What I promise, what I don’t, and how you can check.

What a local installation guarantees

• Custody. Your documents and prompts are processed on hardware you own. There is no vendor whose retention policy, terms change, or breach exposes that data — because there is no vendor in the processing path.
• Verifiability. The claim above is testable: we unplug the network cable and your workflow keeps working. Every setup ends with that test.
• Auditability. Agents I install log their actions to records on your disk that you can open and read. What the AI did is never a mystery.
• Continuity. No subscription in the processing path means nothing shuts off when a payment lapses. Caretaking is optional and cancelable.

What it does not guarantee

• Regulatory compliance. Running AI locally changes who holds your data; it does not by itself satisfy any regulation or professional obligation. Whether your obligations are met is a question for you and your counsel — I’ll give you an honest data-flow write-up to take into that conversation.
• Invulnerability. No system is “100% secure,” and I won’t tell you otherwise. A local machine can be stolen, infected, or misconfigured. What changes is that securing it is in your hands, and I set it up so the defaults are sane: local-only network binding, no telemetry pipelines, least-privilege agents that fail closed.
• Frontier-model quality. Open local models are genuinely good at drafting, dictation, and searching your own documents. For some tasks, cloud models are still better. If your job is one of them, I’ll say so at intake.

If your work is regulated

I’ve built for regulated data before, and that work taught me the discipline this page reflects: consent and review gates in the product, logs that don’t contain the sensitive payload, and claims kept within what’s provable. Bring your compliance officer to the intake call — the architecture conversation is exactly what they’ll want to have.

The claims standard for this site

Every factual claim on this site is one I can substantiate — from project pages that state what’s unfinished, to the absence of testimonials I don’t have. If something here is wrong, tell me and it gets corrected, visibly. This page is the standard the rest of the site is audited against, and that audit is repeated before major changes ship.

If a claim seems too smooth, ask me for the receipt: brian@niceley.ai.