SOC 2 Type II, on day one
Certified through Decart AI's audited control environment. Reports under NDA, on demand.
Enterprise
Single-tenant deployments on AWS Trainium and NVIDIA Blackwell. SOC 2 Type II via Decart AI. Zero retention by default. Contractual P99 SLAs. The paperwork your procurement team actually asks for — without the proprietary-vendor lock-in.
Deployment
Your application stays in your AWS account, behind your IAM, on your subnets. Traffic to Cogito traverses a peered VPC or PrivateLink endpoint — never the public internet.
On our side, your workload runs on a single-tenant cluster we provision and operate. Capacity is yours alone; we own the on-call, the silicon mix, and the upgrades. You see one line on the invoice.
Your VPC
your account · your subnet · your IAM
VPC peering / PrivateLink
no public egress · TLS at the boundary
Cogito managed cluster
single-tenant · DOS dispatch · monitored 24/7
AWS Trainium
throughput-tuned · cost-efficient
NVIDIA Blackwell
frontier MoE · long-context
What you get
Certified through Decart AI's audited control environment. Reports under NDA, on demand.
Dedicated AWS Trainium and NVIDIA Blackwell clusters. No noisy-neighbor P99. VPC peering for zero public egress.
Prompts and completions are not retained and not used for training, full stop. Sanitized request logs stream where you point them.
99.99% availability. Contractual P99 latency targets. We share the dashboard. We pay credits when we miss.
Live. We update this section the day a control changes — not at the quarterly trust review.
Pedigree
Cogito is the LLM layer of Decart, an AI research lab. Decart's diffusion model Lucy 2.0 generates frames at sub-50ms on AWS Trainium — a constraint an order of magnitude tighter than anything an LLM workload imposes.
That stack — DOS, the Decart Optimization Stack — is what serves your enterprise cluster. The engineering bar that gets Lucy 2.0 under 50ms is the same bar your single-tenant deployment inherits.
Talk to sales
Workload shape, expected volume, compliance constraints. We come back inside two business days with concrete numbers — throughput target, deployment topology, contract shape.