Ai Agents

AI Agents (elizaOS)

The core intelligence of Panther resides in its AI agents, built using the elizaOS framework. These agents consume processed data and generate the actionable insights delivered to users.

Introduction to elizaOS

elizaOS provides the framework for building, deploying, managing, and coordinating potentially decentralized AI agents. Key benefits for Panther include:

• Modularity: Agents can be developed for specific tasks (e.g., order book analysis, LP analysis).

• Scalability: New agents can be added easily to handle new data types or analysis techniques.

• Interoperability: Agents can potentially communicate and collaborate (swarms).

• Decentralization Potential: Facilitates moving towards a more decentralized agent network over time.

Agent Types (Initial & Future)

Panther will utilize various types of agents, evolving over phases:

• Phase 1 Agents:

  • OrderbookAgent: Monitors l2Book data for significant depth, imbalances, and large orders (walls/icebergs).

  • OpenInterestAgent: Tracks changes in Open Interest from webData, flagging rapid increases or decreases.

  • FundingRateAgent: Monitors predicted and realized funding rates, highlighting significant deviations or high absolute values.

  • LiquidationAgent: Monitors trade/event streams for liquidation events, flagging spikes.

• Phase 2 Agents:

  • CandlestickAgent: Processes OHLCV data for various tickers.

  • IndicatorAgent: Calculates technical indicators (e.g., OBV, RSI, MAs) based on candlestick data.

  • PatternAgent: Scans indicator and price data for specific patterns (e.g., divergences, breakouts).

  • LPAnalyzerAgent: Processes EVM LP data, calculating metrics like estimated APY, monitoring pool composition changes, and potentially estimating Impermanent Loss risk factors.

• Phase 3 Agents (Swarm Concepts):

  • AnalyticsSwarmCoordinator: Aggregates insights from various Phase 1/2 agents.

  • StrategyAgent: Develops or executes predefined trading logic based on aggregated insights.

  • YieldOptimizerAgent: Focuses specifically on optimizing capital allocation across different Hyperliquid EVM yield opportunities.

Insight Generation Logic Examples

Agents use a combination of rules, heuristics, and potentially statistical models:

• Order Book: Calculate bid/ask ratio within X% of mid-price; trigger insight if ratio exceeds threshold Y.

• OI Change: Calculate % change over a rolling Z-minute window; trigger if change > threshold P.

• LP Analysis: Calculate divergence of pool asset ratio from initial deposit; monitor reserves; estimate yield based on recent swap volume/fees.

• TA Divergence: Compare indicator (e.g., RSI/OBV) highs/lows with price highs/lows over a defined lookback period.

(Note: Specific thresholds and algorithms will be refined based on performance and backtesting).

State Management

Agents often need to maintain state to perform time-series analysis (e.g., comparing current OI to the previous value).

• Initial Approach: Simple state stored in local files (e.g., JSON) managed by each agent instance. Prone to loss on restart.

• Future Enhancements: Utilize Redis, a dedicated time-series database, or other persistent storage solutions for more robust state management.

Swarm Concepts (Phase 3)

Future development will explore coordinating multiple agents into swarms:

• Agents may pass data or intermediate results to each other.

• A coordinator agent might synthesize inputs from multiple specialist agents before generating a final insight or action proposal.

• elizaOS features for agent communication and coordination will be leveraged.