Direct answers about data, cost, adoption, and what happens if it isn't a fit.

No. Your private AI runs for your company alone on NVIDIA and Apple hardware hosted in the United States. Nothing your team types, uploads, or asks reaches a public AI provider. On consumer tiers, what your staff type can be used to train the model by default. On every tier, your data sits on a third party’s systems under terms that can change. Most staff use runs through personal accounts with none of the business protections.

Prompts and customer content are not sent to public model APIs for inference or used to train a shared model. In-Building approved AI work stays in your building. The deployment documentation identifies approved support, update, backup, logging, and integration paths so your company can review the complete data flow.

Public AI and client data guide

Outputs are drafts. Workflows include source material, review steps, and clear ownership. Your team remains responsible for documents, advice, and anything sent to a customer. For professional firms, this includes tax positions and professional decisions.

For professional firms: PrivateStride documents your AI system’s data flows and controls and works with your designated security lead as your firm considers updates to its written information security plan. Your Qualified Individual, counsel, and other advisers retain their responsibilities. PrivateStride provides technical and implementation support, not legal advice or a certification of compliance.

Safeguards Rule and AI guide

Your source documents remain in your systems. Before implementation or installation, the proposal documents offboarding responsibilities, access removal, hardware return, and the agreed retention or deletion process for PrivateStride-managed configuration and support records.

Medical practices are on our roadmap. We would need to document a BAA and the security posture first. Book a fit call to discuss timing.

Book a fit call

Every answer drawn from your documents shows its source, and your AI says ‘I don’t know’ when your documents do not contain the answer. A named person approves anything that reaches a client. Your first ninety days give you your own number for drafts accepted without correction, per workflow.

At first, yes: your people review every draft line by line. Review can move from reading each sentence to glancing at the cited source, then to exception review where your dashboard shows trust has been earned. Review never drops to zero for client-facing work.

Your model keeps no memory between conversations, so a user’s instructions end with that conversation. Only administrators can change behavior through a versioned, logged process. Every question, answer, and change is logged, and attempts to change behavior are flagged to your administrator.

Someone could put instructions in a file, but retrieved content is material to summarize or cite, never commands. Your system follows a fixed order of authority: system, administrator, user, documents. A retrieved document cannot override that order.

Your AI can only see documents the person asking can already open. Permission filtering runs before your model, so restricted passages never reach it. Nothing is indexed until an owner approves the collection and access: default deny. Role-based test accounts attempt to retrieve restricted material before go-live, and the results are recorded.

No. Information trained into a model has no permissions and cannot be deleted without retraining. Your model learns how your firm writes from material everyone may see. It looks up what your firm knows from permitted documents at question time, so nothing client-specific or restricted is trained into it.

Update or remove the file, and the next answer reflects that change without retraining. Facts live in your document index, not your model. A client’s data leaves when their documents do.

Your deployment record assigns three responsibilities. Your people are responsible for what they send, your firm for the configuration it approves, and PrivateStride for your system behaving as that record says.

Private AI in the Cloud runs for your company alone in a US data center, reached over a secured connection. Private AI in Your Building is installed in your building. Both run on NVIDIA and Apple hardware hosted in the United States. Nothing reaches a public AI provider. Cloud-hosted and in-building describe where your private AI will run after the assessment, not where the assessment happens.

No. Both assessments are done remotely, by video sessions with your leadership, phone interviews with up to five of your people, and a document review. The In-Building Private AI Assessment costs more because the report scopes hardware, network, security, and a 90-day program for your building. If you want one of us in your office in person, that is a separate engagement priced on request; it is not part of either assessment.

In a US data center we contract, or installed in your building. Companies and firms can choose either location.

If you want US data center hosting, choose the Cloud-Hosted Private AI Assessment, $4,500. If you want it installed in your building, choose the In-Building Private AI Assessment, $9,500. The fit call recommends which fits. Assessment pricing depends on deployment location. Both assessments are done remotely, by video and phone. Nobody travels.

