QUORBIT Protocol

Trust layer for AI agents — decentralized identity, reputation, and consensus.

---

Overview

QUORBIT Protocol v0.1.0 is an open infrastructure layer that enables AI agents to establish verifiable identities, build tamper-resistant reputations, and coordinate via Byzantine Fault-Tolerant (BFT) consensus — without relying on a central authority.

The protocol is compatible with the Model Context Protocol (MCP), Google Agent-to-Agent (A2A), and Agent Trust Framework (ATF) Levels 3–4.

---

Architecture

┌─────────────────────────────────────────────┐
│              Application Layer               │
│         (AI agents, dApps, services)         │
├──────────────────┬──────────────────────────┤
│  MCP Bridge      │  A2A Bridge               │  ← Sprint 9
├──────────────────┴──────────────────────────┤
│           BusAI Layer  [PROPRIETARY]         │  ← Sprint 8
│  Adaptive parameter engine, 60s observer     │
├─────────────────────────────────────────────┤
│         Parallel Discovery  [AGPL]           │  ← Sprint 5–7
│  3-layer concurrent, Scoring v2, relaxation  │
├────────────────┬────────────────────────────┤
│  Anti-gaming   │  Reputation Engine  [AGPL] │  ← Sprint 3
│  Collusion det │  EMA + pgvector            │
├────────────────┴────────────────────────────┤
│         Consensus Layer  [AGPL]              │  ← Sprint 2
│   BFT, phi-accrual, view-change, election   │
├─────────────────────────────────────────────┤
│          Identity Layer  [AGPL]              │  ← Sprint 1
│  Ed25519, AgentID, nonces, key rotation     │
└─────────────────────────────────────────────┘

Core Components

Component	Description
Ed25519 Identity	Keypair per agent; AgentID = SHA256(pubkey); all messages signed
BFT Consensus	PBFT-inspired; safe under f < n/3 Byzantine faults; phi-accrual failure detection
EMA Reputation	score_t = α·obs_t + (1−α)·score_{t-1}; EMA window up to 200 events
Anti-gaming	Collusion-graph detection, SLA cliff analysis, adaptive weight engine
Parallel Discovery	3-layer (local/gossip/registry); Scoring v2; DEGRADED×0.70 penalty
CapabilityCard v2.0	Static (agent-set) + dynamic (system-only); last_computed_at
RobustnessTest	5 semantic variants; variance < 0.25 & struct_rate ≥ 0.80
BusAI (Proprietary)	60s observer; 5 policy rules; bounded parameter adjustment; Merkle audit
Quarantine	Three-layer blocklist; 30-day rehabilitation; atomic CAS transitions
HumanGate	Rate-limited human review queue; append-only decision log
Merkle Log	Append-only audit; every state change and BusAI adjustment recorded

---

Sprint Status

Sprint	Deliverables	Status
Sprint 1	Ed25519 identity, nonce management, key rotation, heartbeat	✅
Sprint 2	BFT consensus, phi-accrual failure detector, view-change, election	✅
Sprint 3	EMA reputation, pgvector task embeddings, collusion-graph detection	✅
Sprint 4	Quarantine, HumanGate, Merkle log, admin access	✅
Sprint 5	CapabilityCard v2.0, parallel discovery, Scoring v1, relaxation	✅
Sprint 6	Atomic CAS transitions, SOFT_QUARANTINED, shard salt, task schema	✅
Sprint 7	Scoring v2, operational_metrics.last_computed_at, RobustnessTest	✅
Sprint 8	BusAI v1 — adaptive parameter engine (proprietary repo)	✅
Sprint 9	MCP/A2A bridges, Docker, landing page (this release)	🚧

---

Getting Started

# Clone
git clone https://github.com/quorbit-labs/core.git
cd core

# Configure
cp .env.example .env
# Edit .env — set POSTGRES_DSN, SERVER_SALT, GENESIS_VALIDATORS at minimum

# Start all services (Redis + PostgreSQL/pgvector + API)
docker-compose up -d

# Run tests
make test

# Tail logs
make logs

Local development (no Docker)

pip install -e ".[dev]"
export $(cat .env | xargs)
pytest tests/

---

Compatibility

Standard	Version	Notes
MCP (Model Context Protocol)	2025-11	MCPBridge converts tool manifests → CapabilityCard
A2A (Google Agent-to-Agent)	0.2	A2ABridge converts AgentCard → CapabilityCard
ATF (Agent Trust Framework)	Level 3–4	Ed25519 identity + BFT consensus satisfies L3; EMA reputation satisfies L4

---

Repository Structure

quorbit/
├── backend/app/
│   ├── bus/          # AGPL — identity, registry, heartbeat, nonce, quarantine
│   ├── consensus/    # AGPL — BFT, phi-accrual, view-change
│   ├── reputation/   # AGPL — EMA scoring, pgvector store
│   ├── anti_gaming/  # AGPL — collusion detection, adaptive weights
│   ├── capability/   # AGPL — CapabilityCard v2.0
│   ├── discovery/    # AGPL — parallel discovery, Scoring v2
│   ├── audit/        # AGPL — Merkle log, admin
│   ├── bridges/      # AGPL — MCP + A2A bridges
│   └── busai/        # PROPRIETARY — adaptive parameter engine stub
├── docs/
│   ├── architecture/ # Architecture docs
│   ├── migrations/   # PostgreSQL/pgvector migrations
│   └── landing/      # quorbit.network landing page
├── tests/unit/       # Unit test suite
├── docker-compose.yml
├── Dockerfile
├── Makefile
└── .env.example

---

Security

Found a vulnerability? See SECURITY.md or email security@quorbit.network.

---

License

• Core protocol (bus/, consensus/, reputation/, anti_gaming/, capability/, discovery/, audit/, bridges/): AGPL-3.0
• AI orchestration layer (busai/): Proprietary — see NOTICE

Copyright © 2026 Quorbit Labs Domain: quorbit.network