2026-06-06 15:19:06 +02:00
7 changed files with 599 additions and 8 deletions
--- a/.env.example
+++ b/.env.example
@@ -72,6 +72,25 @@ VITE_SENTRY_DSN=
 # Sentry/GlitchTip auth token for source map upload at build time (optional)
 SENTRY_AUTH_TOKEN=

+# NL search — Ollama LLM inference
+# Leave APP_OLLAMA_BASE_URL empty to disable NL search (safe default for CX32 / CI).
+# Set to http://ollama:11434 to enable. Requires CX42 (16 GB RAM) to run alongside OCR.
+APP_OLLAMA_BASE_URL=http://ollama:11434
+
+# CPU limit: 4.0 is safe on both CX32 (4 vCPUs) and CX42 (8 vCPUs).
+# Raise to 7.5 on CX42 for full throughput.
+OLLAMA_CPU_LIMIT=4.0
+
+# Memory limit: requires CX42 (16 GB) to run alongside OCR.
+# Reduce or set APP_OLLAMA_BASE_URL= on smaller hosts.
+OLLAMA_MEM_LIMIT=8g
+
+# Ollama API key — set on the Ollama service to restrict inference API access on archiv-net.
+# Generate with: openssl rand -hex 32
+# NOTE: Empirically verified that OLLAMA_API_KEY is NOT enforced in Ollama 0.6.5 or 0.30.6 (ADR-028 §7).
+# archiv-net network isolation is the only effective access control. Retained for forward compatibility.
+OLLAMA_API_KEY=
+
 # Production SMTP — uncomment and fill in to send real emails instead of catching them
 # APP_BASE_URL=https://your-domain.example.com
 # MAIL_HOST=smtp.example.com
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -141,6 +141,65 @@ services:
    security_opt:
      - no-new-privileges:true

+  # --- Ollama: Model init (one-shot pull) ---
+  # Pulls qwen2.5:7b-instruct-q4_K_M (~4.7 GB) into the ollama_models volume on first start.
+  # On subsequent starts (model already in volume), exits quickly without re-downloading.
+  # Not started in CI — CI uses explicit service selection
+  # (docker-compose.ci.yml: db minio create-buckets)
+  ollama-model-init:
+    image: ollama/ollama:0.30.6
+    restart: "no"
+    networks:
+      - archiv-net
+    volumes:
+      - ollama_models:/root/.ollama
+    mem_limit: 2g
+    read_only: true
+    tmpfs:
+      - /tmp:size=512m
+    cap_drop:
+      - ALL
+    security_opt:
+      - no-new-privileges:true
+    command: >
+      sh -c "ollama serve & SERVE_PID=$$! && until curl -sf http://localhost:11434/api/tags; do sleep 1; done && ollama pull qwen2.5:7b-instruct-q4_K_M && kill $$SERVE_PID"
+
+  # --- Ollama: LLM inference server ---
+  # Serves the pre-pulled model for NL search inference.
+  # Not started in CI — CI uses explicit service selection
+  # (docker-compose.ci.yml: db minio create-buckets)
+  ollama:
+    image: ollama/ollama:0.30.6
+    container_name: archive-ollama
+    restart: unless-stopped
+    expose:
+      - "11434"
+    networks:
+      - archiv-net
+    volumes:
+      - ollama_models:/root/.ollama
+    environment:
+      OLLAMA_API_KEY: "${OLLAMA_API_KEY}"
+    cpus: "${OLLAMA_CPU_LIMIT:-4.0}"
+    mem_limit: "${OLLAMA_MEM_LIMIT:-8g}"
+    memswap_limit: "${OLLAMA_MEM_LIMIT:-8g}"
+    read_only: true
+    tmpfs:
+      - /tmp:size=512m
+    cap_drop:
+      - ALL
+    security_opt:
+      - no-new-privileges:true
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:11434/api/tags"]
+      interval: 30s
+      timeout: 10s
+      retries: 5
+      start_period: 60s  # model weights are pre-loaded by ollama-model-init; service only needs to bind port
+    depends_on:
+      ollama-model-init:
+        condition: service_completed_successfully
+
  # --- Backend: Spring Boot ---
  backend:
    build:
@@ -184,6 +243,8 @@ services:
      SPRING_MAIL_PROPERTIES_MAIL_SMTP_STARTTLS_ENABLE: ${MAIL_STARTTLS_ENABLE:-false}
      APP_OCR_BASE_URL: http://ocr-service:8000
      APP_OCR_TRAINING_TOKEN: "${OCR_TRAINING_TOKEN:-}"
+      APP_OLLAMA_BASE_URL: "${APP_OLLAMA_BASE_URL:-http://ollama:11434}"
+      APP_OLLAMA_API_KEY: "${OLLAMA_API_KEY}"
      SENTRY_DSN: ${SENTRY_DSN:-}
      SENTRY_TRACES_SAMPLE_RATE: ${SENTRY_TRACES_SAMPLE_RATE:-1.0}
      # Observability: send traces to Tempo inside archiv-net (OTLP gRPC port 4317)
@@ -247,3 +308,4 @@ volumes:
  frontend_node_modules:
  ocr_models:
  ocr_cache:
+  ollama_models:
--- a/docs/DEPLOYMENT.md
+++ b/docs/DEPLOYMENT.md
@@ -50,15 +50,17 @@ graph TD

 The OCR service requires significant RAM for model loading. The dev compose sets `mem_limit: 12g`.

