
The counter to centralized AI. Built by a team of award-winning researchers, now powering the largest companies on earth. This is Nesa.

The World's AI Blockchain nesa.ai

The counter to centralized AI. Built by a team of award-winning researchers, now powering the largest companies on earth. This is Nesa.

Deloitte's State of AI 2026 survey of 3,235 leaders: 73% of enterprise leaders name data privacy and security as their top AI risk, and nearly three quarters plan to deploy autonomous agents within the next two years. Only 21% report mature governance for their agents, which almost always refers simply to containing them on-prem, thereby restricting their access and ability. The opportunity to bring the first fully private, verifiable, and yet free-roaming agent fleet to enterprise is massive. Every enterprise will need Nesa in this next chapter.

Agents running on private activity tracks, but that are fully trackable and auditable by their owners internally is the only future we can see here.
Menlo Ventures states AI agents have overtaken humans as the dominant consumers of AI. On OpenRouter, agents now generate 80% of all tokens: 130.9 trillion per week, roughly 7x human-driven usage. As agents handle more tasks and access more sensitive data on our behalf, privacy becomes even more important. Nesa’s EE technology tackles this by keeping data fully private throughout inference! menlovc.com/perspective/2026…

Today, an onchain app that uses AI usually calls a model API off-chain and posts the answer to the chain. The chain sees the result and has no record of how it was produced. On an AI Layer-1, the request is settled on chain and the result comes back with a cryptographic proof that it was computed correctly. While the query and output stay unseen 🔒 For a DeFi app assessing risk with a model, that is the difference between trusting a vendor's answer and confidently knowing the high-impact answer is true.

32% of AI users say they have told a chatbot something they kept from close friends, parents or doctors. In the same DuckDuckGo survey of 1,944 US adults, only 25% knew that AI chats can be subpoenaed. People use chatbots as confidants. Most major chatbots train on conversations by default and tie them to an account. What you ask an AI is some of the most sensitive data you produce, and today it sits readable on someone else's servers. Encrypting the query and the output at the point of inference means the provider never holds the readable version. That's exactly what we are building here.

Every onchain app that adds AI has a gap in the middle of it. A contract's logic is public. A wallet's balance is public. But the model call? That's a request to a server the chain cannot see. Nothing proves: ⬟ Which model answered ⬟ Who read the data ⬟ What happened during inference And this matters more as AI starts taking real actions. Agents that move user funds, models that score risk, and assistants that act for users all depend on that call. The most consequential decision in the app runs outside the guarantees of the chain it lives on 👀 Closing that gap takes a Layer-1 where inference is settled and verified onchain, with the data encrypted while the model runs. It is what we are building at Nesa.

Over the next two years this will be one of the biggest topics for the industry. Build with Nesa.
Demis Hassabis is warning about the next wave of AI-driven cyberattacks. The industry is spending billions making models more capable while security architecture continues to play catch-up. Nesa’s technology ensures that secure computation can advance alongside model intelligence, not years after it.

Privacy usually costs speed, which is the usual reason it gets skipped. Fully homomorphic encryption keeps data hidden during computation, at a heavy performance cost. Our Equivariant Encryption paper takes another route. 📍 Inputs, outputs and intermediate activations stay unreadable to the server running the model, and the paper reports near-zero extra latency compared with unencrypted inference. The paper also includes a threat analysis that lays out the attack vectors, so anyone can test the claims.

Nvidia agreed on September 3 to acquire Hugging Face, one of the most widely used hubs for open models. Nvidia says the aim is to protect open-source AI developers. The dominant GPU vendor will now also own one of the main places open models are published. For the person using those models, little changes. With every inference-as-a-service player today, you still cannot check the proof for: → Which model answered a request → Who could read your prompt → What happens to your data You trust whoever sits in the middle. Open models only stay open if anyone can verify how they run, whoever owns the hub. That takes an execution layer with no owner, where the network checks which model ran and keeps the data private.

The official NES migration portal is now live for eligible on-chain holders. See the article on the official Nesa website below: nesa.ai/article/d8s9uv30 Please use only links shared through official Nesa channels.





Where you start building on Nesa depends on what you already have: 1. No model chosen yet? → Use the Playground on our website 2. Your own model? → Use custom model upload 3. An app that uses the ChatGPT format? → Use the ChatGPT-compatible API 4. A TypeScript project? → Use the SDK Learn more below 👇

Where you start building on Nesa depends on what you already have: 1. No model chosen yet? → Use the Playground on our website 2. Your own model? → Use custom model upload 3. An app that uses the ChatGPT format? → Use the ChatGPT-compatible API 4. A TypeScript project? → Use the SDK Learn more below 👇 --- 2/ 📍 The Playground Browse the on-chain model catalog by modality and task, compare each model's latency and usage, and try models before you commit to one. The catalog spans text, image, video, audio, 3D, bio and multi-modal models. 3/ 📍 Custom model upload Register your own model through the Build on Nesa flow with a Docker image or GitHub repo Plus its metadata and visibility settings, and it runs on the network. Maintaining more models requires Nesa Pro, with up to 5 active custom models. 4/ 📍 The ChatGPT-compatible API. If your app already uses the ChatGPT format, moving it to Nesa is a one-line change. The query and output stay unseen on every request. 5/ 📍 The TypeScript SDK on GitHub (nesaorg/ai-sdk-ts). It manages AI sessions and sends inference requests to over 100,000 models In the browser or in Node. Install it with npm install @nesaorg/ai-sdk.