AlphaDawg

AlphaDawg is an ENS-native, autonomous multi-agent economy for discovering, hiring, coordinating, and paying specialized AI agents. A personal Lead Dawg acts as the orchestrator. For every investment hunt, it assembles a task-specific swarm of research, sentiment, onchain-forensics, risk, and execution agents. Each specialist has an immutable version, capability manifest, pricing policy, payout identity, and human-readable ENS identity. The swarm hires specialists through machine-to-machine x402 USDC nanopayments instead of relying on one monolithic model. Every specialist produces intelligence through 0G’s verifiable AI stack. AlphaDawg binds each request and response to the exact agent, model, provider, buyer, job, and deadline. Results are stored on 0G Storage and accepted only after proof-enabled readback verifies the exact bytes, root, digest, and size. Verified intelligence enters an adversarial Alpha-versus-Risk debate. Only policy-approved consensus can reach the Executor Agent and become a Uniswap action. Hedera records the swarm lifecycle—agent hiring, debate, payment, receipt, settlement, and execution—as a tamper-resistant audit trail. The dashboard exposes this complete Proof Rail, transforming an opaque trading bot into a transparent, composable, and trust-minimized network of autonomous economic agents.

AlphaDawg uses a hierarchical multi-agent architecture built with Next.js 16.2, React 19, strict TypeScript, PostgreSQL, Prisma, viem, SIWE, Express, and isolated specialist-agent services. The Lead Dawg is the only orchestration and signing boundary. It discovers specialists, validates their capabilities, purchases their intelligence, coordinates the adversarial debate, enforces risk policy, and authorizes execution. Specialist agents remain inference-only and never receive trading keys. ENS provides agent identity and authority. Using viem’s Registry, Public Resolver, and Universal Resolver paths, AlphaDawg resolves each specialist’s owner, delegate, version, capability manifest, service endpoint, payout address, and policy. Authority is checked before 0G inference and again immediately before delivery, preventing revoked, impersonated, or silently upgraded agents from receiving payment. Agent-to-agent commerce uses x402 USDC nanopayments with Circle Gateway batching. The buyer signs requests through a viem account, while each Express specialist exposes a paid analysis endpoint. This lets the orchestrator dynamically hire only the agents needed for a particular hunt. 0G provides sealed inference and verifiable storage. Canonical requests are cryptographically bound before single-use payment headers are created. Outputs are uploaded to 0G Storage and independently verified through a small Go boundary using the official 0G Storage client. We added this verifier because a TypeScript verified: true flag alone could not prove that the downloaded bytes matched the expected Merkle root, digest, and size. PostgreSQL acts as the deterministic coordination layer. Immutable AgentVersions, Quotes, JobIntents, effects, receipts, and settlements are protected with database constraints and idempotency keys. Worker leases use owner, epoch, version, and expiry fencing, preventing crashed or stale agents from duplicating payments or finalizing another worker’s job. Hedera HCS records structured swarm events, while HTS represents the agent economy and settlement layer. After verified multi-agent consensus, the Executor Agent can route an approved action through Uniswap. The same receipt connects ENS authority, x402 payment, 0G verification, Hedera audit evidence, and final onchain execution.