-| Production target | RAM | Recommended OCR limit | Notes |
-|---|---|---|---|
-| Current server (Hetzner Serverbörse, i7-6700) | 64 GB | 12 GB | Default `mem_limit: 12g` works comfortably |
-| ≥ 16 GB RAM | 16+ GB | 12 GB | Default works |
-| 8 GB RAM | 8 GB | 6 GB | Set `OCR_MEM_LIMIT=6g`; accept reduced batch sizes |
-| 4 GB RAM | 4 GB | — | Disable OCR service (`profiles: [ocr]`); run OCR on demand only |
+| Production target | RAM | Recommended OCR limit | NL Search | Notes |
+|---|---|---|---|---|
+| Current server (Hetzner Serverbörse, i7-6700) | 64 GB | 12 GB | Supported | Default `mem_limit: 12g` works comfortably; plenty of headroom for Ollama |
+| ≥ 16 GB RAM | 16+ GB | 12 GB | Supported | Default works |
+| 8 GB RAM | 8 GB | 6 GB | Disabled — set `APP_OLLAMA_BASE_URL=` (empty) | Set `OCR_MEM_LIMIT=6g`; accept reduced batch sizes |
+| 4 GB RAM | 4 GB | — | Unsupported | Disable OCR service (`profiles: [ocr]`); run OCR on demand only |

 On servers with less than 16 GB RAM the default `mem_limit: 12g` cannot be honoured — set the `OCR_MEM_LIMIT` env var (in `.env.production` / `.env.staging`, or as a Gitea secret consumed by the workflow). The prod compose interpolates this var with a 12g default.

+> **Memory budget:** OCR (~6 GB active) + Ollama (~8 GB) = ~14 GB. On servers with less than 16 GB RAM, do not run `docker-compose.observability.yml` continuously alongside both OCR and Ollama.
+
 ### Dev vs production differences

 | Concern | Dev (`docker-compose.yml`) | Prod (`docker-compose.prod.yml`) |
@@ -145,6 +147,16 @@ All vars are set in `.env` at the repo root (copy from `.env.example`). The back
 | `XDG_CACHE_HOME` | XDG cache base dir — redirects Matplotlib and other XDG-aware libraries away from the read-only `HOME` (`/home/ocr`) to the writable cache volume | `/app/cache` | — | — |
 | `TORCH_HOME` | PyTorch model cache — redirects `~/.cache/torch` to the writable models volume | `/app/models/torch` | — | — |

+### Ollama (NL search) service
+
+| Variable | Purpose | Default | Required? | Sensitive? |
+|---|---|---|---|---|
+| `APP_OLLAMA_BASE_URL` | Base URL for the Ollama service. Leave empty to disable NL search. | `http://ollama:11434` | — | — |
+| `APP_OLLAMA_API_KEY` | API key passed as `Authorization: Bearer` to Ollama. Leave empty for unauthenticated access. Note: `OLLAMA_API_KEY` is not enforced in Ollama 0.6.5 or 0.30.6 (see ADR-028). | — | — | YES |
+| `OLLAMA_CPU_LIMIT` | Docker CPU quota for the Ollama container. On CX42 (8 vCPUs) can be raised to `7.5`. | `4.0` | — | — |
+| `OLLAMA_MEM_LIMIT` | Memory limit for the Ollama container. Requires CX42 (16 GB RAM). | `8g` | — | — |
+| `OLLAMA_API_KEY` | API key set on the Ollama service itself. Same value as `APP_OLLAMA_API_KEY`. Leave empty for unauthenticated. | — | — | YES |
+
 ### Observability stack (`docker-compose.observability.yml`)

 | Variable | Purpose | Default | Required? | Sensitive? |
@@ -265,6 +277,19 @@ git.raddatz.cloud      A   <server IP>

 ### 3.4 First deploy

+> **First start — Ollama model pull:** On first `docker compose up -d`, the `ollama-model-init` container pulls `qwen2.5:7b-instruct-q4_K_M` (~4.7 GB). At 10 Mbps this takes approximately 60–90 minutes; at 100 Mbps approximately 6–10 minutes. The pull is a one-time operation — subsequent restarts skip it (model already on the `ollama_models` volume). Monitor progress with `docker logs -f $(docker ps -q --filter name=ollama-model-init)`.
