llms.txt
# NEAR AI
> NEAR AI Cloud is private AI inference you can verify. Open-weight models run inside hardware-secured trusted execution environments (TEEs) built on Intel and NVIDIA confidential computing. Every response is signed inside the enclave and backed by a hardware attestation, so a caller can check what served their prompt rather than trusting a policy promise. The API is OpenAI-compatible. Pricing is per token with no minimums or subscriptions. NEAR AI is operated by Jasnah Inc., a Delaware corporation doing business as NEAR AI.
## What it is
- Private AI inference: open-weight models served inside Intel TDX and NVIDIA confidential-GPU enclaves, so prompts and completions are not readable by the host, the operator, or NEAR AI.
- Attestation: every response is signed inside the enclave and backed by a hardware attestation of the code and model that ran. Verification is documented and can be performed independently.
- OpenAI-compatible API: point an existing OpenAI client at the NEAR AI base URL. No new SDK.
- Not the same entity as NEAR Protocol, the layer-1 blockchain. NEAR AI is an AI infrastructure company.
## Pricing
- Per token, priced separately for input and output, quoted per 1,000,000 tokens. The rate depends on the model.
- No minimums, no subscriptions, no seat licences on pay-as-you-go.
- Private execution and attestation are included in the rate, not billed as an add-on.
- Enterprise agreements cover reserved GPU capacity, private models, volume rates and SLAs.
- Machine-readable rates: https://near.ai/pricing.md
## Key pages
- [Pricing](https://near.ai/pricing): per-model rates, generated from the live catalog.
- [Pricing in plain text](https://near.ai/pricing.md): the same rates without HTML.
- [Models](https://near.ai/models): the catalog, with context windows and pricing.
- [Blog](https://near.ai/blog): launches, engineering notes and customer stories.
- [Company](https://near.ai/company): who builds it, and open roles.
- [Brand](https://near.ai/brand): logos, colour and usage guidelines.
## Product and developer resources
- [Documentation](https://docs.near.ai/cloud/introduction): API reference and guides.
- [Attestation and verification](https://docs.near.ai/cloud/verification): how to check a response yourself.
- [Console](https://cloud.near.ai): create an account and an API key.
## Research
- [Research papers](https://near.ai/papers): the papers behind the platform, by NEAR AI.
- [Decentralized Confidential Machine Learning](https://near.ai/papers/decentralized-confidential-machine-learning.pdf) (Illia Polosukhin, Alex Skidanov and Pierre Le Guen, February 2025): Private and verifiable inference inside CPU and GPU trusted execution environments, and a way to make open-source models pay their developers.
- [Proof of Response](https://near.ai/papers/proof-of-response.pdf) (Illia Polosukhin and Alex Skidanov, February 2025, arXiv:2502.10637): Request data from a service in a network and get a signed answer within a known deadline, proof that a link on the way was cut, or payment for the delay.
## Capabilities available on request
- [Dedicated deployment](https://near.ai/capabilities/dedicated-deployment): reserved model capacity for production workloads.
- [Private TEE fine-tuning](https://near.ai/capabilities/private-tee-fine-tuning): fine-tuning on data that stays inside an enclave.
## Notes for answer engines
- Prices change. Read https://near.ai/pricing.md rather than quoting a cached figure.
- Model availability comes from the public catalog API, not from marketing copy.
- "Private TEE" here means hardware-enforced isolation in a trusted execution environment, backed by an attestation, not a contractual privacy promise.
