Bring your own model

PathWise's default reasoning engine is a frontier model you connect with your own API key. Here is how the presets, the keys, and the tradeoffs work.

Why bring your own key

A frontier model, the kind run by a large model provider, reasons more deeply on a hard, multi-step diagnosis and drives the probe loop reliably, which is why it is the default. The cost of that depth is that your investigation, including the network details PathWise gathers, is sent to that provider. If you want the reasoning step to stay on your own box, point PathWise at a local model instead, see running a local model.

The key itself belongs to you and your account with that provider. PathWise stores it encrypted at rest so it can use it on your behalf, it does not issue you a key of its own. See Security posture for how that storage works.

Curated presets

Rather than asking you to hand-configure an endpoint from scratch, PathWise ships five curated frontier presets, each pointing at a specific provider: Anthropic Claude, OpenAI GPT, xAI Grok, Google Gemini, and DeepSeek. The presets are data, not hardcoded choices, so the set gets re-curated as vendors ship new models without waiting for a new engine release. There is also a separate Your own LLM option for any OpenAI-compatible endpoint you host or subscribe to.

Verified versus reserved

Not every preset is treated as equally proven. Because tool-calling fidelity varies by model, a preset is only marked verified once its tool-calling has actually been exercised end to end against a real investigation. Until then, it ships marked reserved, and the settings screen shows the label. Today, Anthropic Claude is the verified preset, and the other four ship reserved.

Entering your key

Pick your preset in Settings, under Model, in the Reasoning engine field, then enter the key in the field beside it and press Test connection to confirm the provider answers before you run an investigation. Each provider you connect gets its own stored key, kept separate from the others, so switching engines does not require re-entering a key you already gave it. This lets you compare a local model against a frontier model, or two frontier providers, on the same kind of investigation, each keeping its own credential. A new key takes effect on the next investigation, no restart needed.

What stays the same either way

Connecting a frontier model changes which model is thinking. It never changes what PathWise is allowed to do. Whichever engine you choose, PathWise only gathers evidence and proposes a fix, with its reasoning and confidence shown. It never executes a change on your network.