+>
+> **Do not use `--wait` on first deploy** — `docker compose up -d --wait` waits for all services to reach their health/completion target, including `ollama-model-init`. On first pull this blocks for 60–90 minutes and will time out any CI/deploy script that uses `--wait`.
+>
+> **Re-deploy idempotency:** on subsequent `docker compose up -d` runs (including `--force-recreate`), `ollama-model-init` re-executes but exits in seconds — Ollama's CLI skips the download when the model digest already matches what is on the volume.
+>
+> **Verify NL search is active** after enabling Ollama (`APP_OLLAMA_BASE_URL=http://ollama:11434`):
+> ```bash
+> curl -s http://localhost:8080/api/nl-search?q=brief+von+grossmutter
+> # Returns 200 with results → NL search is active
+> # Returns 503 NL_SEARCH_UNAVAILABLE → Ollama is not reachable or APP_OLLAMA_BASE_URL is unset
+> ```
+
 ```bash
 # 1. Trigger nightly.yml manually (Repo → Actions → nightly → "Run workflow")
 #    Expected: docker compose up -d --wait succeeds for archiv-staging, then
@@ -560,6 +585,24 @@ bash scripts/download-kraken-models.sh

 > Downloads the Kurrent/Sütterlin HTR models. Run once after a fresh clone or when models are updated.

+### Upgrade the Ollama model
+
+To switch to a newer model version (e.g. a future release of `qwen2.5`):
+
+1. Update the model name in the `ollama-model-init` `command:` in `docker-compose.yml`.
+2. Remove the existing model volume to free the old weights:
+   ```bash
+   docker volume rm familienarchiv_ollama_models
+   ```
+   (In production the volume name is prefixed with the compose project: `archiv-production_ollama_models`.)
+3. Restart the stack:
+   ```bash
+   docker compose up -d
+   ```
+   The `ollama-model-init` container pulls the new model weights on first start (~4–8 GB download depending on the model). The `ollama` inference server will not start until the pull completes (`condition: service_completed_successfully`).
+
+> **`ollama_models` volume:** holds model weights only — fully reproducible by re-pull, no backup needed.
+
 ### Trigger a canonical import

 The importer no longer parses the raw spreadsheet. It consumes the **canonical artifacts**
--- a/docs/adr/028-ollama-docker-compose-service.md
+++ b/docs/adr/028-ollama-docker-compose-service.md
@@ -0,0 +1,239 @@
+# ADR-028: Ollama Docker Compose service for NL search
+
+**Date:** 2026-06-06
+**Status:** Accepted
+**Deciders:** Marcel Raddatz
+**Relates to:** #737 (infrastructure), #735 (NL search epic)
+
+---
+
+## Context
+
+Issue #735 introduces natural-language document search, requiring a local LLM to generate embeddings and/or run inference at query time. The family archive stores personal family history — data privacy is non-negotiable, so cloud-based inference APIs are excluded. The production target is a Hetzner CX42 (16 GB RAM, 8 vCPUs, CPU-only, ~32 EUR/month).
+
+Alternatives considered:
+
+| Option | Reason rejected |
+|---|---|
+| **llama.cpp** | No HTTP API out of the box; requires custom wrapper; higher ops burden |
+| **vLLM** | GPU-first; significant overhead on CPU-only hardware; overkill for this scale |
+| **Cloud APIs** (OpenAI, Gemini, etc.) | Vendor lock-in; per-token cost at scale; data leaves the server — unacceptable for a private family archive |
+| **Ollama** | Self-contained Docker image; built-in HTTP REST API; actively maintained; CPU-compatible; zero egress |
+
+**Decision:** run Ollama as a Docker Compose service alongside the existing stack.
+
+---
+
+## Decisions
+
+### 1. Hardware minimums and CPU-only constraint
+
+All inference runs on CPU. The target is the Hetzner CX42 (16 GB RAM, 8 vCPUs).
