Platform
The NEXUS OS Product Suite
Six core modules, one operating system. Each module is independently useful — together, they form a closed loop from plain-language spec to production-merged, security-validated code.
NEXUS Forge
Multi-agent project builder
The orchestration engine at the heart of NEXUS OS. Forge takes a plain-language feature description, decomposes it into a dependency-ordered task graph, and dispatches each task to the right agent — from Orchestrator down to MicroAgent — until the whole delivery merges.
- Codebase-aware task decomposition
- DAG-based dependency scheduling
- Dispatch → execute → merge pipeline
- Full Orchestrator → MicroAgent hierarchy
NEXUS CORTEX
Adaptive routing, reputation & skills
CORTEX is the intelligence layer that decides who does the work and how. It combines UCB1 multi-armed-bandit routing, EWMA reputation tracking per agent role, and a composable skill library — so every task lands on the agent and model tier most likely to succeed.
- UCB1 adaptive tier routing (haiku/sonnet/opus)
- EWMA quality reputation per agent role
- 42-skill library across 6 domains
- Cost router KPI: 74% saved vs. baseline
Guardian
Auto-patching for production
Guardian watches deployed code for drift, regressions, and newly disclosed vulnerabilities, then opens an automated, MergeGate-validated patch. It closes the loop between 'we found a problem' and 'a verified fix is ready to merge' — no idle backlog tickets.
- Continuous drift & regression detection
- Automated patch generation
- MergeGate-validated before merge
- Works alongside AutoFix for security CVEs
AutoFix
Security remediation loop
When the SecurityScanner gate flags a finding — secrets, injection, XSS, vulnerable dependencies — AutoFix generates a targeted remediation, re-runs the scan, and repeats up to a bounded number of retries. In the validation campaign, this took a cmd-injection finding from 1 to 0 with a clean MergeGate PASS.
- Triggered automatically on Q < 0.75
- Bounded retry loop (max attempts)
- Re-runs SecurityScanner after each fix
- Proven: 1 → 0 cmd-injection findings
Governance / MergeGate
Quality gates before merge
Every delivery is scored on a 5-axis composite Quality Score Q before it can merge: tests, security, token efficiency, self-correction, and constitution adherence. Q ≥ 75 passes, Q < 75 triggers AutoFix, and Q < 60 blocks with human escalation.
- 5-axis weighted Quality Score (Q)
- Automatic AutoFix trigger below threshold
- Hard BLOCK + escalation below safety floor
- 100/100 MergeGate pass on live pilot
Agent Marketplace
Discover & deploy specialized agents
Browse and deploy SpecialistAgents and pre-built skill packs contributed by the NEXUS community and core team. Every listing carries its EWMA reputation history and Quality Score track record, so you know exactly what you're adding to your team.
- Browse SpecialistAgents & skill packs
- Reputation & Q-score history per listing
- One-click deploy into your agent registry
- Community + core-team contributions
Every module works standalone — and better together
Start with Forge for your next feature, or drop Guardian into an existing production service. The full loop — plan, build, validate, secure, deploy — is one command away.