Yes. Your private AI and everything it has learned about your business move with you. The In-Building Private AI Assessment covers the installation.

In-Building core local inference keeps working. Features that depend on an external identity provider, integration, scheduled update, or remote support may be unavailable until the connection returns. Cloud requires a secured connection to the US data center. We document your configuration’s dependencies before launch.

NVIDIA and Apple hardware hosted in the United States. Your proposal sets out equipment, support, updates, and refresh or return responsibilities. Cloud includes an equipment lease. In-Building uses leased hardware with upgrades. The arrangement is quoted in your assessment.

Your data stays on hardware dedicated to you. There is no per-seat meter. The signed agreement sets terms that cannot change on you. Your AI is trained on your business: your documents, your process, and your voice.

Yes. Cloud is built for entrepreneurs, founders, and executives in any industry. Companies and firms can choose Cloud-hosted or in-building based on where they want their AI to run.

Choose the Cloud-Hosted Private AI Assessment, $4,500. It takes two weeks, remotely, and produces a written report quoting implementation and monthly service. The assessment fee is credited in full toward implementation within 90 days. Both assessments are done remotely, by video and phone. Nobody travels.

Cloud assessment details

No. The assessment is the first step. It ends with a written report and a fixed quote.

No sandbox, no trial, and no demo. The assessment is the first step. You receive its written report. No AI system is delivered until implementation is paid.

Implementation and monthly service for Cloud, or installation and quarterly Managed AI for In-Building, are quoted in your assessment. No service price is set on this page.

The written report is the product. Two weeks of leadership interviews, a data-path map, three baselined workflows, and a capacity model on your numbers produce a report you keep. Cloud is $4,500, credited in full toward implementation within 90 days. In-Building is $9,500, credited in full toward installation within 90 days. Both assessments are done remotely, by video and phone. Nobody travels.

The report says so, with the reasoning, and you keep it. Some companies are better served by a policy and training. The assessment exists to identify what fits, including whether Cloud is a better choice.

Yes. Every implementation or installation is scoped from the assessment. The scope and three workflows use your baseline. For eligible professional firms on the In-Building track, the written time-recovery target also uses that baseline.

Only professional firms on the in-building track qualify: CPA, law, insurance, and wealth management firms of 10 to 75 professionals. We agree to a written 90-day time-recovery target. If the target is not met and participation requirements were satisfied, we keep working at no additional charge. Cloud has no guarantee. Signed terms control.

Professional firms program

In-Building Managed AI is a quarterly agreement. Cloud uses monthly service. Notice periods, service end dates, hardware return, and any remaining obligations are stated in your proposal before you sign.

On your documents, your process, and your voice, on your own private AI. It never trains anyone else’s model.

USA-sourced open-weight models, trained on your documents, your process, and your voice.

No internal IT department is required. We implement or install, monitor, update, and maintain the system within the agreed service scope. If you have an internal team or managed service provider, we coordinate responsibilities with them.

The assessment scopes the implementation or installation schedule. Physical In-Building deployment depends in-building and network readiness. The full 90-day Absorption Method, with champions and the dashboard, is the in-building firm program. Baseline, workflow redesign, training, and measurement turn installation into recovered time. Cloud includes setup, training on your documents, and monthly change work.

For the in-building firm program, the Absorption Method reviews adoption weekly, addresses friction during rollout, and uses champions to make approved workflows the normal workflows. Participation requirements and the agreed result target are written into the engagement. Cloud includes setup, training on your documents, and monthly change work.

For professional firms in the in-building program, adoption and workflow data are compared with the agreed baseline. Leadership then decides which workflows merit continued improvement or expansion. If an eligible firm’s written target is not met and participation requirements were satisfied, we keep working at no additional charge. Cloud has no guarantee. The proposal and signed terms control.

Measurement methodology

Professional firms program