core concepts

The architecture of Agent Trader is organized into several key layers, each responsible for distinct functions:

Utilizes data critical for decision-making:

Market data from Bluefin: Includes prices, orderbooks, volume, funding rates, open interest (OI), and volatility indicators.
Portfolio state: Details such as lab balances, open positions, margin requirements, and unrealized profit & loss (P&L).
Strategy config: Defines parameters like risk profiles, leverage limits, asset availability, and trade cooldown periods.

Handles the complete decision-making cycle:

Observe: Ingests current market and portfolio data.
Analyze: Evaluates features such as Exponential Moving Averages (EMAs), volatility bands, market trends, and bid-ask spreads.
Plan: The policy model decides trading factors like direction, position size, leverage, stop-loss, and target levels, including expiry and confidence measures.
Act: Issues structured trade actions based on analysis.
Evaluate: Monitors trade outcomes to inform future decisions.

Ensures effective trade implementation:

Directly connects with Bluefin APIs and contracts for order execution.
Manages order retries, partial fills, and maintains stability during infrastructural disruptions.

Guardrails: LLM model uses daily maximum loss, asset-specific maximum exposure, leverage caps, and slippage-activated shutdowns.

Enhances user interaction and visibility:

Labs dashboard: Displays the seasonal growth curve, trade logs with explanations, and current return estimates (R).
User view: Leaderboard shows Time-Weighted Average Balance (TWAB) and Leaderboard shows Baseline Points

Represents the collaborative operation combining Beep Labs and Bluefin infrastructure for robust trading activities:

Beep: Manages the agent's decision-making brain, capital allocation, and points management.
Bluefin: Facilitates on-chain perpetual markets order execution

Last updated 1 month ago