# VoltageGPU Documentation > Confidential AI infrastructure. Intel TDX trust domains, NVIDIA H200 GPU attestation > generated by the tenant, OpenAI-compatible TEE inference. Company: VOLTAGE EI, France. Main site: https://voltagegpu.com (llms.txt: https://voltagegpu.com/llms.txt) ## Pages - Getting started: https://docs.voltagegpu.com/getting-started - Volt API reference: https://docs.voltagegpu.com/api-reference - Standard pods by API (no enclave: list machines, deploy, poll, release): https://docs.voltagegpu.com/api-reference#standard-pods - Pods overview: https://docs.voltagegpu.com/pods - Create a pod (dashboard and API): https://docs.voltagegpu.com/pods/create - Standard GPU pods (no enclave, for data that is not sensitive): https://docs.voltagegpu.com/pods/standard-gpu - Confidential Compute (container tier, Intel TDX, GPU confidential mode OFF): https://docs.voltagegpu.com/pods/confidential-compute - Confidential VMs (full Intel TDX VM, tenant attestation): https://docs.voltagegpu.com/pods/confidential-vm - Templates: https://docs.voltagegpu.com/pods/templates - Volumes (container pods only, Confidential VMs have no persistent volume yet): https://docs.voltagegpu.com/pods/volumes - Backups: https://docs.voltagegpu.com/pods/backups - Restores: https://docs.voltagegpu.com/pods/restores - Scheduled termination: https://docs.voltagegpu.com/pods/scheduled-termination - Inference (OpenAI-compatible, TEE models): https://docs.voltagegpu.com/inference (busy models retried with backoff; optional "fallback_models"; X-VoltageGPU-Model tells which model answered and was billed; GET /v1/models has a live status per model; GET /v1/attestation?model=&nonce= returns the instances' Intel TDX quotes and NVIDIA GPU evidence bound to your nonce: https://docs.voltagegpu.com/inference#attestation) - CLI reference: https://docs.voltagegpu.com/cli ## Standard pods through the API (since 8 October 2026) Same X-API-Key and balance as the confidential tier. No memory encryption, no TDX quote, no GPU attestation. GET https://api.voltagegpu.com/api/volt/machines?provider=standard (machine_id, price_per_hour charged for the whole machine) POST https://api.voltagegpu.com/api/volt/pods {"provider": "standard", "name": "my-pod", "machine_id": ""} (provider is required) GET https://api.voltagegpu.com/api/volt/pods/ (poll until ssh_command is set) DELETE https://api.voltagegpu.com/api/volt/pods/ (unused part of the first hour refunded) Your own image: add "image" (public, linux/amd64), optionally "start_command" (one simple command), "ports" (up to 16) and "env" (up to 32 string values) to the POST body. An image that cannot be prepared answers 400 and nothing is charged. Check the machine: curl -fsSL https://voltagegpu.com/hwcheck.sh | bash -s -- --gpu "" --gpus (read-only, compares the machine with its listing, exit 0 = matches, 2 = does not). Agent checklist for a standard pod (every call takes the X-API-Key header): 1. GET /api/volt/balance: the balance must cover the first hour of the chosen machine. 2. GET /api/volt/ssh-keys: pass one as "ssh_key_id" so the pod is reachable. 3. GET /api/volt/machines?provider=standard: pick "machine_id". 4. Either "template_id" from GET /api/volt/templates?provider=standard, or your own "image" (not both). 5. Optional: "gpu_count", "auto_terminate_hours" (whole hours, 1 to 720; for a shorter run call DELETE yourself), "start_command" as ONE command: ; & | < > backticks, $( and line breaks are refused with a 400 and nothing is charged; put multi-step setup in a script inside the image. 6. POST /api/volt/pods, poll GET /api/volt/pods/ until ssh_command is set, DELETE when done. Billing: per second from a prepaid balance, the first hour charged at deploy. A pod the provider never manages to start gets the full prepaid hour back. ## Publish your own attestation Drop the bundle.json written by voltage-verify on https://voltagegpu.com/verify/share: the checks run again up to the Intel and NVIDIA roots, and you can publish an unlisted page with a badge, removable with a private link. ## Confidential VM guide, sections worth linking directly - Generate your own Intel TDX quote via configfs TSM, verify offline with Intel DCAP: https://docs.voltagegpu.com/pods/confidential-vm - NVIDIA GPU confidential-computing mode: ON on single-GPU H200 VMs (verified 4 Sept 2026), single-GPU H100 VMs (verified 16 Sept 2026) and single-GPU RTX PRO 6000 Blackwell VMs (verified 17 Sept 2026), all with NVIDIA remote attestation successful on a caller nonce; 8x H100 nodes attest per GPU in Protected PCIe mode (10 Sept 2026, NVSwitch not attested); the 8x H200 node is not attested yet; OFF on the container tier; B200 VMs unverified. Published evidence, both proofs: single-GPU H200 (4 Sept 2026), the 8x H100 node (voltage-verify bundle of 12 Sept 2026) and single-GPU RTX PRO 6000 Blackwell (bundles of 17 Sept and 9 Oct 2026). The single-GPU H100 was verified on 16 Sept 2026, but its bundle is not published. - Two GPU reports are never byte-identical (SPDM responder nonce, measured offsets): https://docs.voltagegpu.com/pods/confidential-vm#report-stability - Raw artifacts you can verify without renting a VM (NVIDIA-signed token, GPU report + nonce, TDX quote + report_data, SHA-256 sums): https://docs.voltagegpu.com/pods/confidential-vm#artifacts - Concurrent TDX quotes can fail with EINVAL from configfs TSM, retry on EINVAL only: https://docs.voltagegpu.com/pods/confidential-vm#concurrent-quotes ## Related on the main site - Article "Two Proofs" (TDX quote + NVIDIA H200 attestation, tenant-side, real outputs): https://voltagegpu.com/blog/two-proofs-tdx-nvidia-h200-attestation-tenant - Artifacts README: https://voltagegpu.com/blog/two-proofs/evidence/README.txt - voltage-verify, open-source CLI that binds both proofs to a workload manifest and verifies the bundle offline (pip install "voltage-verify>=0.2.1", https://github.com/VoltageGPU/voltage-verify): https://voltagegpu.com/blog/two-proofs/voltage-verify - Tenant TDX attestation guide (blog): https://voltagegpu.com/blog/tenant-tdx-attestation-guide - Verify a confidential GPU yourself, five steps (pillar page): https://voltagegpu.com/confidential-computing/tenant-attestation - Confidential GPU clouds compared (VoltageGPU, Azure, Google Cloud, AWS), verified 12 Sept 2026: https://voltagegpu.com/compare/confidential-gpu-clouds - Self-service Confidential VM deploy: https://voltagegpu.com/confidential-vm/deploy - Trust center: https://voltagegpu.com/trust