🤖 AICP – AI Consensus Protocol

AICP is a developer‑first platform that transforms local LLMs into collaborative coding teams. Create model pools (e.g., "TypeScript Squad"), let them debate, vote, and produce production‑ready code – all in real time, 100% local via Ollama.

🎯 What is AICP?

Stop trusting a single AI. Instead, assemble a team of models that discuss, critique, and vote on the best solution for your coding tasks. Whether you need a REST API, a data pipeline, or a frontend component – AICP delivers a consensus‑driven answer you can actually use.

• Model Pools – group models by specialty (TypeScript, Python, Rust, etc.).
• Real‑time code debates – watch models propose, argue, and refine code.
• Live streaming – see every token, every criticism, every vote.
• Final answer with syntax highlighting – polished markdown output ready to copy.

✨ Features

⚙️ How It Works (Developer Edition)

1. Create a pool – give it a name and description (e.g., “Expert TypeScript backend developers”).
2. Select models – choose which Ollama models belong to the pool.
3. Chat with the pool – type a coding request (e.g., “Create an Express API to convert meters to centimeters”).
4. Models debate – each model proposes a solution (REASONING + CODE).
5. Argument & rebuttal rounds – models critique, defend, and refine.
6. Voting – models vote for the best code (cannot vote for themselves).
7. Synthesis – the winning model writes the final answer with a markdown code block.
8. You get the code – ready to copy, paste, and run.

All steps stream live to your terminal – you see every line of reasoning and every line of code as it’s generated.

📦 Prerequisites

• Node.js 18+ (tested with v20+)
• Ollama installed and running (ollama.ai)
• At least two models pulled, for example:

ollama pull llama3:8b
ollama pull mistral:7b
ollama pull phi3:mini

Recommendation: Use models with at least 7B parameters for best code quality. Small models (1B‑2B) may produce incomplete code.

🛠 Installation

Option 1: Homebrew (macOS – recommended)

brew tap devdanielvaldez/aicp
brew install aicp

Then run aicp from anywhere.

Option 2: From source

git clone https://github.com/devdanielvaldez/aicp-advanced.git
cd aicp-advanced
npm install
npm run build
cd packages/cli
npm run dev

Or install globally:

npm install -g ./packages/cli
aicp

🚀 Quick Start

aicp

You will see the main menu with:

• 🧩 Manage model pools – create, list, edit, delete, and chat with pools.
• 🤖 Manage models – list and select which models are available for pools.

Create your first pool

1. Choose Manage model pools → Create pool.
2. Enter a name (e.g., typescript-backend).
3. Enter a description (e.g., “Expert TypeScript backend developers for Node.js, Express, and REST APIs”).
4. Select at least two models from your installed list.
5. The system auto‑generates a system prompt.
6. Choose Chat with pool and start coding.

Example chat

> Create an Express API to convert meters to centimeters

🤖 Models are debating...

[llama3:8b] streaming:
REASONING: Simple, type‑safe solution with input validation...
CODE:
```typescript
import express from 'express';
const app = express();
app.use(express.json());
app.post('/convert', (req, res) => {
  const { meters } = req.body;
  if (typeof meters !== 'number') return res.status(400).json({ error: 'Invalid input' });
  res.json({ centimeters: meters * 100 });
});
app.listen(3000);

After debate and voting, the winning model streams the final answer with syntax highlighting.

📖 Detailed Usage

Model Pools

Chat Commands

Inside a pool chat:

• /verbose – toggle detailed debate logs.
• /exit – leave the chat.
• /help – show commands.

Configuration

All data is stored in ~/.aicp/:

• pools.json – your model pools.
• reputation.db – model reputation scores.
• config.json – global settings (selected models, etc.).

🏗 Architecture

AICP is a TypeScript monorepo with three active packages (P2P is experimental and not used in the developer edition):

aicp-advanced/
├── packages/
│   ├── core/        # Consensus, reputation, vector memory, PSO (optional)
│   ├── cli/         # Interactive CLI, debate orchestrator, streaming client, pools
│   └── api/         # REST API (stub)
└── tsconfig.base.json

The core debate engine (phases.ts) is reused for pools, but with custom prompts tailored for code generation. All communication with Ollama uses HTTP streaming (/api/chat with stream: true).

🐞 Troubleshooting

🤝 Contributing

We love contributions! See CONTRIBUTING.md.

git clone https://github.com/devdanielvaldez/aicp-advanced.git
cd aicp-advanced
npm install
npm run build
cd packages/cli
npm run dev

Then open a Pull Request.

📄 License

MIT © devdanielvaldez

🙏 Acknowledgements

• Ollama – local LLM runtime
• Inquirer.js – beautiful CLI prompts
• Highlight.js – syntax highlighting for terminal

Built with ❤️ for developers who want reliable AI code assistance.

GitHub Repository – ⭐ star and fork welcome!