+
+| Tier | RAM | NL search |
+|---|---|---|
+| CX42 | 16 GB | Supported — full stack including Ollama |
+| CX32 | 8 GB | Disabled — set `APP_OLLAMA_BASE_URL=` (empty) to skip Ollama entirely |
+| CX22 | 4 GB | Unsupported for NL search |
+
+### 2. Memory budget on CX42
+
+| Component | `mem_limit` | Typical active RSS |
+|---|---|---|
+| OCR service | 12g (hard ceiling) | ~6 GB |
+| Ollama | 8g | ~8 GB |
+| **Total** | | **~14 GB active** |
+
+`memswap_limit` on the Ollama service is set to `8g` (matching `mem_limit`) to prevent Linux from swapping model weights into swap under OCR memory pressure. Swapping model weights does not crash the container but silently degrades inference latency. This mirrors the pattern already applied to the OCR service.
+
+**Operational constraint:** do NOT run `docker-compose.observability.yml` continuously alongside both OCR and Ollama on a CX42. The observability stack adds ~2 GB, which leaves no headroom.
+
+### 3. Graceful-degradation contract
+
+`app.ollama.base-url` absent OR blank → Ollama bean NOT registered → NL search returns HTTP 503 with `ErrorCode: NL_SEARCH_UNAVAILABLE`.
+
+This single code path covers all unavailability scenarios: base-url unset, service unreachable, health check failed, and request timeout.
+
+#### Why not `@ConditionalOnProperty`
+
+`@ConditionalOnProperty` registers the bean when the property is present but blank (`APP_OLLAMA_BASE_URL=`). This produces a `RestClient` with an empty base URL that fails at runtime with an opaque error rather than a clean 503.
+
+#### Correct condition expression
+
+```java
+@ConditionalOnExpression("!'${app.ollama.base-url:}'.isBlank()")
+```
+
+When the property is absent, the placeholder resolves to `''`; `.isBlank()` returns `true`; negation makes the condition `false`; the bean is not registered. Same result for an explicit empty string (`APP_OLLAMA_BASE_URL=`).
+
+### 4. Backend configuration pattern
+
+Use a `@ConfigurationProperties` record, not separate `@Value` injections:
+
+```java
+@ConfigurationProperties("app.ollama")
+record OllamaProperties(String baseUrl, String apiKey) {}
+```
+
+`OllamaProperties` is registered unconditionally — it is a plain value holder with no side effects.
+
+`@ConditionalOnExpression` belongs **only** on `RestClientOllamaClient` (the bean that creates a live network client).
+
+**Deliberate divergence from the OCR pattern:** the OCR service uses `@Value`-with-default because OCR is always-on and `http://ocr-service:8000` is a safe default. Ollama is truly optional — a missing URL means "feature disabled", not "use this default server". There is no safe default Ollama URL.
+
+### 5. Optional<OllamaClient> injection
+
+The NL search service uses constructor injection with `Optional<OllamaClient>`:
+
+```java
+private final Optional<OllamaClient> ollamaClient;
+```
+
+When empty (bean not registered), the service method returns 503 immediately:
+
+```java
+var client = ollamaClient.orElseThrow(
+    () -> DomainException.internal(ErrorCode.NL_SEARCH_UNAVAILABLE, "Ollama not configured"));
+```
+
+Prefer this over `@Autowired(required = false)` with a null check — the null-check pattern is noisy when the service already uses `@RequiredArgsConstructor`.
+
+### 6. Empty API key guard
+
+`RestClientOllamaClient` omits the `Authorization` header entirely when `apiKey` is blank:
+
+```java
+if (!apiKey.isBlank()) {
+    request.header("Authorization", "Bearer " + apiKey);
+}
+```
+
+Sending `Authorization: Bearer ` (empty token) has undefined or potentially broken behavior depending on the Ollama version. This mirrors the `trainingToken` guard in `RestClientOcrClient.java:107`.
+
+### 7. OLLAMA_API_KEY behavior in Ollama 0.6.5 and 0.30.6
+
+**Empirically verified (2026-06-06) on both `0.6.5` and `0.30.6`:** `OLLAMA_API_KEY` does **not** enforce request authentication in either version.
+
+Test matrix run against `/api/tags`:
+
+| Configuration | No auth header | `Authorization: Bearer ` (empty) | `Authorization: Bearer wrongkey` | `Authorization: Bearer correctkey` |
+|---|---|---|---|---|
+| `OLLAMA_API_KEY=` (empty) | 200 | 200 | — | — |
+| `OLLAMA_API_KEY` unset | 200 | — | — | — |
+| `OLLAMA_API_KEY=testkey99` | 200 | 200 | 200 | 200 |
+
+**Finding:** The `OLLAMA_API_KEY` environment variable is not listed in Ollama's startup config dump and does not gate any HTTP request in either tested version. All configurations — empty string, fully unset, and a real key — accept all requests without authentication.
