September 2026: Where Neuron Runs, and the Languages It Writes In
September had four stories on neuron.ws, more than any month before it. Read together, they answer a question prospective customers reasonably ask before anything else: will this fit the way we are required to work?
Review with a memory
We opened the month with a short video arguing that track changes has a flaw (opens in a new tab). Once a change is accepted in a typical editor, any trace of it disappears. In Neuron, every field keeps a permanent record of its edits, which changes how review works in practice. A reviewer can read only what a tester added to approved library text, instead of re-reading the whole finding to spot what moved. The track changes page (opens in a new tab) shows the idea in under two minutes.
AI on hardware you own
A few days later came Neuron AI on your own hardware (opens in a new tab), the most complete account yet of how our AI is deployed and managed.
The models live on the customer’s server and the work happens there. Teams can send different tasks to different models, so a heavyweight model handles executive summaries while a faster one helps with everyday drafting. Drafts are matched against the findings library by meaning rather than by exact wording, and a built-in benchmark measures how a given server and model actually perform, so model choices rest on numbers from the team’s own hardware. Every generation is logged, step by step. Nothing about it is meant to be a black box.
Running anywhere, including nowhere near the internet
Mid-month we published how to deploy Neuron anywhere, including air-gapped networks (opens in a new tab). In government work and regulated industries, the location of engagement data is usually settled by contract long before anyone installs software.
The story covers three pieces. Offline Mode, a single setting that blocks every outbound connection Neuron would otherwise make. A guided first run that takes a new server from installation to a first client. And deployment guides for the major public clouds, alongside one for networks with no outside connection at all. Single sign-on and multi-factor requirements round out the picture for organizations with an identity team of their own.
One engagement, many languages
We finished September with reporting in every language your clients read (opens in a new tab). Firms that deliver across borders often bolt a translation step onto the end of each engagement, and their carefully maintained library stops helping the moment the language changes.
In Neuron, language belongs to the engagement. Library findings, reporting briefs and report templates each carry versions per language, and an engagement automatically picks up the right one. Neuron AI drafts in the engagement’s language, and because library matching works on meaning, a draft can still find the right entry when the two are written in different languages.
The theme of the month
Each of these stories is about control. Control over what changed in a finding, over which hardware runs the AI, over what the server is allowed to reach, and over the language a client receives. Those are the conditions under which many security teams are allowed to adopt a platform at all, and they have shaped Neuron from the start.
All four stories are on neuron.ws/news (opens in a new tab). To talk through a deployment that fits your environment, schedule a demo (opens in a new tab).
Get the next one by email.
Neuron product news and what we are building at PenTest.WS, sent to your inbox. Unsubscribe anytime.
See Neuron on your terms.
Tell us about your team and environment. We will show you Neuron running the way you would run it: on your infrastructure, under your control.