diff --git a/README.md b/README.md
index e69de29..2565106 100644
--- a/README.md
+++ b/README.md
@@ -0,0 +1,5 @@
+## DEV
+Run in DEV with command:
+```
+uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload --log-config logging.conf
+```
\ No newline at end of file
diff --git a/app/core/deep_agent.py b/app/core/deep_agent.py
index 4f071a8..5f528f1 100644
--- a/app/core/deep_agent.py
+++ b/app/core/deep_agent.py
@@ -858,25 +858,15 @@ async def _run_single_agent_stream(
                 _gen.update(output=_as_text(response.content), usage_details=extract_usage(response))
                 _gen_ctx.__exit__(None, None, None)
 
-            messages.append(response)
-
             if not response.tool_calls:
-                emitted_any = False
-                async for chunk in llm.astream(messages):
-                    token = _as_text(getattr(chunk, "content", ""))
-                    if token:
-                        streamed_chars += len(token)
-                        streamed_text.append(token)
-                        emitted_any = True
-                        yield "token", token
-
-                # Some providers return final text in `response.content` but stream no chunks.
-                if not emitted_any:
-                    fallback_text = _as_text(response.content)
-                    if fallback_text:
-                        streamed_chars += len(fallback_text)
-                        streamed_text.append(fallback_text)
-                        yield "token", fallback_text
+                # Yield the content from the ainvoke response directly — no second LLM call.
+                # Previously, messages.append(response) was called first, so the re-stream
+                # received [System, Human, AI] and regenerated a response without tools bound.
+                final_text = _as_text(response.content)
+                if final_text:
+                    streamed_chars += len(final_text)
+                    streamed_text.append(final_text)
+                    yield "token", final_text
                 logger.info(
                     "deep_agent: run_single_agent_stream_end trace=%s user=%s tool_calls=%d response_chars=%d",
                     trace_id or "-",
@@ -888,6 +878,7 @@ async def _run_single_agent_stream(
                     _span.update(output="".join(streamed_text))
                 return
 
+            messages.append(response)
             tool_map = {tool_def.name: tool_def for tool_def in tools}
             for call in response.tool_calls:
                 tool_calls_count += 1