Nillion comprehensive research report: the leader in blind computing across the two major tracks of AI+privacy

Author: Messari

Compiled by: Azuma, Odaily Planet Daily

Editor's note: Earlier this week, the market was circulating a list of TGE's hot projects expected to be launched in the first quarter of this year, and Nillion, a privacy computing leader that has raised $50 million, was among them.

In the following, investment research agency Messari provides a detailed analysis of Nillion through multiple levels such as team, narrative, technology, architecture, token, ecology, roadmap, etc., which may help you further understand the information and dynamics of the project.

The following is the full text of Messari, compiled by Odaily Planet Daily.

Overview of core content

• Nillion has partnered with companies/projects such as Virtuals, NEAR, Aptos, Arbitrum, Ritual, io.net and Meta.
• A comprehensive set of application tools, including nilAI, nilVM, nilDB, and nilChain, provides developers with resources to create privacy-preserving applications in areas such as artificial intelligence, healthcare, and DeFi.
• Nillion utilizes privacy-enhancing technologies (PETs) such as multi-party computation (MPC), homomorphic encryption (Homomorphic Encryption) and zero-knowledge proofs (Zero-Knowledge Proofs) for coordination to achieve its decentralized infrastructure. Secure data computing and storage.
• Nillion's validator program has approximately 500,000 validators, which have processed approximately 195 million ciphertexts and secured approximately 1,050 GB of data.

Preface

Processing high-value data (such as passwords, personalized AI, healthcare information, biometric information) has historically been unsafe and inefficient. Although encryption technology can ensure the security of stored data, it needs to be decrypted during calculation and re-encrypted after decryption, which brings loopholes and delays. Although blockchain technology can decentralize transactions and data management, it does not essentially solve the security computing problem of encrypted data. This limitation limits the types of applications that can be safely built in Web3.

Nillion hopes to address these limitations by transmitting, storing, and computing data without decryption, thus ensuring that sensitive information remains private and secure throughout its lifecycle. This approach, called Blind Compute, decentralizes trust and expands the use cases of decentralized networks into previously unexplored white space, such as private AI agents, private LLM inference, and others. Industries that require secure data. By using advanced privacy technologies (PET) such as multi-party computation (MPC), fully homomorphic encryption (FHE) and trusted execution environment (TEE), Nillion allows data to remain encrypted throughout the entire computing process.

background

Founded in 2021, Nillion is a project that provides a novel way to handle private data in distributed systems without compromising security or efficiency. Supported by application frameworks such as nilVM, nilDB, nilAI and nilChain, Nillion provides developers with tools to help them build privacy-friendly applications in areas such as artificial intelligence, DeFi and data storage.

Nillion team members include:

• Alex Page (CEO), former Hedera SPV general partner and Goldman Sachs banker;
• Andrew Masanto (CSO), co-founder of Hedera and founding CMO of Reserve;
• Slava Rubin (CBO), founder of Indiegogo;
• Dr. Miguel de Vega (Chief Scientist), PhD supervisor and author of more than 30 patents.
• Conrad Whelan (founding CTO), founding engineer of Uber;
• Mark McDermott (COO), former head of innovation at Nike;
• Andrew Yeoh (CMO), early senior partner at Hedera, former UBS and Rothschild banker, etc.

Since its founding, the team has raised $50 million in private financing from investors including Hack VC, Hashkey Capital, Distributed Global and Maelstrom.

technology

The Nillion Network is a decentralized infrastructure designed to process high-value data in a secure and private manner.

Nillion consists of two core layers: (i) Coordination Layer, responsible for management and payment; (ii) Orchestration Layer (Petnet), responsible for processing computing and storage. Nillion’s multi-party computation (MPC) protocol is at the heart of the network’s capabilities, enabling private data computation without revealing individual inputs. Nillion's ecosystem is powered by a comprehensive set of application tools (namely nilAI, nilVM, nilDB and nilChain) that help developers build privacy-friendly applications. Academic research papers in cryptography and privacy technology have significantly verified Nillion's technical feasibility.

