getsentry · andreiborza · Jun 24, 2026 · Jun 24, 2026 · nicohrubec · Jun 24, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -4,6 +4,14 @@
 
 - "You miss 100 percent of the chances you don't take. — Wayne Gretzky" — Michael Scott
 
+### Important Changes
+
+- **feat(core): Enable `streamGenAiSpans` by default ([#21732](https://github.com/getsentry/sentry-javascript/pull/21732))**
+
+  The SDK now extracts all `gen_ai` spans out of a transaction and sends them as v2 envelope items by default. This prevents gen_ai spans from being dropped when the transaction payload exceeds size limits. Because they are no longer constrained by transaction size limits, AI message data is also no longer truncated by default. Set `enableTruncation: true` on the respective AI integration to re-enable truncation. To keep the previous behavior, set `streamGenAiSpans: false`.
+
+  Self-hosted Sentry users should opt out with `streamGenAiSpans: false`, since streamed gen_ai spans may not be ingested by their Sentry instance.
+
 ## 10.60.0
 
 ### Other Changes

@@ -6,6 +6,8 @@ Sentry.init({
   release: '1.0',
   tracesSampleRate: 1.0,
   dataCollection: { genAI: { inputs: true, outputs: true } },
+  // This suite asserts on gen_ai spans embedded in the transaction, so opt out of span streaming.
+  streamGenAiSpans: false,
   transport: loggingTransport,
   beforeSendTransaction: event => {
     // Filter out mock express server transactions

diff --git a/packages/core/src/tracing/ai/utils.ts b/packages/core/src/tracing/ai/utils.ts
@@ -63,7 +63,7 @@ export function resolveAIRecordingOptions<T extends AIRecordingOptions>(options?
  * Otherwise, truncation is disabled whenever gen_ai spans are sent through the span streaming / v2
  * span path, i.e. full span streaming (`traceLifecycle: 'stream'`) or `streamGenAiSpans`. That path
  * is not subject to the transaction payload-size limits that truncation works around, so the full
- * message data can be retained.
+ * message data can be retained. `streamGenAiSpans` is opt-out (on unless explicitly set to `false`).
  */
 export function shouldEnableTruncation(enableTruncation: boolean | undefined): boolean {
   if (enableTruncation !== undefined) {
@@ -75,7 +75,7 @@ export function shouldEnableTruncation(enableTruncation: boolean | undefined): b
     return true;
   }
 
-  return !hasSpanStreamingEnabled(client) && !client.getOptions().streamGenAiSpans;
+  return !hasSpanStreamingEnabled(client) && client.getOptions().streamGenAiSpans === false;
 }
 
 /**

diff --git a/packages/core/src/tracing/spans/extractGenAiSpans.ts b/packages/core/src/tracing/spans/extractGenAiSpans.ts
@@ -20,7 +20,7 @@ export function extractGenAiSpansFromEvent(event: Event, client: Client): SpanCo
     event.type !== 'transaction' ||
     !event.spans?.length ||
     !event.sdkProcessingMetadata?.hasGenAiSpans ||
-    !client.getOptions().streamGenAiSpans ||
+    client.getOptions().streamGenAiSpans === false ||
     hasSpanStreamingEnabled(client)
   ) {
     return undefined;

diff --git a/packages/core/src/types/options.ts b/packages/core/src/types/options.ts
@@ -565,16 +565,16 @@ export interface ClientOptions<TO extends BaseTransportOptions = BaseTransportOp
   orgId?: `${number}` | number;
 
   /**
-   * If set to `true`, gen_ai spans will be extracted from transactions and sent as v2 span envelope items.
+   * Unless set to `false`, gen_ai spans will be extracted from transactions and sent as v2 span envelope items.
    *
    * This enables streaming gen_ai spans, avoiding payload size limits of usual transactions.
    *
    * Because the v2 span format is not subject to the transaction payload-size limits that gen_ai message
-   * truncation exists to work around, enabling this option also disables gen_ai input truncation (and the
-   * inline-media redaction that rides along with it) by default. Set `enableTruncation: true` on the
-   * respective AI integration to opt back into truncation.
+   * truncation exists to work around, this also disables gen_ai input truncation by default. Set
+   * `enableTruncation: true` on the respective AI integration to opt back into truncation, or set this
+   * option to `false` to send gen_ai spans as part of the transaction (which re-enables truncation by default).
    *
-   * @default false
+   * @default true
    */
   streamGenAiSpans?: boolean;
 

diff --git a/packages/core/test/lib/tracing/ai/utils.test.ts b/packages/core/test/lib/tracing/ai/utils.test.ts
@@ -92,6 +92,7 @@ describe('shouldEnableTruncation', () => {
     getCurrentScope().clear();
     getIsolationScope().clear();
     getGlobalScope().clear();
+    getCurrentScope().setClient(undefined);
   });
 
   afterEach(() => {
@@ -110,8 +111,13 @@ describe('shouldEnableTruncation', () => {
     expect(shouldEnableTruncation(undefined)).toBe(true);
   });
 
-  it('defaults to true with a default client (no streaming)', () => {
+  it('defaults to false with a default client (streamGenAiSpans is opt-out)', () => {
     setupClient();
+    expect(shouldEnableTruncation(undefined)).toBe(false);
+  });
+
+  it('defaults to true when streamGenAiSpans is explicitly disabled', () => {
+    setupClient({ streamGenAiSpans: false });
     expect(shouldEnableTruncation(undefined)).toBe(true);
   });
 

diff --git a/packages/core/test/lib/tracing/spans/extractGenAiSpans.test.ts b/packages/core/test/lib/tracing/spans/extractGenAiSpans.test.ts
@@ -37,7 +37,6 @@ function makeClient(options: Partial<Parameters<typeof getDefaultTestClientOptio
   return new TestClient(
     getDefaultTestClientOptions({
       dsn: 'https://dsn@ingest.f00.f00/1',
-      streamGenAiSpans: true,
       ...options,
     }),
   );
@@ -121,7 +120,17 @@ describe('extractGenAiSpansFromEvent', () => {
     expect(extractGenAiSpansFromEvent(event, makeClient())).toBeUndefined();
   });
 
-  it('returns undefined when streamGenAiSpans is not enabled', () => {
+  it('extracts gen_ai spans by default when streamGenAiSpans is unset', () => {
+    const event = makeTransactionEvent([makeSpanJSON({ op: 'gen_ai.chat', span_id: 'genai001' })], true);
+
+    const result = extractGenAiSpansFromEvent(event, makeClient());
+
+    expect(result).toBeDefined();
+    expect(result![0].item_count).toBe(1);
+    expect(event.spans).toHaveLength(0);
+  });
+
+  it('returns undefined when streamGenAiSpans is explicitly disabled', () => {
     const event = makeTransactionEvent([makeSpanJSON({ op: 'gen_ai.chat' })]);
     const client = makeClient({ streamGenAiSpans: false });