Skip to content

Models: describe the three privacy tiers used in the catalog #61

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
alejoras merged 1 commit into from
Oct 5, 2026
Merged

Models: describe the three privacy tiers used in the catalog #61

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion cloud/inference-sdk/python.mdx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,7 +48,7 @@ The client connects to `https://cloud-api.near.ai/v1` by default. Reuse a client

End-to-end encryption is disabled by default. `e2ee=True` encrypts supported Chat fields to a verified model key and decrypts the response automatically.

In Python, model attestation requires catalog metadata with `providerType="vllm"` and `attestationSupported=True`. Other providers, including Chutes, use Incognito mode: Gateway verification only. E2EE and model deployment policies reject Incognito models before sending Chat. Use `e2ee=False` for Gateway-only Chat without a claim about model TEE execution.
In Python, model attestation requires catalog metadata with `providerType="vllm"` and `attestationSupported=True`. Other models, including Chutes-hosted [3P Confidential TEE](/cloud/models) models, are verified at the Gateway only. The SDK calls this Incognito mode, which is separate from the Incognito privacy tier in the model catalog. E2EE and model deployment policies reject these models before sending Chat. Use `e2ee=False` for Gateway-only Chat without a claim about model TEE execution.

## Verification

Expand Down
7 changes: 4 additions & 3 deletions cloud/models.mdx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,10 +6,11 @@ icon: "brain-circuit"

NEAR AI Cloud provides access to leading AI models, each optimized for different use cases ranging from advanced reasoning and tool calling to long-context processing, embeddings, reranking, image generation, and audio transcription — all with transparent, pay-per-use pricing.

The catalog contains two kinds of models:
The [model catalog](https://near.ai/models) labels every model with one of three privacy tiers:

- **TEE-hosted models** (marked <span className="doc-badge">TEE</span>) run on NEAR AI's own GPU fleet inside [Trusted Execution Environments](/cloud/private-inference). They support [attestation, signatures, and verification](/cloud/verification), and nobody — not even NEAR — can see your prompts or outputs.
- **Third-party models** (OpenAI, Anthropic, Gemini, and others) are proxied to their providers through the NEAR AI gateway for convenience. They offer the same unified API and billing, but the TEE privacy and verifiability guarantees do not extend to the upstream provider.
- **Confidential TEE** (marked <span className="doc-badge">Confidential TEE</span>) models run on NEAR AI's own GPU fleet inside [Trusted Execution Environments](/cloud/private-inference). They support [attestation, signatures, and verification](/cloud/verification), and nobody, not even NEAR, can see your prompts or outputs.
- **3P Confidential TEE** models run inside a third-party provider's Trusted Execution Environment. Their provider is listed as **Attested 3P**. The request path to the provider's enclave is encrypted, and the provider publishes its own [attestation evidence](/cloud/verification/cloud-api/model-attestations). Model-level verification depends on the client: the JavaScript SDK verifies Chutes model attestation, while the Python SDK verifies the Gateway only.
- **Incognito** models (OpenAI, Anthropic, Gemini, and others) are routed through the NEAR AI Gateway, which runs in a TEE, and sent to the provider under NEAR AI's provider account. Your NEAR AI credentials are not forwarded to the provider. The model itself runs on the provider's infrastructure, outside a TEE, so the TEE privacy and verifiability guarantees do not extend to the upstream provider.

<div className="doc-lead">
<a className="doc-button" href="https://near.ai/models" target="_blank" rel="noreferrer">Browse Model Catalog →</a>
Expand Down
Loading