{ "id": "12", "slug": "confidentialai", "name": "Confidential AI", "tagline": "Confidential inference on Intel TDX with per-workload attestation receipts", "description": "Confidential AI (Inexorable, Inc.) serves open-weight models inside Intel TDX confidential VMs on its open-source c8s platform, with NVIDIA GPUs in confidential-compute mode. The public /attestation endpoint returns fresh nonce-bound c8s receipts for every workload in the serving cluster — gateway, router, inference workers, and supporting services — including Intel TDX quotes, pinned image measurements, mesh identity proofs, and NVIDIA GPU evidence. A post-quantum over-encryption tunnel (ML-KEM-768 + X25519) terminates inside the attested TEE.", "logo_url": null, "website_url": "https://confidential.ai", "api_docs_url": "https://confidential.ai/docs/inference-api/get-started", "founded_year": null, "headquarters": null, "status": "active", "tee_types": [ { "type": "intel_tdx", "attestation_method": "c8s attestation bundle: per-workload Intel TDX quotes whose report_data binds a SHA-384 transcript over the caller nonce, session keys, and mesh identity", "notes": "GET https://api.confidential.ai/attestation?nonce=<32-byte b64url> returns c8s/attest-pq/v1 receipts for every cluster workload plus a c8s/attest-lb/v1 front-door receipt that binds the live TLS serving certificate." }, { "type": "nvidia_h100_cc", "attestation_method": "NVIDIA GPU evidence relayed inside the cluster attestation bundle (nvidia_gpu per inference worker, gpuEvidence aggregate)", "notes": "GPU evidence is collected and relayed (Blackwell-class devices); independent GPU verdict is not yet produced by this directory's verifier." } ], "models": [ { "name": "DeepSeek V4 Flash", "slug": "deepseek-v4-flash", "source_id": "deepseek-ai/DeepSeek-V4-Flash-0731", "family": "DeepSeek", "context_window": 1048576, "max_output_tokens": null }, { "name": "MiniMax M3 (MXFP8)", "slug": "minimax-m3-mxfp8", "source_id": "MiniMaxAI/MiniMax-M3-MXFP8", "family": "MiniMax", "context_window": 1048576, "max_output_tokens": null } ], "pricing": [ { "model_name": "DeepSeek V4 Flash", "price_per_million_input_tokens": 0.2, "price_per_million_output_tokens": 0.4, "price_per_hour": null, "pricing_model": "per_token", "free_tier": false, "notes": "Cached input $0.018 per 1M tokens. Capacity-available tier per public pricing page." }, { "model_name": "MiniMax M3 (MXFP8)", "price_per_million_input_tokens": null, "price_per_million_output_tokens": null, "price_per_hour": null, "pricing_model": "custom", "free_tier": false, "notes": "Bespoke deployment — no public per-token price; contact hello@confidential.ai." } ], "api_features": { "openai_compatible": true, "streaming": true, "function_calling": false, "batch_inference": false, "vision": false, "embeddings": false, "max_concurrent_requests": null, "avg_latency_ms": null }, "payments": { "accepts_crypto": false, "accepts_bitcoin": false, "accepted_assets": [], "notes": "Monthly invoicing by card or bank transfer." }, "attestation": { "documentation_url": "https://confidential.ai/docs/inference-api/attestation", "endpoint_url": "https://api.confidential.ai/attestation", "verifier_url": "https://github.com/confidential-dot-ai/c8s-verify-js", "evidence_format": "c8s/attest-pq/v1 per-workload receipts and c8s/attest-lb/v1 front-door receipt (Intel TDX report_data binding a SHA-384 identity transcript; ECDSA P-384 mesh proof; cluster allowlist and OCI digests)", "notes": "Live cluster demo: https://confidential-inference-cluster-demo.confidential.ai/. The transcript construction is specified in c8s PROTOCOL.md and the deployed session-pubkey shape was confirmed against live report_data." }, "sdks": [ { "language": "TypeScript", "package_url": "https://github.com/confidential-dot-ai/c8s-verify-js", "docs_url": "https://confidential.ai/docs/inference-api/attestation" }, { "language": "Go", "package_url": "https://github.com/confidential-dot-ai/c8s", "docs_url": "https://confidential.ai/docs/c8s" } ], "api_compatibility": { "openai_chat_completions": true, "openai_responses": false, "anthropic_messages": false, "notes": "OpenAI-compatible /v1/chat/completions and /v1/completions. API keys are provisioned by request (email or request-access form); /health, /attestation, and the c8s well-known endpoints are public." }, "social_links": [], "source_urls": [ { "label": "Public pricing", "url": "https://confidential.ai/pricing", "checked_at": "2026-09-21" }, { "label": "Inference API attestation docs", "url": "https://confidential.ai/docs/inference-api/attestation", "checked_at": "2026-09-21" }, { "label": "Live cluster attestation demo", "url": "https://confidential-inference-cluster-demo.confidential.ai/", "checked_at": "2026-09-21" }, { "label": "OpenAI-compatible model catalog", "url": "https://api.confidential.ai/v1/models", "checked_at": "2026-09-21" } ], "last_verified_at": "2026-09-21", "data_freshness": { "status": "fresh", "last_success_at": "2026-09-21T16:23:14Z", "last_attempt_at": "2026-09-21T16:23:14Z" }, "organization_roles": [ "storefront", "workload-operator", "tee-platform" ], "trust_tier": "hw-verified-tls", "pqc": { "status": "pq-attested-channel", "kex": "X25519MLKEM768", "kex_required": false, "notes": "Wire probe (openssl s_client, 2026-09-21): api.confidential.ai TLS 1.3 refuses all standardized hybrid groups and negotiates classical X25519 only. Post-quantum protection lives one layer up: the c8s over-encryption tunnel (POST /.well-known/c8s/tunnel) performs X-Wing ML-KEM-768+X25519 key agreement whose transcript is bound into the TDX report_data, so PQ confidentiality terminates inside the attested TEE for clients that use it. Cluster identity proof is classical ECDSA P-384 — PQ confidentiality, not PQ authentication.", "source_url": "https://confidential.ai/docs/inference-api/attestation", "checked_at": "2026-09-21" } }