Building Verifiable Onchain AI Agents with LazAI

As decentralized AI becomes a more critical part of the blockchain ecosystem, developers are increasingly seeking ways to make AI operations transparent, verifiable, and reward-driven. In our recent live workshop, Pavel and Thiru from the LazAI team presented a deep dive into building onchain AI agents using the LazAI frameworks, focusing on data contribution, job orchestration, and secure inference execution.

This blog post summarizes the core concepts, workflows, and tools covered during the session. If you prefer video format then check out the YouTube live recording here.

The Need for Verifiable AI

Traditional AI systems are often opaque. As users, we rarely know if the model used the right data, whether the inference process was tampered with, or if results can be trusted, especially in high-stakes financial operations. LazAI addresses this trust gap by creating a verifiable AI marketplace that cryptographically registers data, distributes jobs, and confirms results through a decentralized settlement layer.

Components of the LazAI System

• Agents represent the drivers of AI tasks, executing on top of foundational LLMs.
• Data acts as the navigational input for these agents.
• LLMs (Large Language Models) are the infrastructure powering AI reasoning.
  
  

LazAI’s goal is to create a full-stack solution where data submission, processing, and outcome validation all happen onchain.

The Architecture: Alith, LazAI Client, and Hyperion

At the heart of this architecture lies Alith, a client-side framework for building AI agents. Underpinning Alith is the LazAI Client, which handles all onchain interactions, including job execution and settlement, on the Hyperion blockchain.

Key Functions:

• Data Registration: Allows users to register training or inference data via an IPFS hash.
• Proof Submission: Verifies that submitted data is valid and accessible.
• Inference Execution: Jobs are sent to nodes, which execute the request and post results onchain.
• Tokenized Rewards: Contributors are compensated with DATs (Data Anchoring Tokens), ERC-1155 NFTs that represent data ownership and future royalties.
  
  

Workflow Overview

The lifecycle of a typical onchain AI task includes:

1. Data Contribution:
   
   • Encrypt and upload data.
   • Register the data hash onchain via the LazAI client.
   • Submit the data to IPFS (e.g., via Pinata).
   • A job is created for data verification by nodes.
   • Upon successful verification, a DAT is minted to the contributor’s wallet.
     
     
2. Job Execution and Inference:
   
   • A user deposits funds to a node for inference.
   • The user sends a request to a node using a defined model and prompt.
   • The node processes the request and submits proof of work.
   • The smart contract verifies completion and distributes the reward.
     
     
3. Settlement and Governance:
   
   • Settlement contracts ensure fair payments to node operators.
   • IDOWs (Independent DAOs) manage data registries and act as governing entities.
   • Proxy contracts ensure upgradability without compromising existing registries.

Code Examples

The session featured multiple examples, particularly using Python and Node.js. A few highlights:

Registering Data

const file = await client.addFile("encrypted_file.json");

const proofRequest = await client.requestProof(file.fileId);
‍

Running Inference

const inferenceUrl = await client.getNodeUrl(nodeAddress);

const result = await axios.post(`${inferenceUrl}/v1/infer`, {

  fileId: "74",

  prompt: "Translate to French"

});

Verifying Data onchain

The smart contract stores the file hash and verifies existence through proofs generated by nodes.

Each contract (Data Registry, AI Process, Settlement) has specific functions to ensure modularity and traceability.

Tooling and Developer Support

LazAI offers SDKs in:

• Python
• Node.js
• (Rust examples are in progress.)
  
  

The team maintains open repositories on GitHub under the 0xLazAI organization. Developers can access documentation, boilerplate code, and examples for integrating with the network.

HyperHack Bounty

As part of the HyperHack, participants are encouraged to build architectural diagrams and implementations using the LazAI system. The focus should be on conceptual integration rather than complete technical deployment. Projects can include:

• Building inference agents that consume onchain data.
• Designing DAO-based governance for data verification.
• Creating job queues for specialized AI tasks.

Final Thoughts

The LazAI framework represents a step forward in making AI computation trustless, decentralized, and fairly incentivized. By leveraging cryptographic proofs, tokenized data ownership, and modular smart contracts, developers can now build AI agents that are not only powerful but also transparent and verifiable.

Watch the Full Workshop:

Onchain AI Inferencing Workshop on YouTube

‍