Note
Archived. This repo is part of Cortex, which is no longer under active development. It is kept as a working record: explore, fork and borrow freely, but no fixes or features are planned.
Cortex · cortex · cortex-platform · cortex-gitops · cortex-k3s · cortex-docs · cortex-construction-hq · infrastructure-docs
Deploy a generalist AI stack. Operate it on real tasks. Collect telemetry. Distill the operational knowledge into a specialized model. Graduate from orchestration to a single expert.
Infrastructure as training data.
| Component | Role |
|---|---|
| MoE Router | Routes queries to the right model/tool |
| Qdrant | Vector storage for domain knowledge |
| MCP Server | Tool execution |
| Telemetry Collector | Captures routing decisions, queries, tool calls |
| Distillation Pipeline | Trains LoRA adapters from collected data |
- Scale-to-zero — all components idle to zero via KEDA
- Burst-capable — scales up on demand
- Telemetry-driven distillation — CronJob pipeline: collect → filter → fine-tune → publish
- Output formats — LoRA, full fine-tune, or GGUF
infrastructure-layer · security-layer · networking-layer
Part of the Cortex archive · built with Claude