A self-governing, learning, multi-agent AI infrastructure platform. 7 repos, 339 K8s manifests, 30 microservices, 11 MCP servers. Now consolidated at github.com/cortex-io.
Modeled as a construction company. Learns from YouTube. Deploys via GitOps. Rolls back failures automatically. The brain, the muscle, and the deployment.
February 1, 2026 - The ecosystem now lives under one roof
7 repos transferred, coordinated versioning at v2026.02.0, one-command setup
One command to clone everything: curl -fsSL https://raw.githubusercontent.com/cortex-io/.github/main/setup.sh | bash
Cortex is a self-governing, learning, multi-agent AI infrastructure platform modeled as a construction company. Think Cortex Holdings Inc. - with an Executive level, Shared Services, and 6 specialized Divisions.
Unlike traditional automation, Cortex learns from YouTube videos, validates improvements via RAG, auto-deploys via GitOps, and rolls back failures automatically. The infrastructure that teaches itself.
Now consolidated at github.com/cortex-io with coordinated versioning. 7 repos, 339 K8s manifests, 30 microservices, 11 MCP servers, 113+ tools. All running on a 7-node K3s cluster.
Organized as Cortex Holdings Inc. with Executive level, Shared Services, and 6 specialized Divisions managing 20+ repositories.
Cortex School ingests YouTube videos, validates against existing infrastructure via RAG, auto-approves improvements, deploys via GitOps, and rolls back failures automatically.
Scale-to-zero AI stacks with intelligent cascading: Cache → Keyword → Qdrant → Local LLM → Claude. 80% of queries handled locally.
Mixture of Experts routing with keyword and semantic matching. Routes tasks to 5 Master Agents across specialized domains.
339 K8s manifests in cortex-gitops. ArgoCD auto-syncs every 3 minutes with self-healing and drift correction.
11 MCP servers providing 6-tier capabilities: from simple queries to spawning 10,000 workers to building entire projects.
You issue commands to Claude Code in cortex-platform (dev). Changes commit to cortex-gitops. ArgoCD watches every 3 minutes, syncs to the K3s cluster, and self-heals any drift. The control plane whispers; the cluster thunders.
Every task is analyzed by the MoE router (87-95% accuracy) and sent to the right Master Agent: Coordinator, Development, Security, Inventory, or CI/CD. Each master spawns specialized workers. 7 worker types handle implementation, fixes, tests, scans, security fixes, documentation, and analysis.
Queries flow through 5 tiers: Cache (<1ms) → Keyword (<10ms) → Qdrant vector search (<50ms) → Local LLM (~500ms) → Claude (~2-5s). 80% of queries never leave Tier 3. Scale-to-zero via KEDA means you only pay for what you use.
Cortex learns from YouTube videos. Improvements flow through a 7-stage Redis pipeline: Raw → MoE Router → Expert Analysis → RAG Validation → Auto-Approval (≥90% relevance) → Implementation Workers → GitOps Deploy → Health Monitor. Failures auto-rollback via git revert.
A multi-agent AI system organized as a construction company.
Executive Level
├── Cortex Prime (General Contractor) - Strategic decisions
└── COO (Site Supervisor) - Daily operations
Shared Services (Equipment Yard)
├── Coordinator Master - MoE routing hub (50k tokens)
├── Development Master - Features (30k tokens)
├── Security Master - CVEs (30k tokens)
├── Inventory Master - Cataloging (20k tokens)
└── CI/CD Master - Builds (25k tokens)
6 Divisions (Sub-contractors)
├── Infrastructure - Proxmox, UniFi, Cloudflare, Starlink
├── Containers - Talos (K8s on bare metal)
├── Workflows - n8n automation
├── Configuration - Microsoft Graph
├── Monitoring - Netdata, Grafana, CheckMK, Pulseway
└── Intelligence - AIANA
7 Worker Types (Day Laborers)
├── Implementation, Fix, Test, Scan
└── Security-Fix, Documentation, Analysis
"The infrastructure that teaches itself."
Cortex School learns from YouTube → validates via RAG → auto-deploys via GitOps → monitors health → rolls back failures. No human required.
Vulnerability scanning, CVE remediation, dependency audits, compliance monitoring, and automated security fixes.
Feature implementation, bug fixes, code refactoring, optimization, and technical debt reduction.
Build orchestration, test automation, deployment strategies, release workflows, and pipeline optimization.
Discovery, cataloging, documentation generation, dependency tracking, and health monitoring.
Dive deeper into Cortex's development, architecture, and capabilities through our detailed blog posts.
How Cortex evolved from static routing to a self-learning system that reduces API calls by 65%, improves latency by 67%, and automatically optimizes model selection through vector-based similarity routing
A deep dive into the evolution from simple LLM interactions to autonomous, self-optimizing AI fabric with intelligent routing, scale-to-zero layers, and continuous learning
Transforming Cortex from monolithic to distributed fabric network with six domain-specific AI activators, Redis Streams orchestration, and MCP protocol integration - solving protocol mismatches, cluster capacity, and cross-namespace secrets
How Cortex continuously learns from YouTube channels, validates improvements with RAG, and autonomously implements safe infrastructure changes with automatic rollback.
How we unified three disconnected clients into a resilient event-driven fabric connecting 14 MCP servers across k3s, enabling true session continuity.
From 200+ always-on pods consuming 96% cluster memory to a serverless architecture that scales on-demand. The journey to building the Layer Activator—Cortex's traffic cop for intelligent MCP stack orchestration.
Cortex is now consolidated at github.com/cortex-io. 7 repos, coordinated versioning, one-command setup.
Explore the architecture, fabric layers, autonomous learning pipeline, MCP servers, and the construction company model that makes a complex system navigable.
Start typing to search...