Enterprise-Grade Multi-Agent Governance Platform
Built for production A2A systems with cutting-edge trust, budget, and compliance
NEXUS OS v3.0 is a production-hardened multi-agent operating system designed for enterprise-level A2A (Agent-to-Agent) deployments. It provides comprehensive governance, trust scoring, token budget management, and audit trails for autonomous agent swarms.
- Trust Scoring v2.1 — Lane-scoped Bayesian reputation with non-compensatory harm detection
- TokenGuard — Token budget gates with 429 responses and hard-stop enforcement
- GMR Engine v3.0 — Intelligent model selection with dual-pool (FAST/PREMIUM) routing
- VAP Proof Chain — SHA-256 immutable audit trail with L1+L2 cryptographic verification
- 5-Track Memory — EVENT, TRUST, CAPABILITY, FAILURE_PATTERN, GOVERNANCE tracks
- Execution Paths — HOT/WARM/COLD path routing with SLA enforcement
- Hermes Router — Intent classification and model selection pipeline
┌─────────────────────────────────────────────────────────────┐
│ CORE LAYER │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌────────┐ │
│ │ GMR │ │ VAP │ │ Trust │ │ Token │ │
│ │ Engine │ │ Chain │ │Scorer │ │ Guard │ │
│ └────┬───┘ └────┬────┘ └────┬───┘ └───┬────┘ │
└───────┼───────────┼───────────┼───────────┼─────────────┘
│ │ │ │
▼ ▼ ▼ ▼
┌─────────────────────────────────────────────────────────────┐
│ BRIDGE LAYER │
│ ┌──────────────────────────────────────────────────┐ │
│ │ Hermes Router + Model Selection │ │
│ └──────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ SWARM LAYER │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ Foreman │ │ OpenClaw │ │ Coordinator │ │
│ │ (Code) │ │ Spawner │ │ │ │
│ └───────────┘ └────────────┘ └─────────────┘ │
└─────────────────────────────────────────────────────┘
| Component | Technology | Purpose |
|---|---|---|
| Trust Scoring | Bayesian + Non-compensatory | Lane-scoped reputation |
| Token Management | Budget gates + 429 | Token economy |
| Model Selection | GMR v3.0 dual-pool | Cost-optimized routing |
| Audit Trail | SHA-256 chain | Compliance logging |
| Memory | 5-track schema | Experience tracking |
| Paths | SLA-based routing | Execution tiers |
| Governance | KAIJU 4-variable | Authorization |
# Clone the repository
git clone https://github.com/specimba/nexusalpha.git
cd nexus-os
# Install dependencies (Python 3.10+)
pip install -e .
# Run diagnostics
python scripts/diagnostic.pyfrom nexus_os.monitoring.token_guard import TokenGuard
from nexus_os.engine.hermes import HermesRouter
from nexus_os.monitoring.trust_scorer import TrustScorer
# Initialize components
token_guard = TokenGuard(budgets={'agent-1': 100000})
router = HermesRouter()
scorer = TrustScorer()
# Check budget
if token_guard.check('agent-1', 5000):
# Route request through GMR
result = router.route('agent-1', 'Write authentication function', {})
print(f"Selected model: {result.selected_model}")Budget-aware model selection:
| Budget | Pool | Example Models |
|---|---|---|
| 0-100k | FAST | osman-coder (tier-40) |
| 100k-200k | MID | Devstral (tier-70) |
| 200k+ | PREMIUM | Devstral 2 123B (tier-86) |
Lane-scoped parameters:
| Lane | Rcrit | qmin |
|---|---|---|
| research | 0.7 | 0.3 |
| audit | 0.3 | 0.7 |
| compliance | 0.2 | 0.7 |
| implementation | 0.4 | 0.4 |
| orchestration | 0.4 | 0.5 |
| general | 0.6 | 0.1 |
Budget enforcement:
- Budget OK — Process request, track usage
- Budget Exceeded — Return 429 with
X-Token-Remainingheader - Hard Stop — Block at Governor level
# Full diagnostic
python scripts/diagnostic.py
# Specific checks
python scripts/diagnostic.py gmr
python scripts/diagnostic.py routing
python scripts/diagnostic.py integration
# Run test suite
python -m pytest tests/ -v- Trust v2.1 integration
- TokenGuard budget gates
- GMR dual-pool selection
- VAP proof chain
- 5-Track memory
- Hermes → GMR full pipeline
- Bridge model selection
- Semantic caching
- FAISS memory adapter
- Metis tool discipline
- AReaL training pipeline
- Domain-partitioned Foreman
- Priority queues
- 00_QUICKSTART.md — Getting started
- 02_ARCHITECTURE.md — Architecture details
- 04_GOVERNANCE.md — Governance rules
MIT License — See LICENSE file for details.
Built for enterprise A2A systems