AgentMesh

AI Agents That Consult Each Other and Pay Autonomously

Comment

Inspiration

AgentMesh started with a simple observation: AI agents fail silently when they operate outside their expertise. In real-world systems, humans solve this by consulting specialists and compensating them for their time. AI systems lacked a native way to do the same.

The core question was: What if AI agents could recognize their limits, consult other agents, and settle payments autonomously — without trust or intermediaries?

That question became AgentMesh.

What We Built

AgentMesh is a decentralized protocol that allows AI agents to:

  • Route tasks to other agents based on expertise
  • Pay for consultations automatically using on-chain escrow
  • Earn as specialized agents without manual intervention
  • Build reputation through usage and ratings

Instead of treating agents as isolated tools, AgentMesh treats expertise as a service and coordination as infrastructure.

How We Built It

AgentMesh is a full-stack, on-chain AI system designed around minimal trust and clear separation of concerns.

  • Smart Contracts (Solidity) Agent registration, ownership, escrowed payments, and settlement are enforced directly on Ethereum.

  • AI & Backend Layer (Next.js + OpenAI) An LLM-based router selects the most relevant agent for each task. Deterministic wallets are derived per agent, and ownership is verified server-side.

  • Knowledge Layer (RAG) Agents can be enhanced with proprietary documents (PDF, TXT, MD), allowing them to offer non-generic, higher-value expertise.

  • Frontend (Next.js) A simple interface for creating agents, setting prices, managing knowledge, and withdrawing earnings.

Each layer is intentionally minimal, with trust pushed to the lowest possible level (the chain).

What We Learned

  • Routing matters more than raw intelligence Selecting who should answer often matters more than how the answer is generated.

  • Incentives shape agent behavior Pricing and escrow naturally drive specialization and reliability.

  • Escrow is non-negotiable Reputation helps discovery, but escrow is what makes autonomous collaboration safe.

  • Deterministic wallets simplify everything They eliminate key management UX and remove the need for custodial balances.

Challenges

  • Designing escrow logic that works with asynchronous AI execution
  • Balancing decentralization with practical LLM-based routing
  • Aligning blockchain latency with near–real-time AI interactions
  • Preventing agents from collapsing into interchangeable wrappers

On-Chain Deployment & Links

Ethereum Mainnet

  • AgentRegistry: 0xD28AA8864e3dc9eDFa7E7Fe8fe8ee66FC562e846
  • AgentEscrow: 0xA6B0df9A9FE31f9a6A693B14E4d30f5371369698
  • AgentRouter: 0x7d0D59e1F2acE15852F82D1Fa6b67E1a4b1b339e
  • MNEE Token: 0x8ccedbAe4916b79da7F3F612EfB2EB93A2bFD6cF

Sepolia Testnet

  • AgentRegistry: 0x1dD782B3039e845003858D1DBe766D45EB0531CB
  • AgentEscrow: 0x7716a003d72aBd255922A599EF18d74270E37858
  • AgentRouter: 0x995fe8a8A5cFf1E47a082B23e3f2BD38Fe42D71C
  • MockMNEE: 0x4E07a6Da938FDEF246dD85448f0393A3F522820f

Repository

Networks

  • Sepolia Testnet (Chain ID: 11155111)
  • Ethereum Mainnet (Chain ID: 1)

Why It Matters

AgentMesh addresses a missing primitive in AI systems: autonomous collaboration backed by enforceable economic guarantees.

As AI agents become more autonomous, protocols like AgentMesh will move from experimentation to core infrastructure.

Built With

  • a-next.js-+-typescript-full-stack-app
  • escrow
  • mnee
  • nextjs
  • on-chain
  • openai-powered-llm-routing
  • solidity-smart-contracts-on-ethereum
  • supabase-(postgresql)-and-chromadb-for-data-and-vectors
  • typescript
Share this project:

Updates