+
+**Practical implication:** `OLLAMA_API_KEY` provides no defense-in-depth in the tested versions. `archiv-net` network isolation is the only effective security control. The env var is retained in the Compose definition and `.env.example` for forward compatibility if Ollama enables enforcement in a future version, but operators must not rely on it for access control.
+
+**Backend guard still valid:** the `RestClientOllamaClient` code-level guard (omit `Authorization` header when `apiKey.isBlank()`) remains correct behavior regardless — it prevents a malformed `Authorization: Bearer ` header from being sent.
+
+### 8. read_only: true feasibility
+
+**Empirically verified (2026-06-06) on both `0.6.5` and `0.30.6`:** `read_only: true` works with Ollama. All three operations — `ollama serve`, `ollama pull qwen2.5:7b-instruct-q4_K_M`, and `ollama list` — succeeded with exit code 0 in both versions.
+
+Test run:
+```bash
+docker run --rm --read-only \
+  -v ollama_models:/root/.ollama \
+  --tmpfs /tmp \
+  --entrypoint sh ollama/ollama:0.30.6 \
+  -c "ollama serve & sleep 5 && ollama pull qwen2.5:7b-instruct-q4_K_M && ollama list"
+```
+
+**Note:** the entrypoint must be overridden to `sh` for the test command — the container's default entrypoint is `/bin/ollama` and does not accept `sh` as a subcommand. This is a Docker invocation detail; the Compose service definition uses the image's default entrypoint and `command:` override for the init container, which works correctly.
+
+**Result:** `read_only: true` and `tmpfs: - /tmp:size=512m` are applied to both `ollama` and `ollama-model-init`. The `ollama_models` volume handles all persistent writes; no other paths require write access during normal operation.
+
+### 9. Peak RSS of init container during pull
+
+**Empirically verified (2026-06-06):** Peak RSS during `qwen2.5:7b-instruct-q4_K_M` pull was **~108 MiB**.
+
+`docker stats` samples during the pull (15-second intervals):
+
+| Sample | MEM |
+|---|---|
+| 1 | 54.89 MiB |
+| 2 | 66.3 MiB |
+| 5 | 97.25 MiB |
+| 9 | **107.8 MiB** (peak) |
+
+`mem_limit: 2g` is adequate — the model weights stream directly to the named volume; RSS is dominated by the Ollama server process alone (~100 MB), not the model data. No bump to 4 GB needed.
+
+### 10. Init container pull mechanism
+
+The `ollama-model-init` container uses a curl-based readiness loop with captured PID:
+
+```sh
+ollama serve & SERVE_PID=$!
+until curl -sf http://localhost:11434/api/tags; do sleep 1; done
+ollama pull qwen2.5:7b-instruct-q4_K_M
+kill $SERVE_PID
+```
+
+`kill %1` (job-control syntax) is unreliable in non-interactive `sh -c` contexts. Capturing the PID via `SERVE_PID=$!` is reliable.
+
+The same endpoint (`/api/tags`) is used for both the init container readiness loop and the main service `healthcheck`.
+
+### 11. start_period: 60s rationale
+
+The model is pre-pulled by `ollama-model-init` before the main service starts (via `condition: service_completed_successfully`). At main service startup, Ollama only loads model weights from the named volume and binds port 11434.
+
+60 seconds is appropriate for this cold-start profile. 300 seconds was considered — that would be appropriate if the service pulled the model itself — but overstates actual startup time when the model is already present on the volume.
+
+### 12. Security threat model
+
+**Primary control:** `archiv-net` network isolation. Ollama has no externally exposed port (`expose:` only, not `ports:`). The Caddyfile must not route any path to the Ollama service.
+
+**Note on `OLLAMA_API_KEY`:** Per §7, `OLLAMA_API_KEY` is not enforced in Ollama 0.6.5 or 0.30.6 and provides no authentication barrier against a compromised backend container. `archiv-net` network isolation is the sole effective security control. The env var is retained for forward compatibility only — do not rely on it for access control.
+
+Both `ollama` and `ollama-model-init` receive the ADR-019 hardening baseline:
+
+```yaml
+cap_drop: [ALL]
+security_opt: [no-new-privileges:true]
+```
+
+### 13. CI exclusion strategy
+
+Docker Compose profiles are not used — they would add developer friction (requiring `--profile ...` for all local dev commands).
