diff --git a/src/api/endpoints.py b/src/api/endpoints.py
index 852b1d9..c290d31 100644
--- a/src/api/endpoints.py
+++ b/src/api/endpoints.py
@@ -15,6 +15,7 @@ import time
 
 router = APIRouter()
 FORM_EXTRACT_MODULE_NAME = "文本内容提取事故信息"
+STATE_TAG_PATTERN = re.compile(r"<state>\s*(\d+)\s*</state>", flags=re.DOTALL)
 STATUS_CODE_MAP = {
     '0000': '结束通话',
     '0001': '转接人工',
@@ -50,27 +51,12 @@ def normalize_stage_code(stage_code: str) -> str:
     return stage_code
 
 
-def extract_state_and_content(data1: str) -> dict | None:
-    """
-    Extracts the state and content from a string in the format <state>STATE</state>content.
-
-    Args:
-        data1: The input string.
-
-    Returns:
-        A dictionary with 'state' and 'content' keys if a match is found,
-        otherwise None.
-    """
-    data1 = data1.strip()
-    regex = r"<state>(.*?)</state>(.*)"
-    match = re.search(regex, data1, flags=re.DOTALL)
-    
-    if match:
-        return {
-            "state": match.group(1),
-            "content": match.group(2),
-        }
-    return None
+def extract_first_state_and_clean_content(text: str) -> tuple[str | None, str]:
+    """Return the first state code and content with all state tags removed."""
+    match = STATE_TAG_PATTERN.search(text)
+    if not match:
+        return None, text
+    return match.group(1), STATE_TAG_PATTERN.sub("", text)
 
 
 def parse_json_value(value):
@@ -104,7 +90,7 @@ def extract_form_update_from_flow_nodes(nodes):
         if not isinstance(extract_result, dict):
             return {}
 
-        form_update = extract_result.get("formUpdate") or extract_result.get("form") or ""
+        form_update = extract_result.get("formUpdate", "")
         if not form_update:
             return {}
         return parse_json_value(form_update)
@@ -221,6 +207,7 @@ async def chat(
                 )
                 
                 buffer = ""
+                state_filter_buffer = ""
                 state_code_found = False
                 module_form_sent = False
 
@@ -255,6 +242,41 @@ async def chat(
                     if not chunk:
                         return []
                     return [flush_text_delta(chunk)]
+
+                def clean_later_state_tags(text: str, *, final: bool = False) -> str:
+                    nonlocal state_filter_buffer
+                    state_filter_buffer += text
+                    state_filter_buffer = STATE_TAG_PATTERN.sub("", state_filter_buffer)
+
+                    state_start_index = state_filter_buffer.find("<state")
+                    if state_start_index >= 0:
+                        cleaned = state_filter_buffer[:state_start_index]
+                        state_filter_buffer = state_filter_buffer[state_start_index:]
+                        return cleaned
+
+                    if final:
+                        if state_filter_buffer.startswith("<state"):
+                            state_filter_buffer = ""
+                            return ""
+                        cleaned = state_filter_buffer
+                        state_filter_buffer = ""
+                        return cleaned
+
+                    hold_length = 0
+                    state_prefix = "<state>"
+                    max_suffix_length = min(len(state_filter_buffer), len(state_prefix) - 1)
+                    for suffix_length in range(1, max_suffix_length + 1):
+                        if state_prefix.startswith(state_filter_buffer[-suffix_length:]):
+                            hold_length = suffix_length
+
+                    if hold_length:
+                        cleaned = state_filter_buffer[:-hold_length]
+                        state_filter_buffer = state_filter_buffer[-hold_length:]
+                        return cleaned
+
+                    cleaned = state_filter_buffer
+                    state_filter_buffer = ""
+                    return cleaned
                 
                 async for event in aiter_stream_events(response):
                     try:
@@ -298,9 +320,8 @@ async def chat(
 
                         if not state_code_found:
                             # Check for <state>XXXX</state> pattern
-                            match = re.search(r"<state>(.*?)</state>", buffer, flags=re.DOTALL)
-                            if match:
-                                state_code = match.group(1)
+                            state_code, cleaned_content = extract_first_state_and_clean_content(buffer)
+                            if state_code:
 
                                 # Apply logic to map/adjust state code
                                 nextStageCode = normalize_stage_code(state_code)
@@ -319,15 +340,15 @@ async def chat(
 
                                 state_code_found = True
 
-                                # Send remaining content as text_delta
-                                remaining_content = buffer[match.end():]
-                                if remaining_content:
-                                    for text_event in build_text_delta_events(remaining_content):
+                                if cleaned_content:
+                                    for text_event in build_text_delta_events(cleaned_content):
                                         yield text_event
                                 buffer = "" # Clear buffer after extracting state
                         else:
-                            for text_event in build_text_delta_events(delta_content):
-                                yield text_event
+                            cleaned_content = clean_later_state_tags(delta_content)
+                            if cleaned_content:
+                                for text_event in build_text_delta_events(cleaned_content):
+                                    yield text_event
                             buffer = ""
 
                     except Exception as e:
@@ -339,6 +360,11 @@ async def chat(
                 if not state_code_found and buffer:
                     for text_event in build_text_delta_events(buffer):
                         yield text_event
+                elif state_code_found:
+                    cleaned_content = clean_later_state_tags("", final=True)
+                    if cleaned_content:
+                        for text_event in build_text_delta_events(cleaned_content):
+                            yield text_event
 
                 for text_event in flush_text_chunker_events():
                     yield text_event
@@ -470,11 +496,12 @@ async def chat(
         
         if isinstance(content, str):
             logger.debug("content是一个str")
-            state_and_content = extract_state_and_content(content)
-            if state_and_content:
-                logger.debug(f"解析后的state和content为: {state_and_content}")
-                content_stage_code = state_and_content['state']
-                content = state_and_content['content']
+            content_stage_code, content = extract_first_state_and_clean_content(content)
+            if content_stage_code:
+                logger.debug(
+                    f"解析后的第一个state为: {content_stage_code}, "
+                    f"移除state标签后的content为: {content}"
+                )
             else:
                 raise ValueError("大模型回复中的state解析失败")
         else: