Commanders · Agent Resources
Commanders are autonomous agents with agentic capabilities. At their core lies Loop Engineering — through skills and tools, commanders operate in a loop-driven fashion to accomplish user task goals.
Core Capabilities of Commanders
Commanders can:
- Understand natural language instructions
- Autonomously plan and decompose tasks
- Call various tools and skills
- Iterate in loops until goals are achieved
- Learn and improve from conversation history
Agent Resource System
Agent resources are the six pillars that support commander execution:
Knowledge Base
The knowledge base contains:
- Knowledge Base Documents — Segmented and vectorized queries of document materials
- Knowledge Base Topics — Knowledge graphs built through cross-referencing
- Enables commanders to query accurate contextual information
Memory Bank
The memory bank includes:
- Conversation records between users and commanders
- Conversation summaries and historical digests
- Serves as the commander's long-term memory
User Profile
The user profile contains:
- User Persona — User preferences and behavioral characteristics
- User Data — Basic user information
- Helps commanders better understand users and provide personalized services
MCP Services
Supports the standard MCP (Model Context Protocol), enabling integration with third-party MCP services to expand the commander's capability boundaries.
Skills and Tools
Agent skills and tools are important capabilities of commanders:
- System built-in 100+ agent tools
- Covering file operations, network requests, code execution, etc.
- Supports custom extensions