Multi-Model Agentic AI

Copyright (C) 2025, Shyamal Suhana Chandra

A comprehensive, production-ready C++ multi-agent system where each agent uses llm.c for LLM inference, maintains working memory with MDL-normalized context, ruminates on tasks using chain-of-thought reasoning, and can communicate via direct messaging.

Key Characteristics

The system implements all of the following characteristics:

• ✅ Modular: Component-based architecture
• ✅ Fault-Tolerant: Retry mechanisms, circuit breakers, graceful degradation
• ✅ Secure: Input validation, encryption, secure communication
• ✅ Atomic: Transaction support, rollback capabilities
• ✅ Concurrent: Thread-safe operations, synchronized access
• ✅ Parallel: Thread pools, parallel execution
• ✅ Distributed: Network communication, remote agents
• ✅ Cache Coherent: MESI-like protocol, distributed caching
• ✅ Encrypted: Data at rest and in transit
• ✅ Protocol-Driven: Formal message protocols
• ✅ Robust: Comprehensive error handling
• ✅ Asynchronous: Non-blocking operations
• ✅ Producer-Consumer: Bounded queues, efficient signaling
• ✅ Synchronized: Proper synchronization primitives
• ✅ Optimized: Performance optimizations
• ✅ Lightweight: Minimal overhead

Features

• Multiple Agents: Each agent has its own LLM instance and memory
• MDL-Normalized Context: Custom encoding scheme for efficient context storage
• Chain-of-Thought Reasoning: Agents ruminate on tasks with structured reasoning
• Inter-Agent Communication: Direct messaging between agents
• Configurable Trace Limits: Per-agent working memory with recursion limits
• Multiple Interfaces: CLI, config file, and C++ API

Building

Prerequisites

• C++17 compatible compiler
• CMake 3.15 or later
• Git (for submodules)

Build Steps

# Clone repository with submodules
git submodule update --init --recursive

# Build
mkdir build && cd build
cmake ..
make

# Run
./multi_agent_llm --task "keyword" --agent agent1 --config ../config/agents.yaml

Configuration

See config/agents.yaml for example agent configurations.

Usage

Command Line

# Basic usage with task and agent
./multi_agent_llm --task "research quantum computing" --agent agent1

# Using configuration file
./multi_agent_llm --config config/agents.yaml --task "analyze data"

# With specific agent from config
./multi_agent_llm --config config/agents.yaml --task "research topic" --agent research_agent

C++ API

#include "agent_manager.hpp"
#include "task_interface.hpp"
#include "reporting.hpp"

// Create agent manager
agent::AgentManager manager;

// Create agent from config
agent::AgentConfig config;
config.id = "my_agent";
config.model_path = "models/my_model.bin";
config.trace_limit = 20;
config.memory_size = 4096;
manager.createAgent(config);

// Create task API
task_interface::TaskAPI task_api(&manager);

// Submit task
std::string result = task_api.submitTask("research topic", "my_agent");

// Set up reporting
auto reporter = std::make_unique<reporting::ConsoleReporter>();
// Use reporter to output results

See examples/example_usage.cpp for a complete example.

Architecture

• Agent: Individual agent with LLM, memory, and trace management
• AgentManager: Creates and manages multiple agents
• Memory System: MDL-normalized context with trace recursion limits
• Communication: Direct messaging between agents
• Reporting: Console, file, and callback outputs

License

Copyright (C) 2025, Shyamal Suhana Chandra