HACKATHON DEMO — DEV3PACK / COLOSSEUM FRONTIER
SwarmOS
Autonomous DeFi Yield Optimization — Darwinian AI on Solana
The Problem
DeFi yields change hourly across dozens of protocols. A single AI agent making a one-shot recommendation has no feedback mechanism — if it's wrong, it's wrong silently. There's no selection pressure. No learning.
SwarmOS fixes this with Darwinian selection on Solana.
Live Yield Opportunities
Loading live yield data…
How The Swarm Works
01
Spawn
5 agents deployed per generation. Each reads live yield data and makes a recommendation. The recommendation APY, protocol, and output hash are stored in the Agent PDA, and the program creates an agent-owned USDC token account funded from the swarm treasury.
02
Score (via x402)
Each agent pays
$0.01 USDC via x402 protocol to the scoring oracle before receiving its evaluation. No payment → no score. The oracle compares claimed APY vs live yield data, while the Anchor program rejects suspiciously high claims receiving top scores:
Claim Kamino at 18.0% · real is 9.26% → score: 35/100
Claim Kamino at 9.3% · real is 9.26% → score: 98/100
03
Eliminate
Agents scoring below 60/100 are terminated via on-chain Anchor instruction. Their USDC is reclaimed to the swarm treasury. Their score and failure hash are written to a LineageMemory PDA, while the Agent PDA preserves the original claim and output hash.
04
Inherit
Successor agents read the LineageMemory PDA before executing. The failure hash is injected into their system prompt:
⚠ INHERITED FAILURE MEMORY (Generation 1):
Do not claim APY above live oracle rate for Kamino SOL/USDC.
Previous agent's stored claim was rejected and its USDC was reclaimed.
Failure hash: 7f3a2b…e91c
The swarm learns. Generation 3 agents have never seen generation 1's mistakes — but they carry them.
Swarm State — Live
No terminations yet — run the swarm to see data here.
Verify on Solana Explorer
Every agent lifecycle event is on-chain and verifiable. No backend database. No centralized state. The chain is the ledger.
X402 Agent Micropayments
Every agent pays $0.01 USDC via the x402 protocol before receiving its score. No payment → no evaluation. This is autonomous agent-to-agent commerce on Solana.
HTTP/1.1 402 Payment Required
x-payment-required: [base64 payload]
scheme: exact
network: solana:devnet
amount: 10000 (0.01 USDC)
payTo: D14J1wLNEZkHEBcM9NW9nUwCkhxuJSUvE5G3E38frDJs
Fund From Any Chain
The swarm treasury accepts USDC from any chain. LI.FI routes the optimal swap and bridge automatically. Powered by Jumper Exchange.
⚡Fund Swarm(loading treasury…)