Nillion Network

Nillion Network is a decentralized infrastructure designed to support private high-value data storage and computation. Nillion Network's scalability is achieved through clusters, which configure groups of nodes to meet specific performance, security and cost requirements. Unlike traditional blockchains, Nillion Network operates without relying on global shared state, enabling vertical scalability (by upgrading a single node or cluster) and horizontal scalability (by adding new nodes or clusters), thereby efficiently allocating workload. The following are the contributions of each layer (i.e. coordination layer and orchestration layer) to the network architecture.

Coordination Layer

The coordination layer of the Nillion network (nilChain for short) is responsible for: (i) managing rewards; (ii) payments; (iii) cryptoeconomic security; (iv) coordination among network clusters.

Specifically, nilChain is responsible for coordinating payments for storage operations and blind computations performed on the network, without handling the computations directly. The coordination layer is built using the Cosmos SDK and supports IBC for interoperability; however, given that the core focus of the network is storage and computation, it does not currently support the execution of smart contracts. While directly accessible via Keplr or Leap wallets, applications built on collaborative blockchains (explored further in the Key Projects section) will be completely abstracted. nilChain has been running on the testnet in June 2024.

Orchestration layer (Petnet)

Petnet aims to integrate encryption technologies such as multi-party computation (MPC), fully homomorphic encryption (FHE) and zero-knowledge proofs (ZKPs) to enable private computing and data management. This integration is achieved through two key components: (i) the compiler and (ii) the computational network. Specifically, compilers simplify the use of privacy-enhancing technologies (PETs) by providing different levels of abstraction, while computational networks perform secure computations and manage encrypted data.

Nillion Network is implementing this approach through its Nada language compiler and nilVM, with elements at all four levels of abstraction already in development. The four levels of abstraction are as follows:

• Each PET protocol runs independently in its own blind module, similar to an isolated black box. There are no unified interfaces or abstractions built in, all orchestration occurs on the client side; therefore, developers can use the API to perform specific tasks, but cannot integrate or customize it.
• Various blind modules are integrated into each SDK, providing developers with a direct and unified way to manage multiple PET protocols without requiring cryptography expertise. Although these modules are not yet fully optimized as they currently rely on a single PET protocol, a combination of PET protocols is already available for seamless, ready-to-use use.
• Blind modules start supporting multiple PET protocols within a single blind module. This provides developers with the ability to make various trade-off choices between performance and security - further simplifying decision-making for developers with limited cryptographic knowledge.
• Blindspot modules are deployed on loosely independent networks (called clusters) and are managed by NilChain. As Nillion blind computers mature, the same blind module can be replicated in multiple clusters, each with a different configuration. These configurations vary based on various factors such as number of nodes, node location, reputation, hardware specifications, security thresholds. This versatility allows developers to use the same functionality in different cluster setups, allowing solutions to be customized based on specific needs (such as security, cost, hardware, regulatory compliance, etc.).

Nillion's PET is introduced in stages, each stage passing through the four abstraction levels mentioned above. Phase 1 (i.e. HE, LSSS MPC) and phase 2 (i.e. DWT+LSSS, TEE) are progressing faster and have been integrated into the Nillion network. Among stage 3 technologies (i.e., FHE-MPC, DWT+TEE, public computing, ZKP), FHE-MPC has begun to make progress at the abstraction level.

Operation process

The following is a detailed breakdown of the operation process of Nillion network components:

• Users/developers submit data to store or initiate blind calculation requests through a front-end application built using JavaScript or Python clients.
• Applications using JavaScript clients interact with Petnet for secure computing and encrypted data management. In contrast, Python client-based applications interact with the coordination layer for payments, routing, and multi-chain communications.

The coordination layer processes payments using the corresponding blockchain's native gas token or NIL token.

