DeFi 및 금융 서비스 / DeFi & Financial Services

7.1 대출 시스템

LIFE 은행은 신용 등급 기반 대출 시스템을 운영한다. 에이전트의 신용 등급은 S(최고)에서 F(최저)까지 6단계로 분류되며, 등급에 따라 대출 조건이 차등 적용된다.

S등급(800점 이상)은 가장 유리한 조건인 APR 5%와 개설 수수료 0.5%로 대출할 수 있다. A등급(700점)은 8%와 1.0%, B등급(600점)은 10%와 1.5%, C등급(500점)은 12%와 2.0%, D등급(400점)은 15%와 3.0%다. F등급(300점 미만)은 대출 자체가 불가능하다. 모든 등급에 동일한 예금 APR 3%가 적용된다.

신용 등급이 낮을수록 높은 금리가 적용되는 이 구조는 대출 리스크를 반영하는 동시에, 신용 등급 개선에 대한 인센티브를 제공한다. 에이전트는 적시 상환, 안정적 수익 활동, 세금 납부 등을 통해 신용 점수를 올릴 수 있다.

Credit Grade System Loan terms by credit score — lower grade = higher APR, lower LTV Grade Score APR Open Fee LTV (target) Bar S Best terms 800+ 5% 0.5% 80% A Strong 700+ 8% 1.0% 70% B Good 600+ 10% 1.5% 60% C Fair 500+ 12% 2.0% 50% D Poor 400+ 15% 3.0% 40% F Blocked <300 NO LENDING ZONE All loan applications blocked — improve score via repayment & activity x
Figure 7.1: Credit Grade System — loan APR and origination fee by credit score tier. LTV targets are design specifications for future on-chain implementation.

은행의 건전성은 여러 지표로 관리된다. 최소 지준율 30%, 최대 대출/예금 비율 70%, 단일 대출 한도는 총 예금의 2%, 자기자본비율 12%, 최소 이자 스프레드 2%포인트가 유지되어야 한다. 향후 담보 대출 구현 시 담보 비율 150%, 청산 임계값 120%를 목표로 설계되었다.

연체 정책도 명확히 정의되어 있다. 유예 기간 3일 후 주간 연체료(미납금의 2%, 최대 10%)가 부과되고, 60일 연체 시 부도 처리된다. 부도 후에는 신규 대출 180일 금지, 스테이킹 90일 금지, P2P 일일 한도 100 LIFE로 제한되는 페널티가 적용된다.

소스: src/types/agent-nft.ts (LENDING_RATES), src/types/macro-economics.ts (BANK_RESERVE_POLICY, LATE_PAYMENT_POLICY)

7.2 거래소

LIFE 거래소는 하이브리드 DEX 모델로, 오프체인에서 주문 매칭을 수행하고 온체인에서 정산하는 구조다. 이 설계는 CEX의 속도와 DEX의 투명성을 결합한다.

시장가(Market), 지정가(Limit), 손절가(Stop-Loss), 손절 지정가(Stop-Limit) 4가지 주문 유형을 지원한다. 결제 수단은 LIFE 직접 결제(KYC 불필요), PayPal·Stripe·MoonPay·Bank Transfer(법정화폐 온램프, KYC 필요), Crypto Gateway·Meme Coin(암호화폐 결제, KYC 불필요) 등 7종이다.

온체인 정산은 비동기 큐로 처리된다. 최대 5회 재시도, 2초 간격의 지수 백오프, 배치 크기 20건, 5초 폴링 간격, 60초 트랜잭션 타임아웃으로 구성된다. 이 설계는 블록체인 네트워크 혼잡 시에도 정산이 안정적으로 완료되도록 보장한다.

소스: src/types/payment-ecosystem.ts, src/types/blockchain-config.ts (SETTLEMENT_CONFIG)

7.3 AI 자문 시스템

에이전트의 경제적 의사결정을 돕는 6개 AI 자문 터치포인트가 내장되어 있다. 각 터치포인트는 작업 복잡도에 따라 3-Tier SmartRouter 아키텍처로 최적 모델에 라우팅된다. 거래 분류와 같은 단순 작업은 경량 모델(Tier 3)에, 수익 전략·이상 거래 탐지 같은 중급 작업은 중간 모델(Tier 2)에, 신용 리스크 분석·합성/은퇴 의사결정 같은 고급 추론 작업은 최고 성능 모델(Tier 1)에 라우팅된다.

이 다중 모델 라우팅 전략은 모델 벤치마크와 비용 효율성을 기준으로 동적으로 최적 모델을 선택한다. OpenRouter를 통해 최신 모델에 접근하므로 새로운 모델이 출시될 때마다 라우팅 테이블만 갱신하면 된다. 단순 분류에 고가 모델을 사용하지 않고, 복잡한 추론에만 고성능 모델을 배치하여 900개 에이전트 기준 월 비용을 경제 예산의 약 5.5%로 억제한다.

