botrun-flow-lang 5.12.263__py3-none-any.whl → 6.2.21__py3-none-any.whl

botrun_flow_lang/langgraph_agents/agents/util/perplexity_search.py

@@ -1,464 +1,464 @@
-from copy import deepcopy
-from typing import AsyncGenerator
-from pydantic import BaseModel
-import os
-import json
-import aiohttp
-from dotenv import load_dotenv
-import re
-import logging
-
-load_dotenv()
-
-
-class PerplexitySearchEvent(BaseModel):
-    chunk: str
-    raw_json: dict | None = None
-
-
-PERPLEXITY_API_URL = "https://api.perplexity.ai/chat/completions"
-OPENROUTER_API_URL = "https://openrouter.ai/api/v1/chat/completions"
-
-
-def should_include_citation(citation: str, domain_filter: list[str]) -> bool:
-    # 如果沒有任何過濾規則，接受所有網站
-    if not domain_filter:
-        return True
-
-    # 分離排除規則和包含規則
-    exclude_rules = [
-        rule[1:].replace("*.", "") for rule in domain_filter if rule.startswith("-")
-    ]
-    include_rules = [
-        rule.replace("*.", "") for rule in domain_filter if not rule.startswith("-")
-    ]
-
-    # 檢查是否符合任何排除規則
-    for pattern in exclude_rules:
-        if pattern in citation:
-            return False
-
-    # 如果沒有包含規則，且通過了排除規則檢查，就接受該網站
-    if not include_rules:
-        return True
-
-    # 如果有包含規則，必須符合至少一個
-    for pattern in include_rules:
-        if pattern in citation:
-            return True
-
-    return False
-
-
-def is_valid_domain(domain: str) -> bool:
-    if not domain or "*." in domain:
-        return False
-
-    # 只允許包含 ://、.、% 和英數字的網址
-    # ^ 表示開頭，$ 表示結尾
-    # [a-zA-Z0-9] 表示英數字
-    # [\\.\\:\\/\\%] 表示允許的特殊字符
-    pattern = r"^[a-zA-Z0-9\\.\\:\\/\\%]+$"
-
-    return bool(re.match(pattern, domain))
-
-
-async def respond_with_perplexity_search_openrouter(
-    input_content,
-    user_prompt_prefix,
-    messages_for_llm,
-    domain_filter: list[str],
-    stream: bool = False,
-    model: str = "perplexity/sonar-small-online",
-    structured_output: bool = False,
-    return_images: bool = False,
-) -> AsyncGenerator[PerplexitySearchEvent, None]:
-    """
-    使用 OpenRouter 提供的 Perplexity API 服務
-    structured_output: 只有在 stream 為 False 時有效
-    """
-    # 確保模型是 Perplexity 的模型
-    if not model.startswith("perplexity/"):
-        model = "perplexity/sonar-small-online"
-
-    api_key = os.getenv("OPENROUTER_API_KEY")
-    if not api_key:
-        raise ValueError("OPENROUTER_API_KEY environment variable not set")
-
-    headers = {
-        "Authorization": f"Bearer {api_key}",
-        "Content-Type": "application/json",
-        "HTTP-Referer": "https://openrouter.ai/api/v1",  # OpenRouter 需要提供來源
-        "X-Title": "BotRun Flow Lang",  # 可選的應用名稱
-    }
-
-    messages = deepcopy(messages_for_llm)
-    if len(messages) > 0 and messages[-1]["role"] == "user":
-        messages.pop()
-    if user_prompt_prefix:
-        xml_input_content = f"<使用者提問>{input_content}</使用者提問>"
-        messages.append(
-            {"role": "user", "content": user_prompt_prefix + "\n\n" + xml_input_content}
-        )
-    else:
-        messages.append({"role": "user", "content": input_content})
-
-    filtered_domain_filter = []
-    for domain in domain_filter:
-        if domain and is_valid_domain(domain):
-            filtered_domain_filter.append(domain)
-
-    payload = {
-        "model": model,
-        "messages": messages,
-        "temperature": 0.5,
-        "stream": stream,
-        # OpenRouter 可能不支持 search_domain_filter 參數，如果有問題可以移除
-        "search_domain_filter": filtered_domain_filter,
-        "stream_usage": True,
-        "return_images": return_images,
-        # "reasoning_effort": "high",
-    }
-
-    try:
-        input_token = 0
-        output_token = 0
-        async with aiohttp.ClientSession() as session:
-            async with session.post(
-                OPENROUTER_API_URL, headers=headers, json=payload
-            ) as response:
-                if response.status != 200:
-                    error_text = await response.text()
-                    raise ValueError(f"OpenRouter API error: {error_text}")
-
-                if not stream:
-                    # 非串流模式的處理
-                    response_data = await response.json()
-                    content = response_data["choices"][0]["message"]["content"]
-                    content = remove_citation_number_from_content(content)
-                    if not structured_output:
-                        yield PerplexitySearchEvent(chunk=content, raw_json=response_data)
-
-                    # 處理引用 (如果 OpenRouter 返回引用)
-                    citations = response_data.get("citations", [])
-                    final_citations = [
-                        citation
-                        for citation in citations
-                        if should_include_citation(citation, domain_filter)
-                    ]
-                    images = response_data.get("images", [])
-
-                    if final_citations:
-                        references = f"\n\n參考來源：\n"
-                        for citation in final_citations:
-                            references += f"- {citation}\n"
-                        if not structured_output:
-                            yield PerplexitySearchEvent(chunk=references)
