Open Source Research Project

Solving the
AI Governance
Problem.

As AI agents become autonomous, the lack of governance frameworks creates systemic risk. NEXUS is an open-source platform that brings NIST AI RMF compliance to multi-agent orchestration — making AI systems auditable, accountable, and safe.

0
AI Agents
0
Tools Built
0
Policies
0
LLM Backends
nexus-bridge ~ v3.7

Built on established standards & frameworks

NIST AI RMFISO 42001TOPSIS/AHPA2A ProtocolMCP StandardOpenAPI 3.1
The Problem

AI agents are scaling.
Governance is not.

Multi-agent AI systems are being deployed without standardized governance, creating systemic risks in decision-making, accountability, and compliance. This is a critical gap the U.S. needs addressed.

Current State

Today's multi-agent frameworks lack fundamental governance:

  • No standardized risk assessment for agent decisions
  • No policy enforcement at the orchestration layer
  • No audit trails for automated agent actions
  • No cost controls — agents spend without limits
  • No quality scoring or self-correction mechanisms
  • Single-vendor lock-in with no portability

What NEXUS Solves

NEXUS implements governance-first multi-agent orchestration:

  • NIST AI RMF policies enforced at every dispatch
  • 22 Policy Cards defining risk thresholds and autonomy
  • Ed25519 crypto-signed audit trails
  • Cost cascade from $0 to cloud — controlled by policy
  • Automated quality scoring with RSI self-improvement
  • 8 interchangeable backends — zero vendor lock-in
Technical Contributions

Novel approaches to
AI agent governance.

NEXUS introduces several original contributions to the field of multi-agent AI systems, addressing gaps not covered by existing frameworks.

Declarative Agent Specification System

A novel JSON-based agent manifest schema defining capabilities, constraints, cost tiers, quality thresholds, and behavioral policies. Each of the 12 agents is fully described in a machine-readable spec that enables deterministic routing and automated governance validation.

12
Agent Specs
22
Policy Cards
11
SOP Graphs

MCDM-Based Routing

Multi-Criteria Decision Making (TOPSIS/AHP) applied to AI agent selection. Tasks are scored against agent capabilities to select the optimal executor — a deterministic, auditable process.

Governance-as-Code

NIST AI RMF policies encoded as JSON Policy Cards. Risk thresholds, autonomy levels, and escalation rules enforced automatically at the orchestration layer — not bolted on after deployment.

RSI Quality Loops

Recursive Self-Improvement: every agent output is quality-scored via automated evaluation. Below-threshold results trigger context-enriched re-dispatch. The system measurably improves over time.

Universal Bridge Architecture

A modular relay (9 modules) that abstracts 8 LLM backends behind a unified API. Synaptic packets carry context between agents. Stigmergic traces enable shared memory. Zero vendor lock-in.

Project Scale

Built from scratch. Production-grade.

Every component was designed, engineered, and deployed by a single developer — demonstrating deep expertise in AI systems architecture, DevOps, and governance.

12
AI Agents
9 dispatchers + 3 special ops, each with unique specs
107
Custom Tools
Across 20 MCP servers organized by department
22
Governance Policies
NIST AI RMF aligned, version-controlled
99%
Health Score
99 PASS / 0 FAIL across automated health checks
8
LLM Backends
From local Ollama ($0) to cloud APIs — interchangeable
5
Audit Systems
Automated forensic audits with Ed25519 signatures
1,386
RAG Chunks
Knowledge base with LanceDB + MiniLM embeddings
4
Sprints Done
Foundation → Quality → Interop → Packaging complete
About the Founder

Built by one engineer.
Enterprise-grade output.

JDCM
Founder & Architect

Juan David Cardona Mera

AI Systems Architect & Full-Stack Engineer

Juan David designed and built the entire NEXUS platform — from the multi-agent orchestration bridge to the NIST-aligned governance framework, 20 MCP tool servers, and automated quality assurance pipeline. The project demonstrates deep expertise spanning AI/ML infrastructure, distributed systems, DevOps, and enterprise compliance frameworks.

NEXUS addresses a critical gap in the U.S. AI ecosystem: as autonomous agents proliferate, no existing framework provides standardized governance at the orchestration layer. This project contributes original solutions to a problem of national importance — making AI systems auditable, accountable, and safe for enterprise deployment.

Full-stack AI architecture
NIST AI RMF implementation
Multi-model orchestration
DevOps & infrastructure
Open-source contribution
Enterprise governance design
System Architecture

How NEXUS processes decisions.

Every request passes through a governance-validated pipeline: authentication, policy enforcement, MCDM routing, dispatch, and quality verification.

Request
Bridge v3.7
Governance
MCDM Router
Agent
Quality Check

The Agent Swarm — 12 Specialized Agents

NEXUS Orchestrator
ATLAS Research
FORGE Code Gen
BOLT Rapid Dev
ORACLE Analysis
SHIELD Security
JUDGE Code Review
SCOUT SWE Agent
SPARK Frontend
GHOST Local LLM
ECHO Multimedia
SENTINEL Fan-Out
Competitive Analysis

Addressing gaps in existing frameworks.

NEXUS fills critical governance gaps that no current multi-agent framework addresses.

Governance CapabilityCrewAILangGraphAutoGenNEXUS
NIST AI RMF ComplianceNoNoNoBuilt-in
Declarative Policy CardsNoNoNo22 Policies
SOP Decision GraphsNoPartialNo11 SOPs
MCDM Agent ScoringNoNoNoTOPSIS/AHP
Cryptographic Audit TrailNoNoNoEd25519
Anti-Loop Neural ProtectionNoNoNoActive
Cost Governance ($0 entry)NoNoNoCascading
Interchangeable Backends1118 Backends
Air-Gapped / Offline ModeNoNoNoOllama
Open Source

Deploy in 5 minutes.

NEXUS is released under Apache 2.0. Run locally with zero cost, or deploy to production with Docker Compose.

1

Clone

git clone https://github.com/blackstack-ai/nexus
2

Configure

cp .env.example .env
# Add API keys or use Ollama ($0)
3

Launch

docker compose up -d

Advancing responsible AI infrastructure.

An open-source contribution to AI governance — making multi-agent systems auditable, accountable, and safe for enterprise deployment.

Documenting
Updating
Managing
Auditing
Last audit: 2026-02-13v4.0