diff --git a/src/agent/message.ts b/src/agent/message.ts
index 282da08..8de4580 100644
--- a/src/agent/message.ts
+++ b/src/agent/message.ts
@@ -23,9 +23,11 @@ export async function sendMessageStream(
     background?: boolean;
     // add more later: includePings, request timeouts, etc.
   } = { streamTokens: true, background: true },
+  // TODO: Re-enable once issues are resolved - disabled retries were causing problems
   // Disable SDK retries by default - state management happens outside the stream,
   // so retries would violate idempotency and create race conditions
-  requestOptions: { maxRetries?: number } = { maxRetries: 0 },
+  // requestOptions: { maxRetries?: number } = { maxRetries: 0 },
+  requestOptions: { maxRetries?: number } = {},
 ): Promise<Stream<LettaStreamingResponse>> {
   // Capture request start time for TTFT measurement when timings are enabled
   const requestStartTime = isTimingsEnabled() ? performance.now() : undefined;
diff --git a/src/cli/helpers/stream.ts b/src/cli/helpers/stream.ts
index 299ba08..67c6327 100644
--- a/src/cli/helpers/stream.ts
+++ b/src/cli/helpers/stream.ts
@@ -396,6 +396,7 @@ export async function drainStreamWithResume(
       buffers.interrupted = false;
 
       // Resume from Redis where we left off
+      // TODO: Re-enable once issues are resolved - disabled retries were causing problems
       // Disable SDK retries - state management happens outside, retries would create race conditions
       const resumeStream = await client.runs.messages.stream(
         result.lastRunId,
@@ -403,7 +404,7 @@ export async function drainStreamWithResume(
           starting_after: result.lastSeqId,
           batch_size: 1000, // Fetch buffered chunks quickly
         },
-        { maxRetries: 0 },
+        // { maxRetries: 0 },
       );
 
       // Continue draining from where we left off