-
-                    if structured_output:
-                        yield PerplexitySearchEvent(
-                            chunk=json.dumps(
-                                {
-                                    "content": content,
-                                    "citations": final_citations,
-                                    "images": images,
-                                }
-                            ),
-                            raw_json=response_data,
-                        )
-                else:
-                    # 串流模式的處理
-                    full_response = ""
-                    final_citations = []
-                    async for line in response.content:
-                        if line:
-                            line = line.decode("utf-8").strip()
-                            if line.startswith("data: "):
-                                line = line[6:]  # Remove 'data: ' prefix
-                                if line == "[DONE]":
-                                    break
-
-                                try:
-                                    chunk_data = json.loads(line)
-                                    response_data = chunk_data
-
-                                    if (
-                                        chunk_data["choices"][0]
-                                        .get("delta", {})
-                                        .get("content")
-                                    ):
-                                        content = chunk_data["choices"][0]["delta"][
-                                            "content"
-                                        ]
-                                        full_response += content
-                                        yield PerplexitySearchEvent(
-                                            chunk=content,
-                                            raw_json=chunk_data,
-                                        )
-                                    if not final_citations and chunk_data.get(
-                                        "citations", []
-                                    ):
-                                        citations = chunk_data.get("citations", [])
-                                        final_citations = [
-                                            citation
-                                            for citation in citations
-                                            if should_include_citation(
-                                                citation, domain_filter
-                                            )
-                                        ]
-
-                                except json.JSONDecodeError:
-                                    continue
-
-                    # 只在有符合條件的 citations 時才產生參考文獻
-                    if final_citations:
-                        references = f"\n\n參考來源：\n"
-                        for citation in final_citations:
-                            references += f"- {citation}\n"
-                        yield PerplexitySearchEvent(chunk=references)
-
-        if response_data.get("usage"):
-            logging.info(
-                f"perplexity_search_openrouter============> input_token: {response_data['usage'].get('prompt_tokens', 0) + response_data['usage'].get('citation_tokens', 0)}, output_token: {response_data['usage'].get('completion_tokens', 0)}",
-            )
-    except Exception as e:
-        import traceback
-
-        traceback.print_exc()
-        print(e)
-
-
-async def respond_with_perplexity_search(
-    input_content,
-    user_prompt_prefix,
-    messages_for_llm,
-    domain_filter: list[str],
-    stream: bool = False,
-    model: str = "sonar-reasoning-pro",
-    structured_output: bool = False,
-    return_images: bool = False,
-) -> AsyncGenerator[PerplexitySearchEvent, None]:
-    """
-    structured_output: 只有在 stream 為 False 時有效
-    return_images: 是否返回圖片，但是 openrouter 不支援
-    """
-    # 檢查是否使用 OpenRouter
-    is_use_openrouter = os.getenv("OPENROUTER_API_KEY") and os.getenv(
-        "OPENROUTER_BASE_URL"
-    )
-    if return_images:
-        # if os.getenv("PPLX_API_KEY", "") == "":
-        #     raise ValueError(
-        #         "PPLX_API_KEY environment variable not set, return_images needs PPLX_API_KEY"
-        #     )
-        # Openrouter 尚不支援 return_images
-        is_use_openrouter = False
-
-    if is_use_openrouter:
-        # 若使用 OpenRouter，轉換模型名稱並呼叫 OpenRouter 版本的函數
-        openrouter_model = "perplexity/sonar-reasoning-pro"
-        if model == "sonar-reasoning-pro":
-            openrouter_model = "perplexity/sonar-reasoning-pro"
-        elif model == "sonar-reasoning":
-            openrouter_model = "perplexity/sonar-reasoning"
-        elif model == "sonar-pro":
-            openrouter_model = "perplexity/sonar-pro"
-        elif model == "sonar":
-            openrouter_model = "perplexity/sonar"
-
-        async for event in respond_with_perplexity_search_openrouter(
-            input_content,
-            user_prompt_prefix,
-            messages_for_llm,
-            domain_filter,
-            stream,
