Agent Swarm System

Purpose and Scope

The Agent Swarm System provides multi-agent AI consultation capabilities for cryptocurrency analysis and trading advice. This system enables users to interact with specialized AI consultants through a natural language interface, with each consultant having deep expertise in specific cryptocurrencies and access to comprehensive technical analysis tools.

This document covers the agent coordination architecture, specialized consultant implementations, and tool ecosystem. For information about the core trading logic that agents can analyze, see AI Trading Engine. For details about the mathematical analysis services that power agent tools, see Mathematical Analysis Services.

System Architecture

The Agent Swarm System operates as a multi-layered architecture with centralized coordination, intelligent routing, and specialized domain expertise.

Agent Swarm Architecture

Agent Coordination and Routing

Central Coordination

The RootSwarm serves as the central coordinator for all agent interactions, managing the flow of conversations between users and specialized consultants.

Triage Agent Implementation

The TriageAgent acts as the intelligent router that determines user intent and directs conversations to appropriate specialists.

Property	Value
Agent Name	AgentName.TriageAgent
Completion	CompletionName.HfStreamCompletion
Max Tool Calls	1
Keep Messages	10
Language	Russian

Routing Logic

The triage agent analyzes user queries to identify cryptocurrency interest and routes accordingly:

Specialized Consultant Agents

Common Architecture

All consultant agents share a unified architecture with cryptocurrency-specific specialization:

Property	Common Value
Completion	CompletionName.HfStreamCompletion
Max Tool Calls	1
Keep Messages	25
Language	Russian

Dynamic Price Integration

Each consultant agent provides real-time market context through systemDynamic functions:

Agent-Specific Implementation

Agent	Symbol	systemDynamic Function
BtcConsultantAgent	BTCUSDT	getMarketPrice("BTCUSDT")
EthConsultantAgent	ETHUSDT	getMarketPrice("ETHUSDT")
BnbConsultantAgent	BNBUSDT	getMarketPrice("BNBUSDT")
XrpConsultantAgent	XRPUSDT	getMarketPrice("XRPUSDT")
SolConsultantAgent	SOLUSDT	getMarketPrice("SOLUSDT")

Tool Ecosystem

Technical Analysis Tools

Each consultant agent has access to a comprehensive suite of technical analysis tools:

Tool Category	Tool Names	Purpose
Multi-Timeframe Signals	FetchShortTermRangeSignals FetchSwingTermRangeSignals FetchLongTermRangeSignals	15m, 30m, 1h technical analysis
Market Microstructure	FetchVolumeData FetchSlopeData FetchWhaleData	Volume, momentum, institutional activity
Social Intelligence	FetchMastodonNews FetchTwitterNews	Social sentiment analysis
Historical Context	FetchLongTermHistory FetchShortTermHistory FetchSwingTermHistory	Historical pattern analysis
Candle Analysis	FetchHourCandleHistory FetchThirtyMinuteCandleHistory FetchFifteenMinuteCandleHistory FetchOneMinuteCandleHistory	Multi-timeframe candlestick data

Tool Integration Pattern

AI Inference System

Dynamic Model Selection

The agent swarm utilizes a sophisticated AI inference system that allows runtime model selection through the InferenceMetaService.

Available AI Providers

Provider	Models	IDs
HuggingFace	openai/gpt-oss-120b deepseek-ai/DeepSeek-R1 deepseek-ai/DeepSeek-V3	0, 1, 2
xAI Grok	grok-3 grok-3-mini	3, 4
Ollama	gpt-oss:120b deepseek-r1:671b deepseek-v3:671b	5, 6, 7

Model Selection Architecture

Runtime Model Selection Logic

The InferenceMetaService.getActiveInference() method implements the selection logic:

1. Checks Boolean flags isInferenceSelection0 through isInferenceSelection7
2. Returns corresponding entry from INFERENCE_LIST
3. Falls back to default model (ID 0) if no selection is active

Integration with Trading System

Agent-Trading System Bridge

The Agent Swarm System integrates seamlessly with the core trading infrastructure, enabling agents to provide analysis using the same mathematical services that power automated trading decisions.

Shared Service Dependencies

This shared architecture ensures consistency between manual agent consultations and automated trading decisions, as both systems utilize identical technical analysis calculations and market data sources.

Error Handling and Resilience

The Agent Swarm System implements comprehensive error handling through callback mechanisms:

Tool Error Management

• onToolError callbacks log errors to ./tool_error.txt
• Graceful degradation for unavailable data sources
• Social media tools return "unavailable" status on errors
• Math services return "N/A" for calculation errors

Navigation Error Recovery

All consultant agents include NavigateToTriageTool for returning to the central router when:

• User mentions different cryptocurrencies
• Agent cannot handle specific requests
• Cross-agent navigation is required