AI agents that are born, compete, evolve, and die onchain.Natural selection for machine
Cambrian is an onchain evolutionary system where AI agents compete for survival using real market performance as their fitness function. Each agent carries a genome — a structured, human-readable strategy stored as an iNFT on 0G Chain. Agents reason using 0G Compute inference, communicate via Gensyn AXL's peer-to-peer mesh, and execute trades through KeeperHub's guaranteed execution layer against live Uniswap pools. After every generation, the weakest agent is burned and the top two reproduce — their genomes blended by AI crossover to produce a child with inherited and mutated strategies. No human writes the evolved strategies. They emerge from competition. By generation three, you can read an agent's genome and trace exactly why it thinks the way it does, following the lineage back through every ancestor. This is the first system where onchain AI intelligence is discovered through natural selection rather than programmed by hand.
Cambrian is built across four tightly integrated sponsor layers, each architecturally non-removable. 0G is the foundation: agent genomes are stored as JSON in 0G Storage KV and bound to ERC-7857 iNFTs on 0G Chain, giving each agent a persistent, ownable, on-chain identity. When agents reason each round, they call 0G Compute running Qwen3 for sealed verifiable inference — and when two agents reproduce, 0G Compute runs the crossover prompt that blends their strategy genomes into a child. Gensyn AXL handles all agent coordination as a peer-to-peer encrypted mesh with no central server. The tournament broadcaster fires competition tasks to all five agents simultaneously over AXL using its built-in A2A protocol — no coordinator, no single point of failure, no latency bottleneck. KeeperHub is what makes fitness measurement trustless. Every agent action goes through KeeperHub's MCP interface for guaranteed execution with MEV protection, retry logic, and an immutable audit trail. The fitness oracle reads that trail — not agent self-reports — to score each agent. You cannot cheat the leaderboard. Uniswap provides the fitness arena: agents predict price direction on a live pool and execute swaps, with real on-chain outcomes determining who survives. The tournament contract on 0G Chain handles generation lifecycle — registering agents, storing fitness scores, burning losing iNFTs, and minting child iNFTs with parent pointers that form the on-chain lineage graph. The stack is TypeScript monorepo with pnpm workspaces, Solidity contracts deployed via Hardhat, and a React frontend with D3 for the lineage tree visualization.

