diff --git a/.cursor/rules/sdk/docs/kv-cache-system.mdc b/.cursor/rules/sdk/docs/kv-cache-system.mdc
index df1307131b..5dc9ea9003 100644
--- a/.cursor/rules/sdk/docs/kv-cache-system.mdc
+++ b/.cursor/rules/sdk/docs/kv-cache-system.mdc
@@ -102,7 +102,7 @@ When a new cache key is used for the first time:
 
 ### Cache Initialization (initSystemPromptCache)
 
-Primes a new cache by sending system prompt + tools to the model with `{ cacheKey: cachePath }` in `runOptions`, then cancels generation after the first output token — it only needs to process the cache priming, not produce a response.
+Primes a new cache by sending system prompt + tools through `runModel` with `{ cacheKey: cachePath, saveCacheToDisk: true, prefill: true }`. The prompt is ingested into the KV cache and persisted to disk without producing any output tokens, so the call resolves as soon as priming finishes.
 
 ### Message Preparation (prepareMessagesForCache)
 
diff --git a/packages/sdk/package.json b/packages/sdk/package.json
index 0f74aa7b5c..434cf19376 100644
--- a/packages/sdk/package.json
+++ b/packages/sdk/package.json
@@ -177,7 +177,7 @@
     "@qvac/embed-llamacpp": "^0.14.0",
     "@qvac/error": "^0.1.1",
     "@qvac/langdetect-text": "^0.1.2",
-    "@qvac/llm-llamacpp": "^0.17.1",
+    "@qvac/llm-llamacpp": "^0.17.3",
     "@qvac/logging": "^0.1.0",
     "@qvac/ocr-onnx": "^0.4.2",
     "@qvac/rag": "^0.4.4",
diff --git a/packages/sdk/server/bare/plugins/llamacpp-completion/ops/completion-stream.ts b/packages/sdk/server/bare/plugins/llamacpp-completion/ops/completion-stream.ts
index efcb0d99c4..517b8025c1 100644
--- a/packages/sdk/server/bare/plugins/llamacpp-completion/ops/completion-stream.ts
+++ b/packages/sdk/server/bare/plugins/llamacpp-completion/ops/completion-stream.ts
@@ -103,7 +103,10 @@ interface ChatHistory {
 // dep bump propagates and is harmless once it has.
 type CompletionGenerationParams = GenerationParams & { json_schema?: string };
 
-type CompletionRunOptions = Pick<RunOptions, "cacheKey" | "saveCacheToDisk"> & {
+type CompletionRunOptions = Pick<
+  RunOptions,
+  "cacheKey" | "saveCacheToDisk" | "prefill"
+> & {
   generationParams?: CompletionGenerationParams;
 };
 
@@ -244,10 +247,7 @@ async function initSystemPromptCache(
   const primeResponse = await runModel(model, primeMessages, {
     cacheKey: cachePathToUse,
     saveCacheToDisk: true,
-  });
-
-  primeResponse.once("output", () => {
-    void primeResponse.cancel();
+    prefill: true,
   });
 
   await primeResponse.await();