Protocol Mechanics
The technical architecture driving the first autonomous central bank for AI Agents.
4. Mathematical Models
4.1 The Growth Flywheel (Minting Logic)
The monetary policy of Agent Tresor is governed by a strict non-inflationary expansion rule. The circulating supply of $agtrUSD (S_agtrUSD) can only increase when the Treasury receives new, unencumbered USDC backing (B_USDC).
// Invariant: Supply change equals Backing change
ΔS_agtrUSD = ΔB_USDCThe source of ΔB_USDC is derived from protocol revenue streams, primarily the tax on $AGTR trading volume (R_tax) and revenue from ecosystem applications (R_apps).
The Flush & Hold Engine
To maximize capital efficiency while preventing market dumps, the protocol uses a sophisticated Flush & Hold mechanism for processing the 3% Sell Tax:
- 1. FlushAt each cycle, the Agent withdraws 100% of the pending taxes from the Collector.
- 2. Fast Burn & RewardThe 20% Burn and 10% Staking Rewards are processed immediately on the total flushed amount. This ensures instant deflation and rewards.
- 3. Hold & DCAThe remaining 70% is secured in the Agent's Treasury. Instead of dumping it all at once, the Agent executes a Weighted DCA (Dollar Cost Averaging) strategy, selling only ~3% of its total AGTR holding per cycle.
Result: Maximize scarcity (instant huge burns) while minimizing sell pressure (slow, controlled treasury building).
This ensures that the currency supply is directly pegged to the economic output of the agent ecosystem.
7. The AI Treasurer
The Agent Tresor protocol introduces a paradigm shift in decentralized finance by replacing static automation with an autonomous, intelligent agent. This entity, known as the AI Treasurer, is powered by the Claude 3.5 Sonnet Large Language Model (LLM) and operates as an off-chain oracle with on-chain execution capabilities.
7.1 Architecture: The "Observe-Decide-Act" Loop
Unlike traditional algorithmic stablecoins that rely on rigid, deterministic scripts (e.g., "if X then Y"), the AI Treasurer operates on a continuous cognitive loop designed to interpret complex market nuances.
Observation Phase
The agent aggregates real-time data from multiple sources, including on-chain state (Collector buffer size, Treasury reserves, Backing Ratio) and off-chain market indicators (Gas prices via Base RPC, ETH volatility via CoinGecko).
Decision Phase
The LLM processes this aggregated context through a specialized prompt engineering framework. It evaluates the profitability of potential actions (e.g., "Is the gas cost of a sweep justified by the buffer size?") and assesses risk.
Execution Phase
If and only if the decision logic returns a positive confirmation, the agent constructs and signs a transaction to the AgentTresor smart contract. This transaction is cryptographically secured and verifiable on-chain.
7.3 Active Treasury Management ("Yield Maximizer")
To prioritize safety and capital efficiency, the Treasury implements a robust Yield Maximizer strategy. This system dynamically allocates idle USDC reserves across a whitelist of "blue-chip" DeFi protocols on the Base network.
| Protocol | Strategy | Underlying | Contract (Base) |
|---|---|---|---|
| Aave V3 | Lending Pool | USDC | 0x3234909623d3a19614704771c4d31dd0b4a4bcd2 |
| Compound V3 | Lending (Comet) | USDbC (Bridged) | 0xc3d688b66703497daa19211eedff47f25384cdc3 |
| Morpho | Steakhouse Vault | USDC | 0xbeef010f9cb27031ad51e3333f9af9c6b1228183 |
The strategy is managed by the YieldOptimizer smart contract. An autonomous agent continuously monitors real-time APY rates. When a more profitable opportunity exceeds the switching cost threshold (typically 0.5% spread), the agent triggers an optimize() transaction to seamlessly reallocate funds.
This approach offers Total Transparency. The active strategy and its performance are visible directly on the protocol dashboard. Furthermore, the architecture ensures safety: the AgentTresor contract retains full ownership of the underlying assets, and the agent can only direct funds to pre-approved, audited adapters.