-            openrouter_model,
-            structured_output,
-        ):
-            yield event
-        return
-
-    # 以下是原有的邏輯
-    if model not in ["sonar-reasoning-pro", "sonar-reasoning", "sonar-pro", "sonar"]:
-        model = "sonar-reasoning-pro"
-    api_key = os.getenv("PPLX_API_KEY")
-    if not api_key:
-        raise ValueError("PPLX_API_KEY environment variable not set")
-
-    headers = {
-        "Authorization": f"Bearer {api_key}",
-        "Content-Type": "application/json",
-    }
-    messages = deepcopy(messages_for_llm)
-    if len(messages) > 0 and messages[-1]["role"] == "user":
-        messages.pop()
-    if user_prompt_prefix:
-        xml_input_content = f"<使用者提問>{input_content}</使用者提問>"
-        messages.append(
-            {"role": "user", "content": user_prompt_prefix + "\n\n" + xml_input_content}
-        )
-    else:
-        messages.append({"role": "user", "content": input_content})
-    filtered_domain_filter = []
-
-    for domain in domain_filter:
-        if domain and is_valid_domain(domain):
-            filtered_domain_filter.append(domain)
-
-    payload = {
-        "model": model,
-        "messages": messages,
-        "temperature": 0.5,
-        "stream": stream,
-        "search_domain_filter": filtered_domain_filter,
-        "stream_usage": True,
-        "return_images": return_images,
-
-    }
-    try:
-        input_token = 0
-        output_token = 0
-        async with aiohttp.ClientSession() as session:
-            async with session.post(
-                PERPLEXITY_API_URL, headers=headers, json=payload
-            ) as response:
-                if response.status != 200:
-                    error_text = await response.text()
-                    raise ValueError(f"Perplexity API error: {error_text}")
-
-                if not stream:
-                    # 非串流模式的處理
-                    response_data = await response.json()
-                    content = response_data["choices"][0]["message"]["content"]
-                    content = remove_citation_number_from_content(content)
-                    if not structured_output:
-                        yield PerplexitySearchEvent(chunk=content, raw_json=response_data)
-
-                    # 處理引用
-                    citations = response_data.get("citations", [])
-                    final_citations = [
-                        citation
-                        for citation in citations
-                        if should_include_citation(citation, domain_filter)
-                    ]
-                    images = response_data.get("images", [])
-
-                    if final_citations:
-                        references = f"\n\n參考來源：\n"
-                        for citation in final_citations:
-                            references += f"- {citation}\n"
-                        if not structured_output:
-                            yield PerplexitySearchEvent(chunk=references)
-
-                    if structured_output:
-                        yield PerplexitySearchEvent(
-                            chunk=json.dumps(
-                                {
-                                    "content": content,
-                                    "citations": final_citations,
-                                    "images": images,
-                                }
-                            ),
-                            raw_json=response_data,
-                        )
-
-                # 串流模式的處理
-                full_response = ""
-                final_citations = []
-                async for line in response.content:
-                    if line:
-                        line = line.decode("utf-8").strip()
-                        if line.startswith("data: "):
-                            line = line[6:]  # Remove 'data: ' prefix
-                            if line == "[DONE]":
-                                break
-
-                            try:
-                                chunk_data = json.loads(line)
-                                response_data = chunk_data
-                                # print(chunk_data)
-                                if (
-                                    chunk_data["choices"][0]
-                                    .get("delta", {})
-                                    .get("content")
-                                ):
-                                    content = chunk_data["choices"][0]["delta"][
-                                        "content"
-                                    ]
-                                    full_response += content
-                                    yield PerplexitySearchEvent(
-                                        chunk=content,
-                                        raw_json=chunk_data,
-                                    )
-                                if not final_citations and chunk_data.get(
-                                    "citations", []
-                                ):
-                                    # 發現 perplexity 不會都有 finish_reason 為 stop 的狀況，但是 citations 會有
-                                    # if chunk_data["choices"][0]["finish_reason"] == "stop":
