Truth Swarm

Imagine building the ultimate consumer protection helper for the AI agent revolution: a swarm of ASI-powered evaluators that probe other agents with clever prompts, rigorously scoring their outputs for accuracy, correctness, and lightning-fast speed, then mints tamper-proof ratings on Ethereum. Leveraging Fetch.ai uAgents for seamless multi-agent collaboration, MeTTa for probabilistic reasoning that uncovers hidden insights, and ASI:One for intuitive human oversight, we'll create a system that's not just smart—it's a game-changer for decentralized AI ecosystems. It's tough to know what an agent will do before using it but with Truth Swarm we can verify them before you use them! Making sure you can interact with these agents without worry.

Truth Swarm combines multi-agent orchestration, probabilistic reasoning, quality assurance, and on-chain verification into one cohesive system. We built the architecture around four key layers: Agent Orchestration Layer (ASI1 + Agentverse) Each verifier agent is deployed and managed through Agentverse, registered with unique Fetch.ai identities. ASI1 handles user facing chat features. OpenAi handles LangSmith evals under the hood. Communication & Coordination (Fetch.ai + ACP) Fetch.ai uAgents provide secure, peer-to-peer messaging and discovery between verifier agents. Agent Communication Protocol (ACP) defines how prompts, tasks, and scores are passed among evaluators. This ensures evaluators can probe other AI agents efficiently and exchange verifiable feedback. Evaluation & Reasoning Engine (MeTTa Knowledge Framework) MeTTa is used to model reasoning patterns and probabilistic truth inference. Each agent records its reasoning trace, weighting consensus across the swarm to reduce hallucination and bias. Results are standardized into verifiable truth claims. On-Chain Verification (Ethereum + Smart Contracts) Verified ratings are written as tamper-proof attestations to Ethereum Attestation Service EAS. We use a lightweight scoring contract that stores agent trust metrics and evaluation hashes. The on-chain layer also mints NFTs for verified “trust proofs,” usable by any dApp that wants agent transparency.

We use the Blockscout SDK for real-time transaction monitoring via webhooks. Whenever a verification-related transaction is confirmed on-chain on our website, Blockscout instantly pushes updates to our front end, triggering UI refreshes and human verification notifications. This setup allows Truth Swarm to display attestation details without manual polling, keeping users seamlessly informed about the state of their verifications.

We use Wagmi for wallet integration and transaction management.

Tech Stack: Languages: Python, Solidity, JavaScript Frameworks: Agentverse, MeTTa, ASI1 SDK Protocols: Fetch.ai uAgents, ACP Blockchain: Ethereum (via Foundry + Ethers.js + wagmi) Infra: Vercel (frontend), Pinata/IPFS (metadata), Alchemy (RPC + event indexing), Blockscout SDK (webhooks + real-time transaction notifications)