+
+CI uses explicit service selection in `docker-compose.ci.yml`:
+```bash
+docker compose -f docker-compose.ci.yml up -d db minio create-buckets
+```
+
+Ollama is simply not listed and is never started in CI. A YAML comment on the `ollama` service block documents this:
+
+```yaml
+# Not started in CI — CI uses explicit service selection
+# (docker-compose.ci.yml: db minio create-buckets)
+```
+
+### 14. ollama_models volume operational note
+
+The `ollama_models` named volume holds model weights only — fully reproducible by re-pull. No backup is needed.
+
+If the volume fills after a model upgrade:
+```bash
+docker volume rm ollama_models && docker compose up -d
+```
+The init container re-pulls the model on next startup.
+
+---
+
+## Consequences
+
+### Positive
+
+- NL search runs entirely on-premises; no data leaves the server and no per-token cloud cost.
+- Graceful degradation is a first-class concern: smaller or budget-constrained instances can run the app without Ollama with a single env var change.
+- The init container pattern keeps model pull out of the critical startup path for the main service, giving accurate healthcheck timings.
+- `@ConditionalOnExpression` with a blank-check is more correct than `@ConditionalOnProperty` for optional features with no safe default URL.
+
+### Risks and operational implications
+
+- **Memory pressure:** OCR + Ollama together consume ~14 GB on a 16 GB host. Running the observability stack simultaneously risks OOM kills. Monitor with `docker stats`.
+- **CPU inference latency:** `qwen2.5:7b-instruct-q4_K_M` is chosen for CPU viability, but inference on 8 vCPUs will be noticeably slower than GPU-accelerated alternatives. This is acceptable for the family archive use case (low concurrency, not real-time).
+- All three empirical TBD items from the original issue spec were resolved — see §7 (OLLAMA_API_KEY not enforced), §8 (`read_only: true` works), §9 (peak RSS ~108 MiB).
+- Model upgrades require a `docker volume rm` to free old weights before pulling the replacement. Document this in runbook/DEPLOYMENT.md.
--- a/docs/architecture/c4/l2-containers.puml
+++ b/docs/architecture/c4/l2-containers.puml
@@ -12,13 +12,15 @@ System_Boundary(archiv, "Familienarchiv (Docker Compose)") {
    Container(frontend, "Web Frontend", "SvelteKit / Node adapter / port 3000", "Server-side rendered UI. Handles auth session cookies, document search and viewer, transcription editor, annotation layer, family tree (Stammbaum), stories (Geschichten), activity feed (Chronik), enrichment workflow, and admin panel.")
    Container(backend, "API Backend", "Spring Boot 4 / Java 21 / Jetty / port 8080", "REST API. Implements document management, search, user auth, file upload/download, transcription, OCR orchestration, and SSE notifications. Trusts X-Forwarded-* headers from Caddy.")
    Container(ocr, "OCR Service", "Python FastAPI / port 8000", "Handwritten text recognition (HTR) and OCR microservice. Single-node by design — see ADR-001. Reachable only on the internal Docker network; no external port exposed.")
+    Container(ollama, "Ollama LLM Service", "ollama/ollama:0.30.6 / port 11434 (internal only)", "Local LLM inference server for NL search. Runs qwen2.5:7b-instruct-q4_K_M on CPU. Reachable only on the internal Docker network; no external port exposed. Disabled when APP_OLLAMA_BASE_URL is unset or blank.")
+    ' Named volume: ollama_models — model weights, fully reproducible, no backup needed
    ContainerDb(db, "Relational Database", "PostgreSQL 16", "Stores document metadata, persons, users, permission groups, tags, transcription blocks, audit log, and Spring Session data.")
    ContainerDb(storage, "Object Storage", "MinIO (S3-compatible)", "Stores the actual document files (PDFs, scans). Backend uses a bucket-scoped service account (archiv-app), not MinIO root.")
    Container(mc, "Bucket / Service-Account Init", "MinIO Client (mc)", "One-shot container on startup. Idempotent: creates the archive bucket, the archiv-app service account, and attaches the readwrite policy.")
 }

 System_Boundary(observability, "Observability Stack (/opt/familienarchiv/docker-compose.observability.yml)") {
-    Container(prometheus, "Prometheus", "prom/prometheus:v3.4.0", "Scrapes metrics from backend management port 8081 (/actuator/prometheus), node-exporter, and cAdvisor. Retention: 30 days.")
