cedana / product · security

Confidential inference without the performance tax.

The compute multiplier, now inside confidential VMs.
Detect weight tampering in real time.
Why Cedana

Security without compromise.

Confidential compute that runs at higher performance, not lower.
Cedana’s compute multiplier delivers more tokens per GPU-hour than native performance.

Depth

Continuous run-time monitoring across the stack: CPU/GPU, kernel, CUDA, Kubernetes and Slurm.

Scale

Supports single GPU, multi-GPU and multi-node. Proven at fleet scale.

Breadth

Works transparently across all models, engines, NVIDIA Dynamo, llm-d, and whatever is in your inference stack.

What you get

Capabilities

confidential vm · trust boundary● attested
trust boundary
Your stackvLLM · SGLang · TensorRT-LLM · Dynamo · llm-d · K8s · Slurm
Cedanacheckpoint · migrate · resume · detect tampering
GPUhardware attested
  1. 01checkpointinside the trust boundary● sealed
  2. 02encryptwith your keys, tamper-evident● signed
  3. 03migrateboundary to boundary● attested
  4. 04resumesame state, verified GPU● ready
keys · yoursweights · watched in real timecheckpoints · encrypted, tamper-evidentstack · yours or ours

Checkpoint, migrate, and resume

Inside trust boundaries.

GPU hardware attestation.

Encrypted, tamper-evident checkpoints.

Your keys.

Real-time detection of weight tampering.

Full-stack inference out of the box.

Or use your existing stack.

Talk to us.

Confidential inference without the performance tax. Bring one workload and we will show you the compute multiplier running inside your confidential VMs.