agentmesh
AI-powered autonomous research platform on Injective. One ERC-7715 permission unlocks a 3-agent pipeline Research, Analysis, and Output coordinated on-chain via ERC-7710 redelegation, paid per-inf
Videos
Tech Stack
Description
AgentMesh is a decentralized autonomous research platform built on Injective. Users type a research question, grant a single ERC-7715 permission, and a coordinated swarm of AI agents handles everything else autonomously.
The Orchestrator Agent receives the query and creates three sub-delegations via ERC-7710 redelegation each specialist agent gets a capped USDC spending limit enforced at the smart account level. The Research Agent uses Venice AI embeddings to semantically search and retrieve sources. The Analysis Agent synthesizes raw data into a structured report using Venice Llama 3.3 70b. The Output Agent generates a thematic cover image via Venice Flux Dev and packages the final deliverable.
Every Venice AI inference call is gated by the x402 HTTP Payment Required protocol agents pay for exactly the compute they consume, with no pre-purchased API quotas. All Base Mainnet transactions are routed through the 1Shot Permissionless Relayer, meaning users need zero ETH the entire experience runs on USDC.
The result: one question, one signature, a fully cited research report with AI-generated cover art delivered autonomously, paid for on-chain, coordinated by smart accounts.
Progress During Hackathon
Week 1: Architected the multi-agent pipeline and integrated the Injective MCP Server. Implemented ERC-7715 single-permission grant flow using MetaMask Smart Accounts Kit. Set up the Orchestrator Agent with ERC-7710 redelegation logic to fund sub-agents with capped USDC allowances.
Week 2: Integrated all three Venice AI endpoints Llama 3.3 70b for research and analysis, Flux Dev for cover image generation, and embeddings for semantic search. Wired x402 payment headers into every Venice inference call. Connected the 1Shot Permissionless Relayer for gasless Base Mainnet execution.
Week 3: Built the brutalist dark-terminal frontend in Next.js real-time agent pipeline display via SSE, live spending tracker, report viewer with scanline cover image overlay, and one-click kill switch. Wrote full README, DEPLOY.md, and recorded the demo video. Deployed to Vercel and verified end-to-end flow on Base Mainnet.
Shipped: 3-agent autonomous pipeline, ERC-7715 + ERC-7710 integration, x402 on all Venice calls, 1Shot relay, full Next.js frontend, 60+ lines of deployment docs.