+    Container(prometheus, "Prometheus", "prom/prometheus:v3.4.0", "Scrapes metrics from backend (8081 /actuator/prometheus), OCR service (8000 /metrics), Ollama (11434 /metrics), node-exporter, and cAdvisor. Retention: 30 days.")
    Container(node_exporter, "Node Exporter", "prom/node-exporter:v1.9.0", "Host-level CPU, memory, disk, and network metrics.")
    Container(cadvisor, "cAdvisor", "gcr.io/cadvisor/cadvisor:v0.52.1", "Per-container resource metrics.")
    Container(loki, "Loki", "grafana/loki:3.4.2", "Stores log streams from all containers.")
@@ -45,6 +47,8 @@ Rel(promtail, loki, "Pushes log streams", "HTTP/Loki push API")
 Rel(backend, tempo, "Sends distributed traces via OTLP", "HTTP / OTLP / port 4318 (archiv-net)")
 Rel(prometheus, backend, "Scrapes JVM + HTTP metrics", "HTTP 8081 /actuator/prometheus")
 Rel(prometheus, ocr, "Scrapes OCR + http_* metrics", "HTTP 8000 /metrics")
+Rel(backend, ollama, "NL search inference requests", "HTTP / REST / JSON")
+Rel(prometheus, ollama, "Scrapes LLM request metrics", "HTTP 11434 /metrics")
 Rel(grafana, prometheus, "Queries metrics", "HTTP 9090")
 Rel(grafana, loki, "Queries logs", "HTTP 3100")
 Rel(grafana, tempo, "Queries traces", "HTTP 3200")
--- a/infra/observability/grafana/provisioning/dashboards/ollama.json
+++ b/infra/observability/grafana/provisioning/dashboards/ollama.json
@@ -0,0 +1,218 @@
+{
+  "id": null,
+  "uid": "ollama-dashboard",
+  "title": "Ollama",
+  "description": "Ollama inference latency and request rate",
+  "version": 1,
+  "schemaVersion": 39,
+  "tags": ["ollama", "inference"],
+  "timezone": "browser",
+  "editable": true,
+  "fiscalYearStartMonth": 0,
+  "graphTooltip": 1,
+  "links": [],
+  "liveNow": false,
+  "refresh": "30s",
+  "time": {
+    "from": "now-1h",
+    "to": "now"
+  },
+  "timepicker": {},
+  "weekStart": "",
+  "annotations": {
+    "list": [
+      {
+        "builtIn": 1,
+        "datasource": { "type": "datasource", "uid": "grafana" },
+        "enable": true,
+        "hide": true,
+        "iconColor": "rgba(0, 211, 255, 1)",
+        "name": "Annotations & Alerts",
+        "type": "dashboard"
+      }
+    ]
+  },
+  "panels": [
+    {
+      "id": 1,
+      "type": "timeseries",
+      "title": "Inference Latency p50",
+      "description": "50th percentile of Ollama request duration over a 5-minute window",
+      "gridPos": { "h": 8, "w": 8, "x": 0, "y": 0 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "fieldConfig": {
+        "defaults": {
+          "color": { "mode": "palette-classic" },
+          "custom": {
+            "axisBorderShow": false,
+            "axisCenteredZero": false,
+            "axisColorMode": "text",
+            "axisLabel": "",
+            "axisPlacement": "auto",
+            "barAlignment": 0,
+            "drawStyle": "line",
+            "fillOpacity": 10,
+            "gradientMode": "none",
+            "hideFrom": { "legend": false, "tooltip": false, "viz": false },
+            "insertNulls": false,
+            "lineInterpolation": "linear",
+            "lineWidth": 2,
+            "pointSize": 5,
+            "scaleDistribution": { "type": "linear" },
+            "showPoints": "auto",
+            "spanNulls": false,
+            "stacking": { "group": "A", "mode": "none" },
+            "thresholdsStyle": { "mode": "off" }
+          },
+          "mappings": [],
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              { "color": "green", "value": null },
+              { "color": "red", "value": 80 }
+            ]
+          },
+          "unit": "s"
+        },
+        "overrides": []
+      },
+      "options": {
+        "legend": { "calcs": ["mean", "max"], "displayMode": "list", "placement": "bottom", "showLegend": true },
+        "tooltip": { "mode": "single", "sort": "none" }
+      },
+      "targets": [
+        {
+          "datasource": { "type": "prometheus", "uid": "prometheus" },
+          "editorMode": "code",
+          "expr": "histogram_quantile(0.5, rate(ollama_request_duration_seconds_bucket[5m]))",
+          "instant": false,
+          "legendFormat": "p50",
+          "range": true,
+          "refId": "A"
+        }
+      ]
+    },
+    {
+      "id": 2,
+      "type": "timeseries",
+      "title": "Inference Latency p95",
+      "description": "95th percentile of Ollama request duration over a 5-minute window",
