NeuroBloom.ai is building resilient infrastructure for the future of agent communication and coordination.

We believe the next evolution of AI won't be dominated by monolithic models—but by networks of agents that understand intent, cooperate intelligently, and recover gracefully.

Our open-source protocol, PACT (Protocol for Agent Collaboration & Transfer), lays the foundation for this future.

We’re not just building software—we're building trust, interoperability, and invisible systems that last.

Join us in shaping agent ecosystems that are resilient, ethical, and collaborative by design.

🧩 PACT

Protocol for Agent Collaboration & Transfer (PACT) — Building the universal, lightweight communication layer for intelligent agents.

---

🧩 PACT

---

🌍 Vision

In an increasingly agent-driven world, PACT provides a simple, open, and scalable protocol for intent translation and agent interoperability — enabling diverse AI agents, platforms, and services to collaborate seamlessly.

"Let every agent have its mind... PACT translates their intents."

Why PACT Matters?:

PACT exists to replace confusion with clarity. To transform fragmented systems into collaborative ecosystems. So builders can spend less time duct-taping intent and more time building tools that matter.

---

🌍 Current Protocol Landscape at a Glance

Protocol	Focus	Type	Owned By	Strength
MCP (Model Context Protocol)	App ↔ Model	Vertical	Anthropic	Context & Tool Enrichment
A2A (Agent2Agent)	Agent ↔ Agent	Horizontal	Google	Multi-agent Coordination
PACT (Protocol for Agent Collaboration & Transfer)	Agent ↔ Agent	Horizontal + Middleware	Vendor-neutral	Intent Translation & Interop

---

🔍 How PACT Compares to Other Protocols

Feature	PACT	MCP (Model Context Protocol)	A2A (Agent-to-Agent)
Focus	Intent translation & platform adaptation	Model-to-tool communication	Agent-to-agent collaboration
Complexity	Lightweight	Medium	Comprehensive
ML Integration	Built-in	Limited	Optional
Error Handling	Extensive	Basic	Extensive
Implementation	Simple	Complex	Complex
Use Case	Cross-platform messaging	Tool augmentation	Complex agent interactions

🚀 Quickstart

Installation

git clone https://github.com/aknbloom/pact_adapter_mvp.git
cd pact_adapter_mvp
pip install -r requirements.txt
uvicorn main:app --reload

---

Usage

Send a POST request to the /translate endpoint:

curl -X POST http://localhost:8000/translate \
  -H 'Content-Type: application/json' \
  -d '{
    "pact_version": "0.1",
    "message_id": "abc123",
    "timestamp": "2025-04-14T12:00:00Z",
    "sender": { "agent_id": "agent-A", "platform": "Dialogflow" },
    "recipient": { "agent_id": "agent-B", "platform": "Rasa" },
    "session": { "session_id": "xyz-123", "context": {} },
    "payload": {
      "intent": "check_order_status",
      "entities": { "order_id": "A123456" },
      "text": "Where is my order?"
    }
  }'

Example Response:

{
  "translated_message": {
    "intent": "order.lookup",
    "entities": {
      "order_id": "A123456"
    },
    "text": "Where is my order?"
  }
}

---

🧩 System Architecture

Architecture

PACT As a Translation Layer

The following diagram illustrates the PACT message flow:

• PACT Gateway → ML Intent Classifier → Intent Translator → Agent Router → Adapter Layer → Target Agent → Response Handler
• Resilient design with fallbacks for low-confidence intents, adapter failures, and timeouts.

Key components:

• PACT Gateway: Validates incoming message envelope format
• ML Intent Classifier: Determines intent with confidence score
• Intent Translator: Maps between different intent naming formats
• Agent Router: Selects appropriate target agent
• Adapter Layer: Converts to target platform's message format
• Target Agent: Processes the intent and generates a response
• Response Handler: Wraps response in standard PACT envelope
• Resilient Design: Built-in fallback mechanisms for low-confidence intents, timeouts, and adapter failures
• Platform-Agnostic: Works with any conversational AI platform

---

Protocol design isn't just about what works — it's about what fails gracefully and secures trust at every layer. How PACT's security and resilience loop works (e.g., Inputs → Threat Modeling → Protocol Refinement → Resilience Mechanisms)?

📦 Docker Deployment

docker build -t pact-adapter .
docker run -p 8000:8000 pact-adapter

---

🛠 Features

• FastAPI webhook endpoint /translate
• Static intent mapping (easily extendable)
• Lightweight PACT envelope format
• Ready for extension with ML intent classifiers
• Docker-ready deployment
• Postman collection for local testing

---

PACT Protocol

📍 Project Roadmap

See our development goals and priorities in the Roadmap.

---

🚀 Getting Started

New to PACT? Begin with our Quick Start Guide for a 5-minute onboarding experience.

---

📚 Tutorials

• Quick Start Guide
• Implementing Effective Fallbacks
• Advanced Capability Negotiation
• Production Deployment Guide

---

🤝 Contributing

We welcome contributions!

• Fork the repository
• Submit a PR
• Help extend PACT toward a true open communication standard

See CONTRIBUTING.md for full guidelines.

Good first issues:

• Extend adapter to support new platforms (Intercom, Zendesk)
• Add dynamic intent learning capabilities
• Enhance error and fallback handling

---

📄 License

MIT License - See LICENSE for full text.

---

📬 Connect

For ideas, discussions, or collaborations:

• GitHub Discussions coming soon!
• Contact: founders@neurobloom.ai

Together, let's build the protocol layer for agent collaboration. 🌍