This directory contains the long-form documentation for llm-observability-stack. The top-level README.md is the fast entry point; these documents cover architecture, deployment profiles, local operations, troubleshooting, validation, and publishing.
The documentation is organized around private LLM deployment and observability on k3s/Kubernetes with CPU-only and NVIDIA GPU paths. The verified local reference environment is Xubuntu 24 with k3s and an NVIDIA GPU.
- DEPENDENCY-AUDIT-2026-07-19.md
- QWEN-1.8B-LIVE-VALIDATION-2026-07-18.md
- LIVE-VALIDATION-GO-NATIVE-2026-07-18.md
- LIVE-VALIDATION-2026-07-17.md
- QUICKSTART.md
- cli.md
- K3S-NVIDIA-EDGE-DEPENDENCY.md
- CONFIG-PROFILES.md
- XUBUNTU-K3S-NVIDIA-RUNBOOK.md
- ARCHITECTURE.md
- OPERATIONS-RUNBOOK.md
- PROJECT-DOCUMENTATION.md
- LANGUAGE-BOUNDARIES.md
- GO-NATIVE-MIGRATION-2026-07-18.md
External companion:
gguf-observabilityprovides read-only, model-selectable runtime contract checks and sanitized evidence. It does not replace this repository's Helm, Modelfile, or benchmark assets.
Dashboard presentation is now owned by the Helm-provisioned Grafana JSON in
../dashboards/; there is no separate browser application or
browser-side Kubernetes access path.
-
- Sequential Qwen, Gemma, and Llama deployment with a 900 MiB observed VRAM ceiling.
-
- Fast local setup for k3s, values files, image build/import, install, and first validation.
-
- Go CLI build instructions, commands, profile mapping, install/validate/benchmark/uninstall examples, and two-layer architecture notes.
-
- Required install order and ownership boundary between
k3s-nvidia-edgeand this LLM stack.
- Required install order and ownership boundary between
-
- Canonical comparison of git-tracked defaults, local example values, GPU profiles, CPU profiles, and private overrides.
-
- Current local runbook for Xubuntu 24 + k3s + NVIDIA GPU, based on the active local deployment.
-
- Existing local k3s/NVIDIA command reference.
-
- Component ownership, request paths, service exposure, and configuration boundaries.
-
- Go-first runtime, optional notebook Python, Bash, and Helm/YAML ownership rules.
-
- Day-0 and day-1 tasks: deploy, verify, port-forward, rebuild images, debug, and clean up.
-
- Walkthrough of notebooks
01through10, their prerequisites, and common execution pitfalls.
- Walkthrough of notebooks
-
- Full repository documentation, component walkthroughs, and deployment model.
-
- Current-state summary and hardening priorities.
- KUBERNETES-NETWORKING.md
- Service, EndpointSlice, DNS, ServiceLB, and traffic-path documentation for this stack.
- KUBECTL-COMMAND-REFERENCE.md
- High-signal
kubectlcommand catalog for local operations.
- High-signal
- GO-KUBERNETES-AUTOMATION.md
- Native Go CLI networking inventory, service-path, and endpoint-watch commands.
- VERIFIED-LOCAL-GPU-RESULTS.md
- Historical Gemma-era sanitized GPU benchmark and deployment results.
- LOCAL-K3S-NVIDIA-REPORT-2026-07-02.md
- Historical Gemma-era k3s/NVIDIA status report for the Xubuntu 24 system.
- SINGLE-NODE-K3S-GEFORCE-940M.md
- Low-memory GeForce 940M profile notes and constraints.
- ../validation/README.md
- Validation artifact index.
- GITHUB-PUBLISHING.md
- Remote setup, safe publishing workflow, and repository hygiene guidance.
- First deploy:
- CPU-only or MacOS/minikube:
- Contributor/operator:
- scripts/README.md
- Inventory of standalone helper scripts in
docs/scripts/.
- Inventory of standalone helper scripts in