Historical planning document. The architecture described below is now operational and is documented in the Container Framework README and the Deployment Walkthrough. This page is preserved as the original strategy record.
Restructure Cytognosis infrastructure into two tiers to optimize compute costs, increase deployment flexibility, and support scaling Data Science / MLOps efforts:
1. Core Services Bundle (always-on) — a cost-optimized instance hosting lightweight, persistent services.
2. Heavy Compute Nodes (on-demand) — high-resource instances for heavy workloads, leveraging custom pre-packaged Docker images.
Custom Docker images are pushed to the GCP Artifact Registry (private, IAM-protected) rather than to Docker Hub. The marginal storage / bandwidth cost is acceptable given the integrated IAM security and the fact that we only publish a small number of images.
The original cytohost instance was re-architected as the Core Node running the core stack:
A separate Zotero metadata integration was originally planned in this bundle, but the sovereign paper library architecture replaced it: Zotero metadata syncs against Zotero's free cloud, and PDFs live on Google Drive. The container_framework/configs/services/zotero.yaml file is retained for the optional self-hosted-data-server path, but it is not part of the deployed core stack.
Built as focused images that can be deployed on-demand to heavy GCP instances or local developer machines via the research stack:
container_framework/configs/services/neo4j.yaml. Community Edition + APOC plugin. Sized for OLS4 ontology browsing and KG exploration.container_framework/configs/services/jupyter.yaml. Custom image (cytognosis-compute/datascience:latest) with Python, R, PyTorch, JAX, HuggingFace, and the MLflow client. Boots JupyterLab on port 8888.cytognosis-infrastructurecytognosis-computeus-central1us-central1-docker.pkg.dev/cytognosis-infrastructure/cytognosis-compute/: CI builds and pushes are wired through GitHub Actions Workload Identity Federation (no long-lived JSON keys). See .github/workflows/build-containers.yml.
neo4j and datascience images on every change to compute_images/* and verifies dependencies install cleanly.stack_manager.py CLI.