Yield, Routed.
NanuqFi is an AI-powered, protocol-agnostic yield routing layer built for Solana. Deposit USDC. Pick your risk. Let the protocol work.
See it in 6 minutes
Full walkthrough: marketing → dashboard → vaults & deposit flow → keeper API → on-chain program → build & tests.
DeFi yield is fragmented and fragile
Yield moves constantly
Rates change every hour across 10+ lending protocols on Solana. Stagnant capital leaves returns on the table.
Single vaults are fragile
Concentrated risk. One vulnerability or bad-debt event in a single protocol = total loss for depositors.
Manual routing fails
Retail users don't track live APYs or pool utilization 24/7. It doesn't scale.
One deposit. Three protocols. Zero thinking.
Routing capital intelligently based on real-time risk/reward.
Protocol-Agnostic
Add new yield protocols by implementing a single YieldBackend interface.
AI-Enhanced Validation
Claude validates every rebalance proposal with regime detection and confidence scoring.
On-chain Guardrails
Emergency halt, deposit caps, and drawdown limits enforced by the program, not the keeper.
Backtest-Proven
Algorithm delivered 20.15% return over 469 days of real on-chain historical data.
This isn't a mockup
Live metrics from current devnet deployment.
Don't trust us, verify.
We backtested our router against 469 days of real on-chain data from Kamino, Marginfi, and Lulo. The dynamic routing beats every protocol individually on Sharpe ratio by avoiding utilization spikes.
20.15% RETURN
11.76 SHARPE
Trust the program, not the keeper
The AI keeper is strictly advisory. It proposes allocations; the on-chain allocator validates every constraint before any capital moves.
Yield
AI Keeper Flow
- AI ProposesOff-chain node computes optimal weights and builds TX.
- Program ValidatesOn-chain checks: caps, drawdown, slippage — math, not trust.
- CPI ExecutionCross-program invocations route funds into Kamino/Marginfi/Lulo.
Built as a Ranger Earn adaptor
NanuqFi is deployed as a dedicated Ranger Earn adaptor — a yield-routing primitive that plugs into Ranger's strategy framework. Integration tests passing: 8/8 on devnet.
Why we ship on devnet first
This is a deliberate architectural choice, not a hackathon limitation.
Independent audit pending
We are pricing OtterSec, Kudelski, and Trail of Bits for a formal security review. Real USDC does not touch un-audited custodial contracts.
External dependency simulation
Kamino, Marginfi, and Lulo CPIs need long-running continuous simulation on devnet before customer funds route through them in production.
Keeper adversarial testing
Emergency halt, guardrails, and AI decision paths need real-world stress testing under synthetic volatile market conditions.
Staged rollout model
Launch pipeline: audit → mainnet beta with strict $10k vault caps → gradual TVL cap raise → full launch. 3–6 month horizon.
“We built production-complete logic. What is NOT yet complete is the external validation required to protect user money. That is the standard mainnet requires.”
On the hackathon requirements
Three framings for the Ranger Build-a-Bear judges.
Solana devnet is a real public cluster. All deposits, withdrawals, rebalances, and program state are verifiable through Solscan with ?cluster=devnet. Same primitives, no mainnet liquidity — which is exactly why we chose it for pre-audit deployment.
Current 6.7% reflects a conservative menu — Kamino, Marginfi, Lulo — all straight USDC lending that passes every disqualifying-yield-source filter (no LP, no junior tranche, no leveraged looping). Adding one compliant higher-yield backend via the YieldBackend interface clears 10%+ with a single-file change. We optimized for 11.76 Sharpe and zero drawdown over raw APY.
Program deployed to devnet on April 11 (after finalizing 850+ tests). The public repo carries 150 commits across strategy engine, Anchor program, adaptor, and frontend during March 9 – April 6. On-chain activity starts post-deploy; development is fully auditable in Git.
GitHub history