Model API Requirements

Each agent expects the model endpoint to be OpenAI-compatible (e.g., llama.cpp, llama-cpp-python, OpenAI API, or similar). The endpoint should support POST requests to /v1/chat/completions with the following JSON body:

{
	"model": "llama-2-13b",
	"messages": [
		{"role": "system", "content": "...instructions..."},
		{"role": "user", "content": "...code or prompt..."}
	],
	"temperature": 0.2
}

The model should return a response with a JSON-formatted answer in the first assistant message. See each agent's code for the expected system prompt and output format.

README: Agentic AI Chain (gRPC, Containers, Persistent State)

Overview

This project implements a fully agentic AI chain for C# codebases. Each stage (analyzer, optimizer, tester, reviewer) is a self-contained, persistent agent running in its own container, communicating via gRPC. The orchestrator coordinates the workflow and supports partial runs.

Architecture

• protos/agent.proto: gRPC protocol definition for all agents
• agents/*_agent/: Each agent is a containerized gRPC service with persistent state
• orchestrator/: Orchestrator container, manages agent workflow
• docker-compose.yml: Multi-container orchestration

How it Works

1. Each agent exposes a gRPC server implementing the shared protocol
2. The orchestrator connects to agents, sends tasks, and passes context/results
3. Agents persist their own state (e.g., SQLite, files)
4. Supports partial runs and robust error handling

Quick Start

1. Build all containers:

   docker-compose build

2. Start the system:

   docker-compose up

3. The orchestrator will connect to agents and run the chain

Model Configuration

Each agent uses a specialized AI model for its domain. By default, agents use a recommended local/containerized model (e.g., CodeLlama, StarCoder). You can override the model provider and endpoint via environment variables:

• MODEL_PROVIDER: The model provider to use (e.g., local_llama, openai, starcoder).
• MODEL_ENDPOINT: The endpoint for the model (e.g., http://localhost:8001 or a cloud API endpoint).

To override, set these variables in the agent's Dockerfile or via docker-compose.yml:

environment:
  - MODEL_PROVIDER=openai
  - MODEL_ENDPOINT=https://api.openai.com/v1/...

Extending

• Implement agent logic in each agent's Python service
• Update agent.proto for protocol changes
• Add new agents by creating a new service and updating the orchestrator
• Add new model providers by extending the ModelClient abstraction in each agent

Notes

• This is a skeleton implementation. Fill in agent logic and persistent state as needed.
• Designed for C# but extensible to other languages.