diff --git a/src/utils/ai.js b/src/utils/ai.js
index 7cbc48ee..8c0b6f79 100644
--- a/src/utils/ai.js
+++ b/src/utils/ai.js
@@ -356,6 +356,15 @@ async function callBedrock(messages, model, temperature, maxTokens) {
 // ============================================================
 // CALL LITELLM (OpenAI-compatible proxy)
 // ============================================================
+// A reasoning model spends part of max_tokens thinking before it writes. When
+// the budget is small (a query rewrite at 80, a milestone note at 500) it can
+// hit the limit with the whole budget spent on reasoning and nothing written:
+// a 200, finish_reason=length, content ''. Every caller then saw an empty
+// reply — the take-home sheet was the one the user noticed. The retry gives
+// the same request room to finish and asks for low reasoning effort; it fires
+// only for that signature, never for a genuinely empty answer.
+var REASONING_RETRY_FLOOR = 6000;
+
 async function callLiteLLM(messages, model, temperature, maxTokens, generation) {
   if (!litellmClient) throw new Error('LiteLLM not configured. Set LITELLM_API_BASE in .env');
 
@@ -375,6 +384,13 @@ async function callLiteLLM(messages, model, temperature, maxTokens, generation)
     var reasoningChars = String(choice.message.reasoning_content || choice.message.reasoning || '').length;
     console.warn('[ai] empty reply from ' + model + ': finish_reason=' + (choice.finish_reason || '?') +
       ' completion_tokens=' + (completion.usage && completion.usage.completion_tokens) + ' reasoning_chars=' + reasoningChars);
+    var starved = choice.finish_reason === 'length' && reasoningChars > 0 && !(generation && generation.reasoningRetried);
+    if (starved) {
+      var retryBudget = Math.max(REASONING_RETRY_FLOOR, (maxTokens || 0) * 4);
+      console.warn('[ai] reasoning consumed the budget; retrying ' + model + ' with max_tokens=' + retryBudget);
+      return callLiteLLM(messages, model, temperature, retryBudget,
+        Object.assign({}, generation || {}, { reasoningEffort: 'low', reasoningRetried: true }));
+    }
   }
   return {
     success: true,