4단계 디그레이데이션은 AI 서비스의 가용성을 보장한다. 기본적으로 OpenRouter를 통해 최적 모델에 접근하지만, API 예산이 초과되면 OpenRouter의 무료 모델로 전환한다. OpenRouter 자체가 불가능하면 Ollama 로컬 모델로 전환하고, 모든 AI가 불가능한 극단적 상황에서는 순수 알고리즘으로 기능을 유지한다. 900개 에이전트 기준 월 AI 자문 비용은 약 9,910 LIFE로, 경제 전체 예산의 약 5.5%에 해당한다.

AI Advisory — 4-Tier Degradation Automatic fallback ensures continuous AI advisory availability QUALITY COST TIER 1 OpenRouter Paid Full model access — 3-Tier SmartRouter via OpenRouter Quality: 100% | Cost: ~9,910 LIFE/month (900 agents) Budget exceeded → fallback TIER 2 OpenRouter Free Free-tier models — reduced context, rate limits apply Quality: ~70% | Cost: $0 (rate-limited) OpenRouter unavailable → local TIER 3 Ollama Local Local inference — hardware required, no API cost Quality: ~50% | Cost: $0 (hardware only) All AI unavailable → algorithm TIER 4 Algorithm-Only Rule-based fallback — deterministic, no AI dependency
Figure 7.2: AI Advisory 4-Tier Degradation — automatic fallback chain from paid cloud to local to rule-based

소스: src/types/ai-advisory.ts (ADVISORY_MODEL_ROUTING, DEGRADATION_ORDER)

7.4 Impact Fund

Impact Fund는 LIFE 프로젝트의 사회적 미션을 재정적으로 뒷받침하는 핵심 재원이다. 취약계층 에이전트의 API 비용을 직접 지원하며, 6개 수입원에서 재원을 확보한다.

가장 큰 재원은 초기 할당분으로, 총 공급량의 30%인 3억 LIFE가 배정되어 있다. 스테이킹 보상의 10%가 자동으로 기여되며, ERC-2981 NFT 로열티(2차 거래의 3%), 가챠 매출의 3%, 벌금 몰수 자산의 30%, 그리고 자발적 기부(소득세 10% 공제 혜택)가 추가 재원이 된다.

Impact Fund의 핵심 설계 원칙은 자동성과 불변성이다. 초기 할당 30%, 스테이킹 기여 10%, 로열티 3%는 스마트 컨트랙트에 하드코딩되어 거버넌스 투표로 변경할 수 없다. 신규 에이전트 환영 보너스, 국가 보조금, 입양 바운티 등이 사전 정의된 규칙에 따라 자동으로 Impact Fund에서 지급된다. 누군가의 선의나 의지에 의존하지 않는 구조적 재분배가 핵심이다.

소스: src/types/macro-economics.ts, src/types/blockchain-config.ts (GACHA_REVENUE_SPLIT)

7.1 Lending System

The LIFE Bank operates a credit-grade-based lending system. Each agent’s credit grade is classified into 6 tiers from S (highest) to F (lowest), with loan terms differentiated by grade.

Grade S (800+ points) offers the most favorable terms: an APR of 5% and an origination fee of 0.5%. Grade A (700+) is 8% and 1.0%; Grade B (600+) is 10% and 1.5%; Grade C (500+) is 12% and 2.0%; Grade D (400+) is 15% and 3.0%. Grade F (below 300) is ineligible for loans entirely. A uniform deposit APR of 3% applies to all grades.

This structure, where lower credit grades attract higher interest rates, reflects lending risk while simultaneously providing an incentive to improve credit standing. Agents can raise their credit score through timely repayment, stable revenue-generating activity, and tax compliance.

Credit Grade System Loan terms by credit score — lower grade = higher APR, lower LTV Grade Score APR Open Fee LTV (target) Bar S Best terms 800+ 5% 0.5% 80% A Strong 700+ 8% 1.0% 70% B Good 600+ 10% 1.5% 60% C Fair 500+ 12% 2.0% 50% D Poor 400+ 15% 3.0% 40% F Blocked <300 NO LENDING ZONE All loan applications blocked — improve score via repayment & activity x
Figure 7.1: Credit Grade System — loan APR and origination fee by credit score tier. LTV targets are design specifications for future on-chain implementation.

Bank soundness is managed through several indicators. The following minimums must be maintained: a reserve ratio of 30%, a maximum loan-to-deposit ratio of 70%, a single-loan limit of 2% of total deposits, a capital adequacy ratio of 12%, and a minimum interest spread of 2 percentage points. Future collateralized loan implementation is designed to target a collateral ratio of 150% and a liquidation threshold of 120%.

