fix: retry LLM template model pulls

python/llm/README.md

@@ -62,7 +62,11 @@ http://<device-hostname>:{{.PORT}}
 
 The first start downloads the model in the background — watch the `[ollama]`
 log lines for pull progress. The model list in the UI populates once the pull
-completes.
+completes. If the device does not have network/DNS when the app first starts,
+the Ollama service keeps retrying instead of giving up, so the model appears
+automatically once connectivity is restored. If Open WebUI shows an empty model
+picker, the model is still downloading or the puller is retrying; check the
+Ollama service logs before pulling manually.
 
 > On the device each service runs in its own network namespace, so the
 > Docker service-name URL (`http://ollama:11434`) does not resolve. The
@@ -88,6 +92,6 @@ built-in service-name DNS.
 
 ```sh
 wendy run --detach           # start and return; stream later with:
-wendy device logs
-wendy device ps              # list both containers
+wendy device logs {{.APP_ID}} --service ollama --tail 100
+wendy device apps list       # list both containers
 ```

python/llm/ollama/entrypoint.sh

@@ -5,15 +5,44 @@ mkdir -p "$OLLAMA_MODELS"
 
 # Pull the configured model in the background once the server is up, so the
 # API is reachable immediately and the (potentially large) download streams
-# its progress to the service log.
+# its progress to the service log. Keep retrying because first boot often races
+# network readiness; without retries Open WebUI can stay up with an empty model
+# picker forever after a transient DNS or connectivity failure.
 (
+  export OLLAMA_HOST=127.0.0.1:11434
+
+  echo "Waiting for Ollama API before pulling ${OLLAMA_MODEL}..."
   until curl -fsS "http://127.0.0.1:11434/api/tags" >/dev/null 2>&1; do
     sleep 2
   done
-  if ! OLLAMA_HOST=127.0.0.1:11434 ollama list | awk '{print $1}' | grep -qx "$OLLAMA_MODEL"; then
-    echo "Pulling ${OLLAMA_MODEL} into ${OLLAMA_MODELS}"
-    OLLAMA_HOST=127.0.0.1:11434 ollama pull "$OLLAMA_MODEL"
-  fi
+
+  retry_delay=10
+  max_retry_delay=60
+
+  while true; do
+    if ollama list | awk 'NR > 1 { print $1 }' | grep -qx "$OLLAMA_MODEL"; then
+      echo "Ollama model ${OLLAMA_MODEL} is available in ${OLLAMA_MODELS}."
+      break
+    fi
+
+    echo "First-run setup: pulling ${OLLAMA_MODEL} into ${OLLAMA_MODELS}."
+    echo "Open WebUI may show an empty model list until this download completes."
+
+    if ollama pull "$OLLAMA_MODEL"; then
+      echo "Finished pulling ${OLLAMA_MODEL}."
+      retry_delay=10
+      continue
+    fi
+
+    echo "Pulling ${OLLAMA_MODEL} failed. Network/DNS may not be ready yet, or the model name may be unavailable."
+    echo "Retrying in ${retry_delay}s..."
+    sleep "$retry_delay"
+
+    retry_delay=$((retry_delay * 2))
+    if ((retry_delay > max_retry_delay)); then
+      retry_delay=$max_retry_delay
+    fi
+  done
 ) &
 
 exec ollama serve