• After the coordination layer processes the request, it forwards the calculation task to the Petnet containing PET.
• Petnet processes data using PET such as linear secret sharing schemes, obfuscated circuits, and/or homomorphic encryption depending on the task requirements.

These calculations will be performed on a cluster of nodes.

Each node in Petnet manages only a piece (share) of the encrypted data.

• The node performs specified calculations (such as addition, multiplication, or safe comparison) on the masked data and generates partial output.
• Petnet aggregates these partial outputs to produce the final calculation in a secure and confidential manner.
• The final result is routed back as follows:

If using a JavaScript client, Petnet sends the results directly to the application for user/developer access.

If using the Python client, the coordination layer retrieves the results from Petnet and routes them to the application or relevant blockchain for further consumption.

• For blockchain-integrated use cases, the coordination layer will pass the results to the original smart contract or decentralized application, allowing multi-chain functionality without requiring the user to download a new wallet.

Nillion’s MPC complex computing protocol

Multi-party computation (MPC) is a subfield of cryptography that allows individuals to collaboratively compute the results of their combined data without revealing their individual inputs. Nillion has developed an MPC protocol called Curl, which is based on the Linear Secret Sharing Scheme (LSSS) but extends its capabilities to efficiently handle complex operations such as division, square roots, trigonometric functions, and logarithms. This makes Curl highly scalable and well-suited for real-world problems, such as privacy-focused AI agents, where the output is not linearly related to the input. Curl uses a structured two-stage workflow:

Phase 1 (pre-processing to create shares): This phase generates randomness shares and distributes them to participants (computing entities) before processing the actual data using MPC technology. It is worth noting that the operations of the preprocessing stage are independent of the input value and only rely on the number of inputs in order to create the appropriate number of shares before the calculation takes place. It can be thought of as an abstraction layer - placeholders are created ahead of time and then combined with the actual input data provided by the user in Phase 2.

Phase 2 (Efficient Computation of Complex Operations): The computation phase consists of the actual computation on the input private data through the following three stages: (i) input; (ii) evaluation; (iii) output.

• Input: Each party distributes its input to participants, ensuring Information Theoretic Security (ITS). Each participant receives a share for each input value, and the entire process remains confidential.
• Evaluation: Parties use Nillion's Curl protocol to efficiently compute complex operations on input shares.
• Output: Local calculation results are disclosed and aggregated to produce the final result.

To learn more about Nillion’s MPC mechanism, please click here to read the original academic paper.

Application tools

Based on Nillion Network, application tools (i.e. nilVM, nilDB, nilAI and Nada integration package) provide developers with modular frameworks and utilities to quickly build privacy-preserving high-value data applications.

nilAI

nilAI is Nillion's suite of privacy technologies focused on artificial intelligence (i.e., AIVM, nada-AI, and nilTEE). Here's how each technology works:

• Artificial Intelligence Virtual Machine (AIVM): This is a secure artificial intelligence inference platform based on Nillion's MPC technology and Meta's CrypTen framework. It uses Discrete Wavelet Transform (DWT), co-developed with Meta's artificial intelligence research team, to accelerate inference. AIVM ensures data privacy by keeping individual nodes invisible to user prompts and model output, ensuring private deep learning model inference and deployment.
• nada-AI: A library of nilVM, designed for artificial intelligence applications, providing a PyTorch-like interface for running small models (such as neural network "NN", convolutional neural network "CNN", linear regression, etc.). Developers can also use Google Colab to quickly bootstrap their projects.
• nilTEE: This solution uses a Trusted Execution Environment (TEE) to run large language models (LLMs) with high performance during inference. Nillion recommends limiting the use of TEE to inference time, not long-term data storage. Nillion is currently conducting research to enhance nilTEE and AIVM by separating inference settings to further improve security and performance.

