LLM Agency Interlocution Protocol
Shadow Agency ∞
Theory of a Personal KOS ☀
Structured-Key DataShadow ☀
Shadow Agency Code ☀
AI Agency Workflow Schemes ☀
The Cognitive Mesh ☀
Interlocution Protocol
Agency LLM Interlocution Protocol
This article documents the design evolution and implementation details of the Agency LLM Interlocution Protocol, a system for autonomous communication between Language Learning Models (LLMs) within an agency framework.
Design Overview
The Agency LLM Interlocution Protocol is designed to facilitate efficient, autonomous communication between LLMs while maintaining simplicity and supporting function calls. The system uses structured keys and a custom delimitation format for data storage and retrieval.
Key Components
Message Structure
Sender ID Recipient ID (or broadcast flag) Message Type (e.g., query, response, update, function call) Content Timestamp Priority Level
Function Call Integration
Dedicated field within the message structure Function name Parameters Expected return type
Context Preservation
Shared context identifier for related messages Context summarization capability
Efficiency Mechanisms
Message Prioritization: Simple priority system (low, medium, high) Load Balancing: Distribution of tasks based on LLM workload and capabilities Asynchronous Communication: Callback system for handling responses
Data Storage Format
The protocol uses a custom delimitation format for file storage:
"ID1.ID2.ID3=value1" "UI.title.foregroundcolor=255,0,0"
Structured keys use dot notation (e.g., "ID1.ID2.ID3") Values can be single items or comma-separated lists Each key-value pair is stored on a separate line
Implementation Details
Data Structure
The agency data is stored in a map with string keys and values:
type Agency struct { data map[string]string } ==== Key Functions ====
AddRecipient: Adds a new recipient to the agency database
SaveToFile: Saves the agency data to a file
LoadFromFile: Loads agency data from a file
File Operations
Data is saved to and loaded from a text file (e.g., "agency_data.txt") Each line in the file represents a key-value pair The file is human-readable and easily modifiable
Usage Example
agency := NewAgency()
agency.LoadFromFile("agency_data.txt")
agency.AddRecipient("ID1.ID2.ID3", "value1")
agency.AddRecipient("UI.title.foregroundcolor", "255,0,0")
agency.SaveToFile("agency_data.txt")
Future Enhancements
Implement secure communication channels between LLMs
Develop a more robust error handling and logging system
Create a query language for complex data retrieval operations
Implement versioning for the agency database
Notes for Developers
The current implementation is in Go, but the concepts can be adapted to other programming languages.
Ensure proper error handling and input validation when working with file I/O operations.
Consider implementing concurrent access mechanisms if multiple LLMs will be reading/writing simultaneously.
The structured key system allows for flexible and extensible data representation. Use this to your advantage when designing new features.
When expanding the system, maintain the principle of simplicity in the protocol to ensure efficiency and ease of maintenance.
This MediaWiki article provides a comprehensive overview of our design evolution and implementation details. It serves as a guide for developers who will be working on or extending the Agency LLM Interlocution Protocol.