fix: ensure output for non-streaming models (#369)

Co-authored-by: mneves75 <mneves75@users.noreply.github.com>

src/agents/pi-embedded-subscribe.test.ts

@@ -400,6 +400,40 @@ describe("subscribeEmbeddedPiSession", () => {
     expect(subscription.assistantTexts).toEqual(["Hello world"]);
   });
 
+  it("populates assistantTexts for non-streaming models with chunking enabled", () => {
+    // Non-streaming models (e.g. zai/glm-4.7): no text_delta events; message_end
+    // must still populate assistantTexts so providers can deliver a final reply.
+    let handler: SessionEventHandler | undefined;
+    const session: StubSession = {
+      subscribe: (fn) => {
+        handler = fn;
+        return () => {};
+      },
+    };
+
+    const subscription = subscribeEmbeddedPiSession({
+      session: session as unknown as Parameters<
+        typeof subscribeEmbeddedPiSession
+      >[0]["session"],
+      runId: "run",
+      blockReplyChunking: { minChars: 50, maxChars: 200 }, // Chunking enabled
+    });
+
+    // Simulate non-streaming model: only message_start and message_end, no text_delta
+    handler?.({ type: "message_start", message: { role: "assistant" } });
+
+    const assistantMessage = {
+      role: "assistant",
+      content: [{ type: "text", text: "Response from non-streaming model" }],
+    } as AssistantMessage;
+
+    handler?.({ type: "message_end", message: assistantMessage });
+
+    expect(subscription.assistantTexts).toEqual([
+      "Response from non-streaming model",
+    ]);
+  });
+
   it("does not append when text_end content is a prefix of deltas", () => {
     let handler: ((evt: unknown) => void) | undefined;
     const session: StubSession = {

src/agents/pi-embedded-subscribe.ts

@@ -516,8 +516,10 @@ export function subscribeEmbeddedPiSession(params: {
 
           const addedDuringMessage =
             assistantTexts.length > assistantTextBaseline;
-          const chunkingEnabled = Boolean(blockChunking);
-          if (!chunkingEnabled && !addedDuringMessage && text) {
+          const chunkerHasBuffered = blockChunker?.hasBuffered() ?? false;
+          // Non-streaming models (no text_delta): ensure assistantTexts gets the
+          // final text when the chunker has nothing buffered to drain.
+          if (!addedDuringMessage && !chunkerHasBuffered && text) {
             const last = assistantTexts.at(-1);
             if (!last || last !== text) assistantTexts.push(text);
           }