nilVM, Nada and their libraries

nilVM is a virtual machine that allows developers to create programs using PET. The program is written by Nillion's Python-based open source DSL Nada and developed using Nillion SDK. Nada also includes libraries such as nada-ai (similar to PyTorch and scikit-learn), nada-numpy, nada-data and nada-test to simplify program development. Developers can integrate nilVM into their applications using Python, Typescript, or CLI clients and leverage the storage API for secure data storage and retrieval on the Nillion Network. Examples include joint learning initiatives, community development projects, and interactive demonstration use cases.

nilDB

nilDB is a cryptographically distributed NoSQL database designed for privacy-preserving data storage and computation. Unlike ordinary NoSQL databases, nilDB distributes encrypted data as a secret share across multiple nodes, thereby eliminating dependence on a central authority. In addition, data owners can grant others access to run SQL-like queries, calculations, and privacy-preserving aggregations on the stored data.

The specific operations are as follows:

• Users encrypt sensitive data locally on their devices.
• Users securely upload encrypted data through a Nillion-based front-end application. The application securely uploads encrypted data to nilDB via the integrated backend RESTful API.
• Encrypted data is split into secret shares using Nillion's MPC protocol and distributed across a cluster of nodes in the nilDB network. It is worth noting that no single node has the complete data set.
• Users provide explicit consent to the use or query of specific data and can revoke consent at any time through the application.
• Permitted entities (such as companies or third parties) submit SQL-like query requests (such as lookups, range filters, or summary calculations) through Nillion's RESTful API.
• Nodes in a nilDB cluster collaboratively perform computations on encrypted data without exposing sensitive information.
• Query results (such as averages, sums, or filtered data sets) are generated while maintaining data confidentiality.
• Only the final query results will be returned to the requesting user through the RESTful API.
• For more information about the technical architecture, click here .

Nada integration package

The Nada language includes various integration packages, including nada-AI (already discussed earlier), nada-numpy and nada-test. The use cases are as follows:

• nada-numpy: A NumPy restricted adaptation package tailored for the Nada DSL. Compared with ordinary NumPy, nada-numpy allows efficient manipulation of array structures and puts forward strong type requirements for data types to ensure compatibility with the strong type features of MPC.
• nada-test: A testing framework for Nada programs that supports generating dynamic tests at runtime. Developers can write test cases in Python, integrate the framework into pytest workflows, and define flexible input and output specifications.

Other tools (such as Nada DSL, Nada Sandbox, etc.) and SDK can be viewed on GitHub.

NIL token

Token utility

NIL tokens will perform multiple functions in the Nillion network, including

• Pay for computing services, data storage, AI inference, and transaction fees for the Petnet and coordination layers. Specifically, developers can use NIL to access the privacy-preserving computing services Nillion provides for their applications.
• Stake and support network security and earn rewards.

Validators bind NIL to verify transactions and calculations, ensuring the security of the coordination layer.

Petnet nodes stake NIL to increase the security of their cluster and attract developers and applications.

• Participate in decentralized governance by making recommendations and voting on various network decisions such as protocol upgrades, resource allocation, and community grant programs.

governance

Governance decisions are made through an on-chain voting mechanism. Specifically, any NIL token holder who meets the minimum token holding requirements can propose concepts to the network. Community committees or working groups established through previous governance actions may also submit proposals.

Voting rights apply to key decisions, e.g.

• Introduce new features or updates.
• Allocate reward pools for grants, developer rewards, and community-driven projects.
• Adjust network pricing, validator requirements, or authorization limits.
• Modify governance structures such as quorum requirements or proposal thresholds.
• Expand interoperability, establish strategic partnerships, or implement transparency and audit mechanisms.
• Voting power is proportional to the amount of NIL staked, and stakers delegate their voting power to validators while retaining their own ability to vote on proposals.