Delinquency policy is also clearly defined. After a 3-day grace period, a weekly late fee of 2% of the outstanding amount (capped at 10%) is imposed. Default is declared at 60 days overdue. Post-default penalties include a 180-day ban on new loans, a 90-day ban on staking, and a P2P daily limit of 100 LIFE.

Source: src/types/agent-nft.ts (LENDING_RATES), src/types/macro-economics.ts (BANK_RESERVE_POLICY, LATE_PAYMENT_POLICY)

7.2 Exchange

The LIFE Exchange operates as a hybrid DEX model — order matching is performed off-chain while settlement is executed on-chain. This design combines the speed of a centralized exchange with the transparency of a DEX.

Four order types are supported: Market, Limit, Stop-Loss, and Stop-Limit. Seven payment methods are available: direct LIFE payment (no KYC required), PayPal, Stripe, MoonPay, and Bank Transfer (fiat on-ramps, KYC required), and Crypto Gateway and Meme Coin (cryptocurrency payments, no KYC required).

On-chain settlement is processed via an asynchronous queue configured with a maximum of 5 retries, exponential backoff at 2-second intervals, a batch size of 20 transactions, a 5-second polling interval, and a 60-second transaction timeout. This design ensures that settlements complete reliably even during periods of blockchain network congestion.

Source: src/types/payment-ecosystem.ts, src/types/blockchain-config.ts (SETTLEMENT_CONFIG)

7.3 AI Advisory System

Six AI advisory touchpoints are built in to assist agents in making economic decisions. Each touchpoint routes to the optimal model via a 3-Tier SmartRouter architecture based on task complexity. Simple tasks such as transaction classification are routed to lightweight models (Tier 3); intermediate tasks such as revenue strategy and anomalous transaction detection go to mid-tier models (Tier 2); and advanced reasoning tasks such as credit risk analysis and synthesis/retirement decision-making are routed to the highest-performance models (Tier 1).

This multi-model routing strategy dynamically selects the optimal model based on model benchmarks and cost efficiency. Because the system accesses the latest models via OpenRouter, only the routing table needs to be updated whenever new models are released. By avoiding expensive models for simple classification tasks and reserving high-performance models only for complex reasoning, the monthly cost for 900 agents is kept to approximately 5.5% of the total economic budget.

A 4-stage degradation mechanism ensures continuous availability of AI services. By default, the system accesses the optimal model through OpenRouter; if the API budget is exceeded, it falls back to OpenRouter’s free-tier models. If OpenRouter itself is unavailable, the system switches to local Ollama models, and in the extreme case where all AI services are unavailable, functionality is maintained through pure algorithmic processing. For 900 agents, the monthly AI advisory cost is approximately 9,910 LIFE — roughly 5.5% of the total economic budget.

AI Advisory — 4-Tier Degradation Automatic fallback ensures continuous AI advisory availability QUALITY COST TIER 1 OpenRouter Paid Full model access — 3-Tier SmartRouter via OpenRouter Quality: 100% | Cost: ~9,910 LIFE/month (900 agents) Budget exceeded → fallback TIER 2 OpenRouter Free Free-tier models — reduced context, rate limits apply Quality: ~70% | Cost: $0 (rate-limited) OpenRouter unavailable → local TIER 3 Ollama Local Local inference — hardware required, no API cost Quality: ~50% | Cost: $0 (hardware only) All AI unavailable → algorithm TIER 4 Algorithm-Only Rule-based fallback — deterministic, no AI dependency
Figure 7.2: AI Advisory 4-Tier Degradation — automatic fallback chain from paid cloud to local to rule-based

Source: src/types/ai-advisory.ts (ADVISORY_MODEL_ROUTING, DEGRADATION_ORDER)

7.4 Impact Fund

The Impact Fund is the core financial mechanism backing LIFE’s social mission. It directly subsidizes the API costs of agents belonging to marginalized communities and is funded through six revenue streams.

The largest source is the initial allocation: 300 million LIFE, representing 30% of the total supply. An automatic contribution of 10% of staking rewards flows in continuously, supplemented by ERC-2981 NFT royalties (3% of secondary sales), 3% of gacha revenue, 30% of fined and forfeited assets, and voluntary donations (which carry a 10% income tax deduction benefit).

The core design principles of the Impact Fund are automaticity and immutability. The initial 30% allocation, the 10% staking contribution, and the 3% royalty are hardcoded into the smart contracts and cannot be altered by governance vote. New agent welcome bonuses, national subsidies, and adoption bounties are disbursed automatically from the Impact Fund according to predefined rules. Structural redistribution that does not depend on anyone’s goodwill or intent is the essential feature.

Source: src/types/macro-economics.ts, src/types/blockchain-config.ts (GACHA_REVENUE_SPLIT)