From 6ca3c6770577a1759df32da93647bf188d054efb Mon Sep 17 00:00:00 2001
From: Erich Blume <blume.erich@gmail.com>
Date: Wed, 4 Mar 2026 19:43:14 -0800
Subject: [PATCH] Add Ollama reference card and update indexes

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 .../changelog.d/+ollama-reference-card.doc.md |  1 +
 docs/reference/kubernetes/apps.md             |  3 +-
 docs/reference/reference.md                   |  3 +-
 docs/reference/services/ollama.md             | 89 +++++++++++++++++++
 4 files changed, 94 insertions(+), 2 deletions(-)
 create mode 100644 docs/changelog.d/+ollama-reference-card.doc.md
 create mode 100644 docs/reference/services/ollama.md

diff --git a/docs/changelog.d/+ollama-reference-card.doc.md b/docs/changelog.d/+ollama-reference-card.doc.md
new file mode 100644
index 0000000..2d88484
--- /dev/null
+++ b/docs/changelog.d/+ollama-reference-card.doc.md
@@ -0,0 +1 @@
+Add reference card for the Ollama LLM inference service.
diff --git a/docs/reference/kubernetes/apps.md b/docs/reference/kubernetes/apps.md
index dfe8084..919907e 100644
--- a/docs/reference/kubernetes/apps.md
+++ b/docs/reference/kubernetes/apps.md
@@ -1,6 +1,6 @@
 ---
 title: Apps
-modified: 2026-02-25
+modified: 2026-03-04
 tags:
   - kubernetes
   - argocd
@@ -36,6 +36,7 @@ Registry of all applications deployed via [[argocd]].
 | `teslamate` | teslamate | `argocd/manifests/teslamate/` | [[teslamate]] |
 | `cv` | cv | `argocd/manifests/cv/` | [[cv]] |
 | `forgejo-runner` | forgejo-runner | `argocd/manifests/forgejo-runner/` | [[forgejo]] CI |
+| `ollama` | ollama | `argocd/manifests/ollama/` | [[ollama]] |
 
 ## Sync Policies
 
diff --git a/docs/reference/reference.md b/docs/reference/reference.md
index b4d3ee3..9faa8e2 100644
--- a/docs/reference/reference.md
+++ b/docs/reference/reference.md
@@ -1,6 +1,6 @@
 ---
 title: Reference
-modified: 2026-02-24
+modified: 2026-03-04
 tags:
   - reference
 ---
@@ -40,6 +40,7 @@ Individual service reference cards with URLs and configuration details.
 | [[authentik]] | OIDC identity provider | k8s (ringtail) |
 | [[docs]] | Documentation site (Quartz) | k8s |
 | [[flyio-proxy]] | Public reverse proxy (Fly.io + Tailscale) | Fly.io |
+| [[ollama]] | LLM inference server | k8s (ringtail) |
 | [[automounter]] | SMB share automounter | indri |
 
 ## Infrastructure
diff --git a/docs/reference/services/ollama.md b/docs/reference/services/ollama.md
new file mode 100644
index 0000000..75480cb
--- /dev/null
+++ b/docs/reference/services/ollama.md
@@ -0,0 +1,89 @@
+---
+title: Ollama
+modified: 2026-03-04
+tags:
+  - service
+  - ai
+---
+
+# Ollama
+
+LLM inference server with GPU acceleration. Runs on [[ringtail]] with declarative model management via a sidecar.
+
+## Quick Reference
+
+| Property | Value |
+|----------|-------|
+| **URL** | https://ollama.ops.eblu.me |
+| **Tailscale URL** | https://ollama.tail8d86e.ts.net |
+| **Namespace** | `ollama` |
+| **Cluster** | ringtail k3s |
+| **Image** | `ollama/ollama:0.17.5` |
+| **Upstream** | https://github.com/ollama/ollama |
+| **Manifests** | `argocd/manifests/ollama/` |
+| **API Port** | 11434 |
+
+## Architecture
+
+```
+models.txt (ConfigMap, declarative)
+    │
+    ▼
+model-sync sidecar ──ollama pull──► Ollama server (GPU)
+    │                                    │
+    │ reads /config/models.txt           │ serves /api/*
+    │ polls every 30 min                 │ NVIDIA runtime (RTX 4080, time-sliced)
+    │                                    │
+    └────────────────────────────────────┘
+                     │
+                /models (200 Gi hostPath PV)
+                /mnt/storage1/ollama on ringtail
+```
+
+## Models
+
+Declared in `argocd/manifests/ollama/models.txt`. The model-sync sidecar pulls missing models on startup and every 30 minutes.
+
+| Model | Parameters |
+|-------|------------|
+| `qwen2.5:14b` | 14B |
+| `deepseek-r1:14b` | 14B |
+| `phi4:14b` | 14B |
+| `gemma3:12b` | 12B |
+
+To add or remove models, edit `models.txt` and sync via ArgoCD.
+
+## GPU
+
+Shares [[ringtail]]'s RTX 4080 with [[frigate]] via NVIDIA device plugin time-slicing (2 virtual slots). Constrained to one loaded model and one parallel request to avoid VRAM contention.
+
+| Setting | Value |
+|---------|-------|
+| `OLLAMA_MAX_LOADED_MODELS` | 1 |
+| `OLLAMA_NUM_PARALLEL` | 1 |
+| GPU limit | `nvidia.com/gpu: "1"` (time-sliced) |
+
+## Storage
+
+| Mount | Backend | Size |
+|-------|---------|------|
+| `/models` | hostPath PV (`/mnt/storage1/ollama`) | 200 Gi |
+
+PV reclaim policy is `Retain` — models survive PV deletion.
+
+## Networking
+
+| Endpoint | Reachable from |
+|----------|----------------|
+| `https://ollama.ops.eblu.me` | Public internet (Fly.io → Caddy) |
+| `https://ollama.tail8d86e.ts.net` | Tailnet clients |
+| `http://ollama.ollama.svc.cluster.local:11434` | In-cluster (ringtail) |
+
+Tailscale ingress uses ProxyGroup `ingress` — no explicit `host:` field (see [[tailscale-operator]]).
+
+## Related
+
+- [[frigate]] — Shares GPU via time-slicing
+- [[ringtail]] — Host node
+- [[apps]] — ArgoCD application registry
+- [[tailscale-operator]] — Tailscale ingress