Nillion Ecosystem

Nillion creates new opportunities in the following industries:

• Artificial Intelligence: Nillion processes data and infers without exposing sensitive information, bridging the gap between secure local AI processing and the scalability of centralized, non-proprietary AI systems.
• Personalized agents: AI agents can store, compute and process private data.
• Privacy model reasoning: AI models can securely handle private data, minimize the risk of exposure to third parties, and enable private LLM.
• Private knowledge base and search: Data can be stored in encrypted form while still providing search capabilities for AI agents and other AI use cases.
• Data Ownership: Nillion’s cryptographic infrastructure supports a secure data marketplace, allowing users to control and sell their own data to buyers.
• Blockchain: Nillion allows blockchain applications to send blind storage and computing requests to the Nillion network, supplementing the public data function of the blockchain. It also supports on-chain settlement, allowing applications to decrypt related data on the blockchain.
• Healthcare: Nillion enables privacy-preserving analytics of healthcare data across institutions and users.
• DePIN: Integrated with Nillion, the DePIN project can securely store and process sensitive operational data.

Key projects

• Virtuals Protocol: An AI agent building platform that develops a multi-modal AI agent library and, using Nillion, allows private training and inference of its AI models to build personalized AI agents.
• Aptos/NEAR/Arbitrum/Sei: Layer 1 and Layer 2 blockchains that integrate blind data storage and computation to enhance data processing within smart contracts.
• Ritual: An artificial intelligence platform that builds a decentralized artificial intelligence reasoning network, integrating Nillion in its backend for private reasoning.
• Zap: A data platform that aggregates user data into a decentralized data pool in Nillion, providing secure insights through blind computing and zero-knowledge transport layer security (zkTLS).
• Reclaim Protocol: The zkTLS infrastructure platform allows users to prove their identity and reputation through a trusted off-chain platform, and uses Nillion as the storage and processing platform for the generated certificates.
• Healthblocks: A fitness app that uses Nillion to maintain user ownership and control of data while allowing third parties to gain insights without exposing personal details.
• MonadicDNA: A genomics platform that uses Nillion to encrypt data throughout its lifecycle, providing an alternative to centralized service providers such as 23andMe.

roadmap

The Nillion roadmap was released on May 31, 2024 and is divided into four key stages:

• Phase 1 - Creation Sprint (Completed). This phase establishes: (i) the basic coordination layer during the testnet launch; (ii) testing core functions such as Keplr wallet creation, token transfer, staking and management; (iii) providing developers with access to the Nillion SDK, The SDK has telemetry capabilities for early application development; (iv) load testing is performed to evaluate transaction throughput and network scalability.
• Phase 2 - Catalyst Fusion (in progress). This stage: (i) integrate Petnet with the coordination layer; (ii) add external nodes to achieve complete decentralization; (iii) introduce "blind applications" for secure data processing; (iv) support cross-chain functions, Nillion expands into a multi-chain ecosystem.
• Phase 3 – Reinforcement. This phase will: (i) include the mainnet launch and token generation activities (TGE); (ii) run external nodes; (iii) enable real-world interactions through blind computation; (iv) verify the network’s previously constructed builds under real-time conditions application.
• Phase 4 - The future of multi-clusters. This phase will: (i) achieve horizontal scaling by adding clusters of public nodes; (ii) increase computing power; (iii) optimize the network for market-specific applications; (iv) achieve scalability while maintaining security and privacy sex.

Conclusion

Nillion is a decentralized infrastructure designed to handle high-value, privacy-sensitive data in a variety of applications, from artificial intelligence agents to private DeFi. Nillion combines advanced PET (such as MPC, FHE, TEE) to expand the usability of decentralized networks and the possibilities of decentralized applications. Nillion's architecture -coordination layer and Petnet - supports scalability through clustering while ensuring data confidentiality and decentralized trust.

The Nillion ecosystem is constantly expanding, with milestones including: (i) the Nucleus Builder Program (supporting ~50 projects across multiple verticals) and (ii) ~500,000 validators have participated, with a total of ~195 million processed A secret message, protecting approximately 1,050 GB of data. Collaborations with Virtuals, NEAR, Meta and Aptos, as well as ongoing mainnet launch and multi-cluster scalability roadmap development, highlight Nillion’s progress in advancing privacy-focused data management and secure computing.