Context that lasts.
Work you can trace.
Rootweaver connects my coding tools to persistent memory, searchable knowledge and code relationships. I design and operate it, with AI-assisted development.
One platform. Connected layers.
Follow the system from the places I work to the services that remember, retrieve, deliver and observe it.
Architecture / 14 October 2026
Arrows show direction. Select a component to read the relationship. Pan or zoom to explore; use Fit View to return to the whole map.
This map groups the major platform systems. PostgreSQL icons represent logical data roles; model routes have different consumers; hosting lines are representative. It is a dated architecture view, not live traffic or a complete service inventory.
The platform around the tools.
Knowledge, model access and research tools sit on a shared execution and observation layer.
A small, distributed cluster.
K3s runs the service workloads across four machines. The Mac is a native working surface and home to the canonical vault. GitOps declares the cluster state.
Explicit model dependencies.
The current retrieval setup uses external embeddings and selected model routes through LiteLLM. Local GPU services are capacity-dependent; the diagrams do not assume they are always running.
Space to run and examine work.
A JupyterLab workbench, MLflow and Langfuse support research, experiment tracking and model-workflow inspection. These sit alongside the coding and knowledge tools.
See what the system is doing.
Metrics, logs and traces connect runtime behaviour to the work that produced it. Grafana and the monitoring services support investigation; a deployed component is only one part of that evidence.
See how it got here.
The platform page explains how the pieces work together. The journey and writing show what changed and what I learned.
Architecture checked against source, GitOps and a read-only cluster snapshot on 14 October 2026. Flow animation is illustrative. This page does not report live service health.