diff --git a/inventories/development/tools.yml.j2 b/inventories/development/tools.yml.j2
index d8df933297..fa231e4418 100644
--- a/inventories/development/tools.yml.j2
+++ b/inventories/development/tools.yml.j2
@@ -40,6 +40,18 @@ applications:
     services:
       ollama:
         cpus: "{{ lookup('resource', 'host_cpus') }}"
+        min_storage: 4600MB
+        preload_models:
+          - alias: qwen2.5:3b
+            context: 32768
+            name: qwen2.5-3b-instruct
+            repo: lmstudio-community/Qwen2.5-3B-Instruct-GGUF
+            url: https://huggingface.co/lmstudio-community/Qwen2.5-3B-Instruct-GGUF/resolve/main/Qwen2.5-3B-Instruct-Q4_K_M.gguf
+            source: https://huggingface.co/lmstudio-community/Qwen2.5-3B-Instruct-GGUF
+            quant: q4_k_m
+            file: Qwen2.5-3B-Instruct-Q4_K_M.gguf
+            sha256: "d44e2c5d1ec3cae1d5cf6a744bee528e46c65a1e66e741fa92730967e7d625bb"
+            bytes: 1929903008
   svc-ai-s1:
     secrets:
       credentials:
diff --git a/tests/unit/python/utils/inventory/test_tools_model.py b/tests/unit/python/utils/inventory/test_tools_model.py
new file mode 100644
index 0000000000..ddc2cdd30c
--- /dev/null
+++ b/tests/unit/python/utils/inventory/test_tools_model.py
@@ -0,0 +1,36 @@
+"""The alias the readiness probe expects is the one the lane deploys."""
+
+from __future__ import annotations
+
+import unittest
+
+import yaml
+
+from utils import PROJECT_ROOT
+from utils.inventory.tools import LANE_MODEL_ALIAS, render
+
+OLLAMA_ROLE = "svc-ai-ollama"
+
+
+class TestLaneModel(unittest.TestCase):
+    def test_the_rendered_inventory_pins_the_alias_the_probe_expects(self) -> None:
+        """Two places name the model, and nothing else keeps them equal.
+
+        The inventory carries the entry because that is where inventory data
+        belongs, and the probe needs the bare alias before any render has
+        run, because ``ensure`` asks whether the lane serves before the
+        deploy that writes the rendered file. A lane serving one model while
+        the probe expects another never reports ready and redeploys on every
+        call.
+        """
+        rendered = yaml.safe_load(render(PROJECT_ROOT, ["en"]).read_text("utf-8"))
+        ollama = rendered["applications"][OLLAMA_ROLE]["services"]["ollama"]
+
+        self.assertEqual(
+            [LANE_MODEL_ALIAS],
+            [model["alias"] for model in ollama["preload_models"]],
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/utils/inventory/tools.py b/utils/inventory/tools.py
index ae38df1e97..0d53c144ba 100644
--- a/utils/inventory/tools.py
+++ b/utils/inventory/tools.py
@@ -30,6 +30,14 @@ template already carries. Never reuse it anywhere a real caller could reach.
 and under the same standing. A deployment hands each consumer a virtual key
 instead, which is what the role's secret description asks for; the oracle
 judge is not a deployed consumer and has no vault to read one from.
+
+The lane pins its own model in the template, rather than taking the
+repository's ``AI_LOCAL_MODELS`` default: the oracle needs a judge that can
+tell a faithful rendering from an unrelated text, a demand no deployment
+makes. :data:`LANE_MODEL_ALIAS` repeats that one name here because the
+readiness probe compares it against what the gateway actually serves, and
+the rendered inventory is written by the deploy this probe runs before. The
+two are held together by a test that renders the template.
 """
 
 from __future__ import annotations
@@ -54,7 +62,7 @@ LITELLM_ROLE = "svc-ai-litellm"
 OLLAMA_ROLE = "svc-ai-ollama"
 GATEWAY_ROLE = "web-svc-translate"
 LANE_SERVICE = "tools"
-LOCAL_MODELS_FILE = Path("group_vars") / "all" / "16_ai.yml"
+LANE_MODEL_ALIAS = "qwen2.5:3b"
 PROBE_TIMEOUT_SECONDS = 10
 
 
@@ -232,7 +240,7 @@ def _litellm_serves_the_named_model(root: Path) -> bool:
     """
     import json
 
-    alias = _named_model(root)
+    alias = _named_model()
     body = _answers(f"{host_url(root, LITELLM_ROLE, 'litellm')}/models", LITELLM_KEY)
     if body is None or not alias:
         return False
@@ -243,15 +251,9 @@ def _litellm_serves_the_named_model(root: Path) -> bool:
     return alias in served
 
 
-def _named_model(root: Path) -> str:
-    """Return the alias the lane's model server is expected to serve.
-
-    Args:
-        root: repository root.
-    """
-    models = load_yaml(root / LOCAL_MODELS_FILE).get("AI_LOCAL_MODELS") or []
-    first = models[0] if models else {}
-    return str(first.get("alias") or "") if isinstance(first, dict) else ""
+def _named_model() -> str:
+    """Return the alias the lane's model server is expected to serve."""
+    return LANE_MODEL_ALIAS
 
 
 def _gateway_covers_the_catalogs(root: Path) -> bool: