diff --git a/roles/svc-ai-lmstudio/tasks/01_manager_ops.yml b/roles/svc-ai-lmstudio/tasks/01_manager_ops.yml
index 60f97c71e8..24a691460d 100644
--- a/roles/svc-ai-lmstudio/tasks/01_manager_ops.yml
+++ b/roles/svc-ai-lmstudio/tasks/01_manager_ops.yml
@@ -6,36 +6,10 @@
     resolve_output_prefix: LMSTUDIO
     resolve_compose_cid: "{{ LMSTUDIO_CONTAINER }}"
 
-- name: ⬇️ Pre-pull LM Studio models
-  ansible.builtin.shell: > # nocheck: container-exec-resolver  runtime-resolved CID, delegated to the hosting node
-    container exec -i {{ LMSTUDIO_LOCAL_CID | quote }} lms get --yes --gguf {{ model.source | quote }}
-  register: lmstudio_pull_result
-  loop: "{{ LMSTUDIO_PRELOAD_MODELS }}"
-  loop_control:
-    loop_var: model
-    label: "{{ model.alias }}"
-  # rationale: async; model downloads run overlapped, results reaped below before playwright
-  async: "{{ LMSTUDIO_PRELOAD_ASYNC_TIME if ASYNC_ENABLED | bool else omit }}"
-  poll: "{{ ASYNC_POLL if ASYNC_ENABLED | bool else omit }}"
-  retries: "{{ 0 if ASYNC_ENABLED | bool else 5 }}"
-  delay: 15
-  until: >
-    (lmstudio_pull_result.rc | default(0)) == 0 or
-    ('already downloaded' in (lmstudio_pull_result.stdout | default('')))
-  changed_when: false
-  failed_when: >-
-    (lmstudio_pull_result.failed | default(false))
-    and 'already downloaded' not in ((lmstudio_pull_result.stdout | default('')) ~ (lmstudio_pull_result.stderr | default('')))
-  delegate_to: "{{ LMSTUDIO_HOST_NODE }}"
-
-- name: "⏳ Reap the LM Studio model pre-pulls"
-  ansible.builtin.include_tasks:
-    file: utils/async/reap.yml
-    apply:
-      delegate_to: "{{ LMSTUDIO_HOST_NODE }}"
-  vars:
-    async_reap_jobs: "{{ lmstudio_pull_result.results | default([]) }}"
-    async_reap_label: "LM Studio model pre-pull"
-    async_reap_tolerate:
-      - "already downloaded"
-    ASYNC_TIME: "{{ LMSTUDIO_PRELOAD_ASYNC_TIME }}"
+- name: "⬇️ Preload the LM Studio models"
+  block:
+    - name: "⬇️ Preload the LM Studio models (first pass)"
+      ansible.builtin.include_tasks: utils/preload.yml
+  rescue:
+    - name: "♻️ Recover the wedged transfer and preload again"
+      ansible.builtin.include_tasks: utils/recover.yml
diff --git a/roles/svc-ai-lmstudio/tasks/utils/preload.yml b/roles/svc-ai-lmstudio/tasks/utils/preload.yml
new file mode 100644
index 0000000000..9c668a977a
--- /dev/null
+++ b/roles/svc-ai-lmstudio/tasks/utils/preload.yml
@@ -0,0 +1,33 @@
+---
+- name: ⬇️ Pre-pull LM Studio models
+  ansible.builtin.shell: > # nocheck: container-exec-resolver  runtime-resolved CID, delegated to the hosting node
+    container exec -i {{ LMSTUDIO_LOCAL_CID | quote }} lms get --yes --gguf {{ model.source | quote }}
+  register: lmstudio_pull_result
+  loop: "{{ LMSTUDIO_PRELOAD_MODELS }}"
+  loop_control:
+    loop_var: model
+    label: "{{ model.alias }}"
+  async: "{{ LMSTUDIO_PRELOAD_ASYNC_TIME if ASYNC_ENABLED | bool else omit }}"
+  poll: "{{ ASYNC_POLL if ASYNC_ENABLED | bool else omit }}"
+  retries: "{{ 0 if ASYNC_ENABLED | bool else 1 }}"
+  delay: 15
+  until: >
+    (lmstudio_pull_result.rc | default(0)) == 0 or
+    ('already downloaded' in (lmstudio_pull_result.stdout | default('')))
+  changed_when: false
+  failed_when: >-
+    (lmstudio_pull_result.failed | default(false))
+    and 'already downloaded' not in ((lmstudio_pull_result.stdout | default('')) ~ (lmstudio_pull_result.stderr | default('')))
+  delegate_to: "{{ LMSTUDIO_HOST_NODE }}"
+
+- name: "⏳ Reap the LM Studio model pre-pulls"
+  ansible.builtin.include_tasks:
+    file: utils/async/reap.yml
+    apply:
+      delegate_to: "{{ LMSTUDIO_HOST_NODE }}"
+  vars:
+    async_reap_jobs: "{{ lmstudio_pull_result.results | default([]) }}"
+    async_reap_label: "LM Studio model pre-pull"
+    async_reap_tolerate:
+      - "already downloaded"
+    ASYNC_TIME: "{{ LMSTUDIO_PRELOAD_ASYNC_TIME }}"
diff --git a/roles/svc-ai-lmstudio/tasks/utils/recover.yml b/roles/svc-ai-lmstudio/tasks/utils/recover.yml
new file mode 100644
index 0000000000..cea5918d25
--- /dev/null
+++ b/roles/svc-ai-lmstudio/tasks/utils/recover.yml
@@ -0,0 +1,28 @@
+---
+- name: "🩺 Dump the LM Studio transfer record"
+  ansible.builtin.shell: > # nocheck: container-exec-resolver  runtime-resolved CID, delegated to the hosting node
+    container exec -i {{ LMSTUDIO_LOCAL_CID | quote }}
+    cat /root/.lmstudio/.internal/single-downloads-info.json
+  register: lmstudio_transfer_record
+  changed_when: false
+  failed_when: false
+  delegate_to: "{{ LMSTUDIO_HOST_NODE }}"
+
+- name: "🩺 Report the LM Studio transfer record"
+  ansible.builtin.debug:
+    var: lmstudio_transfer_record.stdout
+
+- name: "♻️ Restart LM Studio to release the wedged transfer"
+  ansible.builtin.command: container restart {{ LMSTUDIO_LOCAL_CID }}
+  changed_when: true
+  delegate_to: "{{ LMSTUDIO_HOST_NODE }}"
+
+- name: "🔍 Re-resolve the {{ application_id }} hosting node + container id"
+  ansible.builtin.include_tasks: "{{ lookup('path_absolute', 'roles/sys-svc-compose/tasks/utils/swarm/resolve_host_cid.yml') }}"
+  vars:
+    resolve_service_key: "{{ entity_name }}"
+    resolve_output_prefix: LMSTUDIO
+    resolve_compose_cid: "{{ LMSTUDIO_CONTAINER }}"
+
+- name: "⬇️ Preload the LM Studio models (after the restart)"
+  ansible.builtin.include_tasks: utils/preload.yml