Enterprise
Private, no-egress AI infrastructure deployed into your AWS account.
Frontier models inside a boundary with no route to the public internet.
Six independent isolation layers. Client-owned KMS. Full audit custody.
No standing access. Break-glass only, customer-approved, time-boxed.
Managed software deployed into your AWS account. Not a box. Not an appliance.
The landscape
It sounds like you are being asked to approve AI without a boundary you can defend. You are not alone. Here is what the numbers look like.
$200K+/mo
Opportunity cost of not using AI
100 seats at $100/hr, 5 hours saved per week
Banning AI does not stop your people from using it. It just stops you from knowing how. The alternative is approving something defensible.
The architecture
Every layer is independently enforced. Disabling one does not weaken the others. There is no accidental path out.
Defense in depth: to exfiltrate data, an attacker must defeat all six independent controls. Disabling one does not weaken the others.
1
Dedicated VPC with private subnets only. No internet gateway. No NAT gateway. No public IP addresses anywhere in the environment.
2
Stateless deny-all rules at the subnet level. Only explicitly allowed traffic between internal services passes through.
3
Stateful firewall rules on every resource. Tightly scoped ingress and egress rules per service. No default "allow all" posture.
4
No route to 0.0.0.0/0. Traffic can only reach internal endpoints and AWS services via VPC endpoints. There is no path out.
5
All AWS service communication stays on the AWS backbone via private endpoints. Bedrock, KMS, S3, CloudWatch. None of it touches the internet.
6
Organization-level guardrails that prevent accidental egress. Opening the boundary requires explicit org-admin action and leaves an audit trail.
Public AI vendors
"We won't look."
A policy. Enforced by policy and access controls inside a third-party boundary. Revocable at any time.
Totally Private AI
"We can't look."
An architectural constraint. Enforced by infrastructure. Your keys, your account, no path in for us.
Deployment
You own
AWS account in your organization. You pay AWS directly.
KMS keys generated in your account. Totally Private AI never has access to your key material in production.
VPC and networking. Private subnets, no internet gateway, your route tables.
Audit logs. CloudTrail, VPC Flow Logs, application audit trail. All in your account.
User access. VPN, Direct Connect, or PrivateLink. You decide how users reach the environment.
Identity and provisioning. SAML SSO, Okta, Azure AD, or your existing IdP. You control who gets in and how.
We manage
Deployment and updates. We install the platform and push updates through a controlled pipeline.
Application monitoring. Uptime, performance, and anomaly detection on the application layer.
Model access. Frontier models via Amazon Bedrock, routed through private VPC endpoints.
Support and training. Onboarding, user training, and ongoing technical support.
Custom integrations. API work, workflow automation, and legacy system connections as needed.
Break-glass and incident response. Customer-approved, time-boxed, fully audited. Covered in detail below.
Enterprise deployment moves the environment into your AWS account. During a Proof of Concept, we host in our account so you can go live in days. Same boundary principles and security posture. Enterprise adds customer-owned account, keys, and audit custody.
What changes
5+ hrs/week
Back per professional, measured
Drafting, summarizing, research, and analysis that used to take hours now takes minutes. That capacity goes straight back into billable work and client service.
Weeks
Go-live, not quarters
Not months. Not years. We deploy into your AWS account and your people are working with AI the same week. A DIY build takes 18 to 36 months and requires a dedicated engineering team.
500%+
ROI by Year 2
The productivity gains outpace the investment before the first quarter ends. By year two, you are operating at multiples of your deployment cost.
Zero
Internal platform build required
No platform team. No model hosting. No security review of a homegrown solution. We build, deploy, and maintain the infrastructure. Your engineers stay on their roadmap.
Access controls
1
Engineer submits a break-glass request with a documented reason and scope.
2
Multi-factor authentication before any temporary credential is issued.
3
Your designated contact must approve the access request before it is granted.
4
Credentials expire automatically. Every action is logged. A full audit record is available to you.
No one at Totally Private AI has standing access to your production environment. Break-glass only: customer-approved, time-boxed, fully audited. If we need in, you know about it, you approve it, and it expires automatically.
All chat data is encrypted per-user with your KMS keys. Software updates do not access the database. Your enterprise admins retain full access to organizational data through your own controls.
Evidence and compliance
We publish evidence packs for every region. Architecture diagrams, data-flow documentation, responsibility splits, and regulatory alignment summaries. Your compliance team reviews artifacts, not marketing claims.
ISO 27001
In progress
SOC 2 Type II
Planned
ISO 27701
Planned
ISO 42001
Planned
Architecture diagram and data-flow documentation
Boundary proof checklist: egress controls, encryption, key custody, logging
Regulatory alignment summary for your jurisdiction
Shared responsibility matrix (what you own vs. what we operate)
Pre-filled security questionnaire answers
The math
Without AI
$200K/month
In lost productivity alone. Your professionals spend 5+ hours per week on work AI completes in minutes. That time has a dollar value.
100 seats x $100/hr x 5 hrs/week x 4.3 weeks = ~$215K/month in recoverable capacity
With Totally Private AI
500%+ ROI by Year 2
Payback begins in the first month. Even with installation costs, the productivity gains outpace the investment before the first quarter ends.
Installation pays for itself in weeks, not months
DIY cost (Year 1)
$1.5M - $2.5M
Plus $3.87M in opportunity cost while your team builds instead of ships. 18 to 36 months before production. And a "server in a closet" does not scale past 100 seats.
TPAI Enterprise (Year 1)
Talk to us
Production-ready in weeks. No internal engineering team required. Your people use AI on day one, not year two.
Start with a conversation. We will walk you through the architecture, share the evidence pack for your region, and answer the hard questions your compliance team will ask.
Or email us directly at enterprise@totallyprivate.ai