an outside AI provider. The reasoning runs on your own hardware. Probes still reach the targets they test, and the privacy page covers everything that leaves the box.

The local path is a choice, not the default. PathWise defaults to a frontier preset with your own key, because small local models call tools less reliably on a hard, multi-step diagnosis. Pick the local tier when keeping everything on your own hardware matters more than reasoning depth.

Picking a model that is good at tool-calling

PathWise reasons by calling tools, such as a ping or a DNS lookup, reading the result, and deciding what to check next. Not every local model handles that loop equally well. Some models available through Ollama support this kind of tool-calling reliably, others do not, or do so inconsistently.

Because of that, pick your local model first for how reliably it calls tools, and only second for how small or fast it is. A model that cannot call tools correctly cannot run an investigation, no matter how capable it otherwise seems. The settings screen suggests llama3.1:8b as a starting point. In our testing, models that size handle simple investigations but fall short of a confident answer on hard, multi-step cases, so treat the local path as a privacy choice and switch to a frontier preset when depth matters.

The general shape of setup

Setting up the local path follows a consistent shape.

1. Install Ollama on the machine that will run PathWise, or on a machine PathWise can reach over the network. Use the install method in Ollama's own docs for your platform.
2. Start the Ollama service so it is listening for requests, and pull the model you picked for reliable tool-calling.
3. In PathWise, open Settings, go to Model, and set the Reasoning engine field to the local tier. Enter the endpoint, which defaults to http://localhost:11434/v1, and the model name you pulled.
4. Press Test connection to confirm PathWise can reach the model, then run an investigation to see it reasoning end to end.

If your hardware struggles

A small local model has a real ceiling. On a hard, multi-step diagnosis, it may reason less deeply than a larger model would, or fall short of a confident answer. Switching to a frontier model for that harder case is a setting change, not a reinstall, so you can move back to the local model at any time.