Tolerate a Chutes stream chunk that omits id

crates/api/src/routes/completions.rs

@@ -614,12 +614,12 @@ fn build_final_usage_chunk_bytes(
 }
 
 // Helper function to extract inference ID from a parsed stream chunk
-fn extract_inference_id_from_chunk(chunk: &inference_providers::StreamChunk) -> Uuid {
+fn extract_inference_id_from_chunk(chunk: &inference_providers::StreamChunk) -> Option<Uuid> {
     let id = match chunk {
         inference_providers::StreamChunk::Chat(c) => &c.id,
         inference_providers::StreamChunk::Text(c) => &c.id,
     };
-    hash_inference_id_to_uuid(id)
+    (!id.is_empty()).then(|| hash_inference_id_to_uuid(id))
 }
 
 // Convert MessageContent to serde_json::Value, preserving multimodal parts (images, audio, etc.)
@@ -1635,8 +1635,9 @@ async fn chat_completions_inner(
                                 stream_chat_id = Some(match chunk {
                                     inference_providers::StreamChunk::Chat(c) => c.id.clone(),
                                     inference_providers::StreamChunk::Text(c) => c.id.clone(),
-                                });
-                                break Some(extract_inference_id_from_chunk(chunk));
+                                })
+                                .filter(|id| !id.is_empty());
+                                break extract_inference_id_from_chunk(chunk);
                             }
                             true
                         }
@@ -2730,8 +2731,9 @@ async fn completions_inner(
                                 stream_chat_id = Some(match chunk {
                                     inference_providers::StreamChunk::Chat(c) => c.id.clone(),
                                     inference_providers::StreamChunk::Text(c) => c.id.clone(),
-                                });
-                                break Some(extract_inference_id_from_chunk(chunk));
+                                })
+                                .filter(|id| !id.is_empty());
+                                break extract_inference_id_from_chunk(chunk);
                             }
                             true
                         }
@@ -4162,13 +4164,14 @@ mod tests {
 
     #[test]
     fn test_extract_inference_id_from_chunk_empty_id() {
+        // Given: Chutes omitted its provider inference id at the parsing boundary.
         let chunk = make_chat_chunk("");
+
+        // When: the route derives the optional public inference id.
         let result = extract_inference_id_from_chunk(&chunk);
-        // Empty string should still produce a valid UUID
-        assert!(
-            !result.is_nil(),
-            "empty provider ID should still produce a non-nil UUID"
-        );
+
+        // Then: absence stays absent instead of becoming a hash of the empty string.
+        assert_eq!(result, None);
     }
 
     fn empty_delta() -> inference_providers::models::ChatDelta {

crates/inference_providers/src/attested/chutes/e2ee.rs

@@ -166,6 +166,7 @@ pub struct PreparedRequest {
 /// so this must outlive the whole response.
 pub struct ResponseSession {
     response_dk: ml_kem::DecapsulationKey768,
+    synthetic_stream_id: String,
 }
 
 /// Build an E2EE request blob for `request_json` (the OpenAI request body)
@@ -215,11 +216,18 @@ pub fn build_request(
 
     Ok(PreparedRequest {
         blob,
-        session: ResponseSession { response_dk },
+        session: ResponseSession {
+            response_dk,
+            synthetic_stream_id: format!("chutes-gateway-{}", uuid::Uuid::new_v4()),
+        },
     })
 }
 
 impl ResponseSession {
+    pub(super) fn synthetic_stream_id(&self) -> &str {
+        &self.synthetic_stream_id
+    }
+
     /// Decrypt a non-streaming response blob (`mlkem_ct ‖ nonce ‖ ct ‖ tag`,
     /// keyed with `info="e2e-resp-v1"`, gzip-compressed) into the OpenAI
     /// response JSON bytes.