Agent Architecture and Setup
This document explains the configuration and architecture of the AI agent system built on the agent-swarm-kit framework. It covers the Redis-based persistence layer, agent registration patterns, and session management infrastructure that enables the cryptocurrency consultation agents to operate.
For information about specific agent implementations and their capabilities, see Cryptocurrency Consultant Agents. For details about agent routing and navigation between different specialists, see Agent Navigation and Routing.
Framework Configuration and Setup
The agent system is built on the agent-swarm-kit framework, configured through the main setup file with Redis persistence and logging infrastructure.
Core Framework Configuration
The system initializes with specific configuration parameters and custom adapters:
Chat Session Configuration
The ChatInstance is extended with single-run protection to prevent concurrent message processing:
| Component | Configuration | Purpose |
|---|---|---|
Chat.useChatAdapter |
singlerun(sendMessage) |
Prevents concurrent message processing |
swarm.loggerService |
pinolog logger |
Centralized logging for agent operations |
Logger.useClientAdapter |
Per-client file logging | Individual chat session logs |
Persistence Architecture
The system uses Redis as the primary persistence layer with multiple specialized adapters for different data types.
Redis Persistence Adapters
TTL Configuration
Different persistence layers have specific time-to-live configurations:
| Adapter | TTL | Purpose |
|---|---|---|
PersistEmbedding |
604800s (1 week) | Cache AI embeddings |
History |
900s (15 minutes) | Chat message history |
PersistAlive |
600s (10 minutes) | Client online status |
Agent Registration and Configuration
Agents are registered using the addAgent function with specific configuration objects that define their behavior, tools, and dependencies.
Agent Configuration Structure
Agent Callback System
Agents can define callback functions for error handling and lifecycle events:
| Callback | Purpose | Implementation |
|---|---|---|
onToolError |
Handle tool execution errors | Logs errors to file with client context |
onResurrect |
Handle agent resurrection | Sends notification to Telegram users |
Chat Session Management
The system manages chat sessions through the Chat API with support for multiple client types and swarm configurations.
Session Lifecycle
Integration Points
The agent system integrates with multiple interfaces:
| Interface | Client ID Pattern | Event Handling |
|---|---|---|
| WebSocket | web-{chatId} |
Token streaming, completion events |
| Telegram | telegram-{chatId} |
Message formatting, HTML conversion |
Error Handling and Logging
The system includes comprehensive error handling and logging:
