near-ai-cloud

NEAR AI Cloud

Verifiable private AI inference through Trusted Execution Environments (TEEs). All inference runs inside Intel TDX confidential VMs with NVIDIA TEE GPUs — your data stays encrypted and isolated from infrastructure providers, model providers, and NEAR itself.

Quick Start

The API is OpenAI-compatible. Point any OpenAI SDK at https://cloud-api.near.ai/v1:

import openai

client = openai.OpenAI(
    base_url="https://cloud-api.near.ai/v1",
    api_key="YOUR_API_KEY"  # from cloud.near.ai dashboard
)

response = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-V3.1",
    messages=[{"role": "user", "content": "Hello, NEAR AI!"}]
)
print(response.choices[0].message.content)

import OpenAI from 'openai';

const openai = new OpenAI({
    baseURL: 'https://cloud-api.near.ai/v1',
    apiKey: 'YOUR_API_KEY',
});

const completion = await openai.chat.completions.create({
    model: 'deepseek-ai/DeepSeek-V3.1',
    messages: [{ role: 'user', content: 'Hello, NEAR AI!' }]
});
console.log(completion.choices[0].message.content);

How It Works

• All inference runs inside Intel TDX confidential VMs with NVIDIA TEE GPUs
• TLS terminates inside the TEE, not at a load balancer — prompts are never exposed in plaintext
• TEEs generate cryptographic attestation proofs verifiable via NVIDIA NRAS and Intel TDX
• Every chat response is signed by a key that never leaves the TEE
• You can independently verify hardware attestation and bind it to message signatures

Verification Flow

1. Generate nonce
2. Request model attestation  →  get signing_address, nvidia_payload, intel_quote
3. Verify GPU attestation     →  submit nvidia_payload to NVIDIA NRAS, check JWT fields
4. Verify CPU attestation     →  verify intel_quote via dcap-qvl or TEE Explorer
5. Verify GPU-CPU binding     →  signing_address + nonce bound in TDX report data; same nonce in NRAS eat_nonce
6. Make chat request           →  use the API as normal
7. Fetch chat signature       →  GET /v1/signature/{chat_id}
8. Verify signature            →  recover signer, compare to attested signing_address

API Endpoints

Base URL: https://cloud-api.near.ai

Endpoint	Method	Description
/v1/chat/completions	POST	OpenAI-compatible chat completions
/v1/models	GET	List available models
/v1/attestation/report?model={model}	GET	Model attestation (GPU + CPU) — requires API key
/v1/attestation/report	GET	Gateway attestation — requires API key
/v1/signature/{chat_id}	GET	Chat message signature (also covers Responses API resp_*)

Critical Knowledge

• Base URL is https://cloud-api.near.ai/v1 — use with any OpenAI SDK
• signing_algo can be ecdsa or ed25519
• Attestation report now requires an API key — Authorization: Bearer <key>; retrieval is free and non-billable
• Nonce should be a random 64-char hex string (32 bytes) for attestation freshness
• NRAS response is a two-part array: [["JWT", "..."], {"GPU-0": "..."}] — overall JWT + per-GPU JWTs
• The signing_address from model attestation must match the address that signed chat messages
• Chat signatures are persistent and can be queried at any time after completion

References

Topic	File
Private vs Anonymised Models	references/private-vs-anonymised.md
Model TEE verification	references/model-verification.md
Gateway verification + TLS	references/gateway-verification.md
Chat message verification	references/chat-verification.md
E2E encrypted chat (E2EE)	references/e2ee-chat.md
Specialized endpoints	references/specialized-endpoints.md

Upstream Docs (prefer live values)

• OpenAI compatibility — streaming, reasoning, Files API behavior
• Stateless Responses API — resp_* stateless responses
• Prompt caching
• Web search — server-side live web results
• Staking for inference — pay with staking rewards
• Usage reporting API
• Model discovery & refresh

Resources

• NEAR AI Cloud: https://cloud.near.ai
• Documentation: https://docs.near.ai/cloud/introduction- Verification Example: https://github.com/near-examples/nearai-cloud-verification-example
• Full Verifier: https://github.com/nearai/nearai-cloud-verifier
• NVIDIA NRAS API: https://docs.api.nvidia.com/attestation/reference/attestmultigpu_1
• TEE Attestation Explorer: https://proof.t16z.com/
• DCAP QVL (TDX verification): https://github.com/Phala-Network/dcap-qvl