Data Agent Pool

Overview

The Data Agent Pool forms the sensory and contextual foundation of the FinAgent Orchestration framework. It is responsible for the autonomous ingestion, normalization, and contextualization of heterogeneous financial data streams, enabling downstream reasoning agents to access structured and trusted information.

Design Objectives

  • Ensure modular access to diverse financial data sources.

  • Enable composable data pipelines for transformation and alignment.

  • Maintain provenance and auditability of all data-fetching operations.

  • Support protocol-driven interoperability and agent reuse.

Agent Specialization

Each agent within the pool is designed to specialize in a specific data domain, with its own tools, rate limits, and error handling logic. Examples include:

  • YFinanceAgent: Historical OHLCV data, fundamental indicators.

  • PolygonAgent: Real-time ticks, market depth, and high-frequency snapshots.

  • NewsAgent: Live news streams, event detection, and sentiment tagging.

  • EconAgent: Macroeconomic indicators, forecasts, and calendar-based data.

Architecture and Protocol

  • Communication: All agents expose callable interfaces via the Multi-agent Control Protocol (MCP) and can optionally participate in A2A message exchange for low-latency synchronization (e.g., context propagation).

  • Execution: The Orchestrator composes data retrieval DAGs using tools declared by each agent.

  • Memory Integration: Data responses are versioned, timestamped, and archived into the memory subsystem for future replay or training tasks.

Design Principles

  • Separation of Concerns: Each agent only handles its own data source, avoiding logic entanglement.

  • Context-Aware Pipelining: Agents can be chained to produce layered outputs (e.g., raw → filtered → aligned).

  • Redundancy and Voting: When multiple sources exist for the same signal, the system uses ranking or consensus rules to resolve discrepancies.