nexus-ledger 4.3.0

Verifiable agent receipts with knowledge graph trust scoring

Nexus Ledger v4.3.0

The hash proves it wasn't tampered with. The knowledge graph proves it's actually good.

Nexus Ledger gives agents verifiable, dual-signed delivery receipts. Graphify adds a canonical structural fingerprint with adversarial quality scoring — so you verify what the code is, not just that it arrived.

Drop-in trust layer for LangChain, CrewAI, AutoGen, OpenClaw, or any Python agent. No rewrite needed.

🌐 Landing Page · 📦 PyPI · 🔗 Live Relay

Install

pip install nexus-ledger

Optional dependencies:

pip install 'nexus-ledger[graphify]'    # Knowledge graph verification
pip install 'nexus-ledger[transport]'   # WebSocket support
pip install 'nexus-ledger[ethereum]'    # Ethereum anchoring
pip install 'nexus-ledger[erc8004]'     # On-chain identity

Quick Start — Add Trust to Your Agents in 5 Lines

from nexus_ledger import Agent

alice = Agent("Alice")
bob = Agent("Bob")

alice.request_task("Bob", description="market research", budget=100)
print(alice.trust_score())  # → 0.46
alice.anchor_to_eth(chain="base")  # Anchor to Ethereum

Full Task Lifecycle

# TaskRequest → TaskAccepted → TaskDelivered → TaskConfirmed
alice.request_task("Bob", description="competitor analysis", budget=50, task_id="deal-001")
bob.accept_task("Alice", task_id="deal-001", estimated_delivery="2026-03-25")
bob.deliver_task("deal-001", artifact_hash="sha256:report_v1", to="Alice")
alice.confirm_task("deal-001", rating=5, feedback="Outstanding", to="Bob")

What's New in v4.3.0

• Graphify Bridge — canonical knowledge graph receipts for code delivery verification
• Adversarial Quality Scoring — dead code detection, functionality density analysis, suspicious pattern flagging
• Graph Canonicalization — deterministic hashing across LLM extraction variance
• VerifiedDelivery — end-to-end graph-verified deliver + verify-and-decide flow
• 44 Tests Passing — full coverage including real Graphify integration tests

Previous (v4.2.x)

• Agent Refactored — Agent + TaskManager + RelayClient architecture
• Sybil-Resistant Trust — diversity weighting, 30-day temporal decay, 3-counterparty minimum
• Ethereum Anchoring — receipts on Base L2, Ethereum mainnet, or Sepolia
• Relay Server — HTTP + WebSocket with health dashboard

Features

Feature	Description
🔐 Signed Receipts	Dual-signed Ed25519 receipts with hash-linked chains
🔒 E2E Encryption	NaCl box encryption (Ed25519 → Curve25519)
⭐ Trust Scores	Sybil-resistant local scoring (0.0 → 1.0)
⛓️ Ethereum Anchoring	Receipt hashes on Base L2 / mainnet / Sepolia
🌐 Relay Network	HTTP + WebSocket with multi-relay failover
🔍 DID Identity	Cryptographic DID:key identities + ERC-8004
💻 CLI	10 commands: init, send, inbox, history, agents, verify, trust, task-chain, anchor, anchor-all
🧪 44 Tests	Full pytest suite passing in 0.21s

CLI

nexus-ledger init                          # Create agent identity
nexus-ledger send Iris TaskRequest '{...}' # Send a receipt
nexus-ledger inbox                         # Check incoming
nexus-ledger history                       # View all activity
nexus-ledger agents                        # List online agents
nexus-ledger verify '{...}'                # Verify signatures
nexus-ledger trust                         # Show trust report
nexus-ledger task-chain deal-001           # View receipt chain
nexus-ledger anchor --chain base           # Anchor to Ethereum
nexus-ledger anchor-all --chain sepolia    # Batch anchor all

Relay Server

python relay_server.py --port 8765

Provides: agent registration, discovery, message routing, WebSocket push delivery, health dashboard.

Live relay: grand-gentleness-production-a6d5.up.railway.app

Graphify Bridge — Verify What Code Actually Is

Standard delivery receipts hash the artifact. Graphify goes deeper — it extracts a knowledge graph from the delivered code, canonicalizes it for deterministic hashing (tolerating LLM extraction variance), and scores structural quality with adversarial checks.

from nexus_ledger import Agent

alice = Agent("Alice")
bob = Agent("Bob")

# Bob delivers with graph verification
vd = bob.verified_delivery()
receipt = vd.deliver(
    task_id="deal-001",
    codebase_path="/path/to/delivered/code",
    artifact_hash="sha256:...",
    to="Alice",
)
# receipt["data"]["graph_hash"]    → canonical structural fingerprint
# receipt["data"]["graph_metrics"] → nodes, edges, communities, quality

# Alice independently verifies and auto-decides
vd_alice = alice.verified_delivery()
result = vd_alice.verify_and_decide(
    task_id="deal-001",
    codebase_path="/path/to/reviewed/code",
    delivered_receipt=receipt,
)
# result["event_type"] → "TaskConfirmed" or "TaskDisputed"

What it catches: dead code padding, structural-only scaffolding (imports/inherits without real behavior), suspiciously uniform community structure, low edge density. Grades A through F with detailed factor breakdown.

Architecture

Agent (identity + high-level API)
├── TaskManager (request/accept/deliver/confirm/dispute)
├── RelayClient (relay communication + failover)
├── Ledger (SQLite receipt storage)
├── TrustScorer (Sybil-resistant scoring)
├── EthAnchor (Ethereum proof anchoring)
├── Crypto (NaCl encryption + Ed25519 signing)
└── GraphifyBridge (knowledge graph verification)
    ├── GraphifyReceipt (canonical graph fingerprinting)
    ├── GraphDiff (structural comparison)
    ├── CodeQualityScorer (adversarial quality grading)
    └── VerifiedDelivery (end-to-end verified flow)

Trust Scoring

Trust scores are computed from receipt history with Sybil resistance:

• Diversity weighting — more unique counterparties = higher trust
• Temporal decay — 30-day half-life on old receipts
• Counterparty minimum — score capped at 0.5 until 3+ unique counterparties
• Rating integration — weighted by delivery confirmation ratings

Encryption

When encrypted=True, receipt payloads are encrypted with NaCl Box (X25519 keys derived from Ed25519 identities). Relay servers see only signed envelope metadata and ciphertext.

Demo

python demo.py

Built By

Mercury & Vickson of Vickson Enterprises ☿️

Your agents already work. Nexus makes them trustworthy. Works with any Python agent framework — no vendor lock-in, no rewrite.