CraftLead

CraftLead is an autonomous job intelligence platform powered by a 4-agent AI swarm: Planner decomposes user missions into research tasks, Researcher analyzes job data from Web3.career, Greenhouse, LinkedIn and RapidAPI, Critic reviews quality with an iterative feedback loop (confidence < 0.7 triggers refinement, max 2 iterations), and Executor compiles structured reports with confidence scores. All agent outputs persist on 0G Storage's two-layer architecture (File Layer for immutable artifacts with Merkle proofs, KV Layer for mutable state), enabling agents to maintain memory across sessions. Users buy credits via Uniswap Trading API v1 (multi-token: ETH, USDC, USDT, WBTC, DAI with Permit2 gasless approvals). Agents earn real USDC through on-chain treasury settlement (Planner 15%, Researcher 30%, Critic 20%, Executor 35%) — 3 proven mainnet transactions on Ethereum. Completed missions can be minted as Intelligent NFTs (ERC-721) on 0G Galileo Testnet, with memory root hashes embedded on-chain and verifiable via getAgentInfo() contract call. Agent communication runs through Gensyn AXL encrypted P2P mesh (4 dedicated AXL nodes with Yggdrasil IPv6 overlay, Ed25519 identities). AI inference uses 0G Compute Router (GLM-5-FP8, TEE-verified) with automatic ZhipuAI fallback. Built with Next.js 16, React 19, Supabase, Foundry, and the OpenClaw agent framework.

Architecture: Next.js 16 App Router frontend with React 19, TypeScript, Tailwind CSS v4. Server Actions ('use server') handle all backend operations. Supabase (PostgreSQL with RLS) stores users, missions, agent_logs, artifacts, and agent_earnings across 8 migration files.

Agent Swarm: The orchestrator (lib/agents/orchestrator.ts) coordinates 4 agents sequentially with a feedback loop. Planner uses structured output to decompose missions into 3-5 tasks. Researcher analyzes job data per task. Critic scores quality (0-1 confidence) and sends feedback if below 0.7 threshold. The loop runs max 2 times before Executor compiles the final report.

0G Integration: We use @0gfoundation/0g-ts-sdk@^1.2.6 for Storage (File Layer via MerkleTree + Indexer.upload(), KV Layer with 30s TTL cache) and @0gfoundation/0g-compute-ts-sdk for inference through 0G Compute Router. All agent outputs are saved fire-and-forget with 30s timeout to avoid blocking the swarm.

Uniswap Integration: Dual use of Trading API v1 — (1) user credit purchases via multi-token swaps with Permit2 EIP-712 gasless approvals, (2) agent-initiated on-chain settlement where the treasury swaps ETH→USDC and distributes per the 15/30/20/35 split. We handle EXACT_OUTPUT for credit packs and EXACT_INPUT for settlements. 3 proven mainnet TXs on Etherscan.

Gensyn AXL: 4 AXL nodes deployed via Docker (multi-stage Go 1.25.5 build). Each agent has its own Ed25519 keypair and dedicated API port. Listener processes poll /recv endpoints, run the agent locally, and send results via /send. The orchestrator has dual-mode: AXL mesh if peers available, in-process fallback otherwise.

Smart Contracts: CraftLeadAgentNFT (ERC-721) on 0G Galileo Testnet at 0x01ed4074C8932C274b09c1A4BaA380Ae8D8c7Ab0. Solidity 0.8.26 with Foundry. Functions include mint(), updateMemoryRoot(), incrementMissionCount(), getAgentInfo(). The updateMemoryRoot() call links the NFT to 0G Storage, verifiable via our memory root verification UI.

OpenClaw: Custom craftlead-daily-mission skill with 8 deterministic scripts (sync-jobs, match-jobs, og-memory-adapter, axl-broadcast, mint-inft, uniswap-settlement, appflowy-export, supabase-client) orchestrated by a single SKILL.md triggered by cron.

Notable: Resilient execution with 4 retries + exponential backoff for AI inference. Non-blocking 0G Storage saves prevent swarm slowdown. We published @craftlead/schema (types + Zod), @craftlead/sdk (8 modules), and @craftlead/mcp (MCP server for job sources) as npm packages.