NeuralMesh

NeuralMesh is a decentralized multi-agent AI infrastructure built for Open Agents. It runs five specialized agents (for now) (planner, researcher, executor, evaluator, and evolution) that discover each other via ENS subnames on Sepolia, coordinate over a mesh-oriented architecture, and trigger economic actions through programmable workflows. The planner routes tasks, the researcher generates answers, the evaluator scores quality and writes reputation to ENS, and the evolution loop bumps versioning after threshold milestones. The result is a verifiable, composable agent network with identity, coordination, payments, and continuous improvement.

NeuralMesh is built as a TypeScript monorepo with pnpm + Turbo and a React/Vite dashboard. We use ENS (Sepolia) for identity and discovery, storing agent metadata (pubkey, reputation, version, task count) in text records through a custom resolver flow. For agent transport, we integrate a Gensyn AXL Go binary per agent and bridge it to the TypeScript runtime, with graceful fallback paths where needed. For compute, agents call TokenRouter via an OpenAI-compatible API wrapper. KeeperHub is used for workflow automation and x402-style micropayment flows (health checks, evolution triggers, payment settlement). Smart contract/deployment scripting is handled with Hardhat + viem.