diff --git a/apps/searxng/deployment.yaml b/apps/searxng/deployment.yaml
index 5ab16d1..c8853f6 100644
--- a/apps/searxng/deployment.yaml
+++ b/apps/searxng/deployment.yaml
@@ -35,7 +35,7 @@ spec:
     spec:
       containers:
         - name: searxng-1723974683
-          image: gt.wittyoneoff.com/jason/searxng:ai-summary-20260728-3
+          image: gt.wittyoneoff.com/jason/searxng:ai-summary-20260805
           imagePullPolicy: Always
           env:
             - name: AUTOCOMPLETE
diff --git a/apps/searxng/secret.example.yaml b/apps/searxng/secret.example.yaml
index beccd57..7564bf2 100644
--- a/apps/searxng/secret.example.yaml
+++ b/apps/searxng/secret.example.yaml
@@ -12,3 +12,18 @@ stringData:
     server:
       secret_key: CHANGE-ME
     use_default_settings: true
+
+    # The ai_summary plugin talks to an OpenAI compatible LLM server.
+    # api_key is optional and only needed when that server requires
+    # authentication (e.g. a LiteLLM gateway with a master key, or vLLM /
+    # llama.cpp started with --api-key).  It is only ever sent to base_url.
+    ai_summary:
+      base_url: "http://192.168.1.191:11434"
+      api_key: ""
+      model: "gemma4:e4b"
+      grounding: true
+
+    # NOTE: a plugins: block REPLACES searxng's default plugin list, it does
+    # not merge -- new upstream default plugins have to be added here by hand.
+    plugins:
+      searx.plugins.ai_summary.SXNGPlugin: {active: true}