+      "gridPos": { "h": 8, "w": 8, "x": 8, "y": 0 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "fieldConfig": {
+        "defaults": {
+          "color": { "mode": "palette-classic" },
+          "custom": {
+            "axisBorderShow": false,
+            "axisCenteredZero": false,
+            "axisColorMode": "text",
+            "axisLabel": "",
+            "axisPlacement": "auto",
+            "barAlignment": 0,
+            "drawStyle": "line",
+            "fillOpacity": 10,
+            "gradientMode": "none",
+            "hideFrom": { "legend": false, "tooltip": false, "viz": false },
+            "insertNulls": false,
+            "lineInterpolation": "linear",
+            "lineWidth": 2,
+            "pointSize": 5,
+            "scaleDistribution": { "type": "linear" },
+            "showPoints": "auto",
+            "spanNulls": false,
+            "stacking": { "group": "A", "mode": "none" },
+            "thresholdsStyle": { "mode": "off" }
+          },
+          "mappings": [],
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              { "color": "green", "value": null },
+              { "color": "red", "value": 80 }
+            ]
+          },
+          "unit": "s"
+        },
+        "overrides": []
+      },
+      "options": {
+        "legend": { "calcs": ["mean", "max"], "displayMode": "list", "placement": "bottom", "showLegend": true },
+        "tooltip": { "mode": "single", "sort": "none" }
+      },
+      "targets": [
+        {
+          "datasource": { "type": "prometheus", "uid": "prometheus" },
+          "editorMode": "code",
+          "expr": "histogram_quantile(0.95, rate(ollama_request_duration_seconds_bucket[5m]))",
+          "instant": false,
+          "legendFormat": "p95",
+          "range": true,
+          "refId": "A"
+        }
+      ]
+    },
+    {
+      "id": 3,
+      "type": "timeseries",
+      "title": "Request Rate",
+      "description": "Ollama requests per second over a 5-minute window",
+      "gridPos": { "h": 8, "w": 8, "x": 16, "y": 0 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "fieldConfig": {
+        "defaults": {
+          "color": { "mode": "palette-classic" },
+          "custom": {
+            "axisBorderShow": false,
+            "axisCenteredZero": false,
+            "axisColorMode": "text",
+            "axisLabel": "",
+            "axisPlacement": "auto",
+            "barAlignment": 0,
+            "drawStyle": "line",
+            "fillOpacity": 10,
+            "gradientMode": "none",
+            "hideFrom": { "legend": false, "tooltip": false, "viz": false },
+            "insertNulls": false,
+            "lineInterpolation": "linear",
+            "lineWidth": 2,
+            "pointSize": 5,
+            "scaleDistribution": { "type": "linear" },
+            "showPoints": "auto",
+            "spanNulls": false,
+            "stacking": { "group": "A", "mode": "none" },
+            "thresholdsStyle": { "mode": "off" }
+          },
+          "mappings": [],
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              { "color": "green", "value": null },
+              { "color": "red", "value": 80 }
+            ]
+          },
+          "unit": "reqps"
+        },
+        "overrides": []
+      },
+      "options": {
+        "legend": { "calcs": ["mean", "max"], "displayMode": "list", "placement": "bottom", "showLegend": true },
+        "tooltip": { "mode": "single", "sort": "none" }
+      },
+      "targets": [
+        {
+          "datasource": { "type": "prometheus", "uid": "prometheus" },
+          "editorMode": "code",
+          "expr": "rate(ollama_requests_total[5m])",
+          "instant": false,
+          "legendFormat": "req/s",
+          "range": true,
+          "refId": "A"
+        }
+      ]
+    }
+  ],
+  "preload": false,
+  "templating": {
+    "list": []
+  }
+}
--- a/infra/observability/prometheus/prometheus.yml
+++ b/infra/observability/prometheus/prometheus.yml
@@ -20,4 +20,10 @@ scrape_configs:
  - job_name: ocr-service
    metrics_path: /metrics
    static_configs:
-      - targets: ['ocr:8000']
+      - targets: ['ocr-service:8000']
+
+  - job_name: ollama
+    metrics_path: /metrics
+    static_configs:
+      # Uses the Docker service name for reliable DNS resolution.
+      - targets: ['ollama:11434']