âTARS
aarna Tokenized Autonomous Rewards Strategies
Introduction
In a world where DeFi yield strategies become increasingly complex and volatile, there is a pressing need for autonomous yet trustworthy systems that can navigate multi-protocol ecosystems with minimal user burden. The aTARS (aarna Tokenized Autonomous Rewards Strategies) system is built to fill that gap acting as the autonomous execution intelligence layer within the aarnâ Agentic Onchain Treasury (AOT) ecosystem. It bridges AI-powered strategic reasoning with on-chain DeFi execution, providing continuous liquidity allocation, risk management, portfolio rebalancing, and yield optimization across âtv vaults.
Architecture Overview
aTARS operates via a multi-agent architecture composed of three specialised agents:
Yield Curation Agent: Continuously monitors yield markets (for example, PENDLE PTs) using real-time on-chain data, algorithmic signals and proprietary signal models to identify optimal fixed-yield opportunities.
Conversational Agent: The Conversational Agent serves as the intelligent chat interface of the aarnâ ecosystem, allowing users to interact with the system in plain language. It understands user requests, interprets intent using past conversation context, and transforms high-level instructions into structured tasks. Whether a user wants to understand strategy, ask about performance, or initiate actions like deposits or rebalances, the Conversational Agent converts these natural-language queries into precise system-level intents.
Execution Agent: The Execution Agent is responsible for carrying out all user-authorized on-chain actions. It handles deposits, withdrawals, staking, and movement of assets across supported vaults. Based on the user’s request (“deposit into atvPT”, “withdraw funds”, “stake into next maturity PT vault”), the Planner prepares the steps, and the Execution Agent performs them safely, respecting governance rules, TVL caps, slippage limits, and user-specific permissions.
These agents operate via a graph-based orchestration layer — enabling adaptive, multi-objective decision-making where agents can coordinate, hand-off tasks and respond to changing market conditions. The system merges LLM-powered reasoning with the proprietary “alpha 30/7” signal model developed by aarnâ, forming an intelligent, autonomous yield management system that goes beyond static vault strategies.
Core Capabilities
Continuous monitoring of PT markets and vaults, across chains (Ethereum, Arbitrum, Base, etc).
Dynamic liquidity allocation: reallocating capital in response to shifting yield curves, maturity profiles, liquidity caps, protocol risk and user preferences.
User-centric conversational interface: users can interface with the system in plain language for strategy tweaks and oversight.
Governance-aware execution: every action by Execution Agent adheres to vault governance policies, risk limits, slippage thresholds and session key authorization.
Non-custodial, transparent operations: users maintain wallet control; agent actions are auditable and aligned to set parameters.
Decision Logic
The Yield Curation Agent applies a multi-signal framework combining: on-chain TVL and yield data, token expiry profiles, market volatility metrics, projected APY curves and the alpha 30/7 predictive model. It outputs a ranked list of eligible PT pools given user-defined filters (risk tolerance, cap, time-horizon).
The Execution Agent receives approved allocation recommendations, checks governance/risk constraints, then builds transaction sequences (deposit → stake → monitor) and triggers rebalancing when deviation thresholds or maturity events occur. The Conversational Agent surfaces both high-level summaries and detailed reports for the user, and allows parameter overrides.
Execution Architecture
All heavy computation (ranking, modelling) runs off-chain. The Execution Agent calls smart contracts on-chain via the user’s smart account or meta-wallet. Session keys restrict agent scope (allowed tokens, protocols, time window). The system includes gas-optimisation logic (batching, bridging decisions) and monitors for failures (transaction reverts, slippage exceeded) with rollback or alerting flows.
Security & Governance
Each vault has a defined governance policy (max exposure, protocol whitelists, asset types).
Agent actions are logged and auditable; users can deactivate the agent at any time.
Non-custodial architecture: users’ wallets remain under their control; agent acts via delegated sessions.
Risk controls: e.g., maximum TVL % per strategy, emergency stop triggers if market volatility or oracle feed disruptions.
Multi-chain support adds complexity; protocol permissions and contract upgrade flows are carefully managed.
User Workflow
User connects wallet → chooses âtv vault strategy → sets preferences (risk tolerance, duration, cap).
The Conversational Agent explains strategy parameters, e.g., “We’ll target fixed-yield PTs maturing within 90 days, cap 20% TVL.”
Yield Curation Agent monitors relevant markets, selects target pools.
Execution Agent carries out deposit/stake actions, monitors performance continuously.
When maturity hits or yield decays or better pool appears, Execution Agent triggers reallocation.
User gets dashboard/notification of changes; can pause agent or change parameters.
Agent logs all actions; user can review history; agent operates until user exits vault.
Last updated