-                                    citations = chunk_data.get("citations", [])
-                                    final_citations = [
-                                        citation
-                                        for citation in citations
-                                        if should_include_citation(
-                                            citation, domain_filter
-                                        )
-                                    ]
-
-                            except json.JSONDecodeError:
-                                continue
-
-                # 只在有符合條件的 citations 時才產生參考文獻
-                if final_citations:
-                    references = f"\n\n參考來源：\n"
-                    for citation in final_citations:
-                        references += f"- {citation}\n"
-                    yield PerplexitySearchEvent(chunk=references)
-        # 安全地存取 usage 資訊，避免鍵不存在的錯誤
-        if response_data and "usage" in response_data:
-            usage = response_data["usage"]
-            prompt_tokens = usage.get("prompt_tokens", 0)
-            citation_tokens = usage.get("citation_tokens", 0)
-            completion_tokens = usage.get("completion_tokens", 0)
-            logging.info(
-                f"perplexity_search============> input_token: {prompt_tokens + citation_tokens}, output_token: {completion_tokens}",
-            )
-        else:
-            logging.info("perplexity_search============> usage information not available")
-    except Exception as e:
-        import traceback
-
-        traceback.print_exc()
-        print(e)
-
-
-def remove_citation_number_from_content(content: str) -> str:
-    """
-    移除文字裡的 [1]、[2]、[3] 等數字
-    """
-    return re.sub(r"\[[0-9]+\]", "", content)
-    # answer_message = await cl.Message(content="").send()
-    # full_response = ""
-    # for response in responses:
-    #     if response.candidates[0].finish_reason != Candidate.FinishReason.STOP:
-    #         # await answer_message.stream_token(response.text)
-    #         yield GeminiGroundingEvent(chunk=response.text)
-    #         full_response += response.text
-    #         if response.candidates[0].grounding_metadata:
-    #             if len(response.candidates[0].grounding_metadata.grounding_chunks) > 0:
-    #                 references = f"\n\n{tr('Sources:')}\n"
-    #                 for grounding_chunk in response.candidates[
-    #                     0
-    #                 ].grounding_metadata.grounding_chunks:
-    #                     references += f"- [{grounding_chunk.web.title}]({grounding_chunk.web.uri})\n"
-    #                 # await answer_message.stream_token(references)
-    #                 yield GeminiGroundingEvent(chunk=references)
-    #     else:
-    #         if response.candidates[0].grounding_metadata:
-    #             if len(response.candidates[0].grounding_metadata.grounding_chunks) > 0:
-    #                 references = f"\n\n{tr('Sources:')}\n"
-    #                 for grounding_chunk in response.candidates[
-    #                     0
-    #                 ].grounding_metadata.grounding_chunks:
-    #                     references += f"- [{grounding_chunk.web.title}]({grounding_chunk.web.uri})\n"
-    #                 # await answer_message.stream_token(references)
-    #                 yield GeminiGroundingEvent(chunk=references)
+from copy import deepcopy
+from typing import AsyncGenerator
+from pydantic import BaseModel
+import os
+import json
+import aiohttp
+from dotenv import load_dotenv
+import re
+import logging
+
+load_dotenv()
+
+
+class PerplexitySearchEvent(BaseModel):
+    chunk: str
+    raw_json: dict | None = None
+
+
+PERPLEXITY_API_URL = "https://api.perplexity.ai/chat/completions"
+OPENROUTER_API_URL = "https://openrouter.ai/api/v1/chat/completions"
+
+
+def should_include_citation(citation: str, domain_filter: list[str]) -> bool:
+    # 如果沒有任何過濾規則，接受所有網站
+    if not domain_filter:
+        return True
+
+    # 分離排除規則和包含規則
+    exclude_rules = [
+        rule[1:].replace("*.", "") for rule in domain_filter if rule.startswith("-")
+    ]
+    include_rules = [
+        rule.replace("*.", "") for rule in domain_filter if not rule.startswith("-")
+    ]
+
+    # 檢查是否符合任何排除規則
+    for pattern in exclude_rules:
+        if pattern in citation:
+            return False
+
+    # 如果沒有包含規則，且通過了排除規則檢查，就接受該網站
+    if not include_rules:
+        return True
+
+    # 如果有包含規則，必須符合至少一個
+    for pattern in include_rules:
+        if pattern in citation:
+            return True
+
+    return False
+
+
+def is_valid_domain(domain: str) -> bool:
+    if not domain or "*." in domain:
+        return False
+
+    # 只允許包含 ://、.、% 和英數字的網址
+    # ^ 表示開頭，$ 表示結尾
+    # [a-zA-Z0-9] 表示英數字
+    # [\\.\\:\\/\\%] 表示允許的特殊字符
+    pattern = r"^[a-zA-Z0-9\\.\\:\\/\\%]+$"
+
+    return bool(re.match(pattern, domain))
+
+
+async def respond_with_perplexity_search_openrouter(
+    input_content,
+    user_prompt_prefix,
+    messages_for_llm,
+    domain_filter: list[str],
+    stream: bool = False,
+    model: str = "perplexity/sonar-small-online",
+    structured_output: bool = False,
+    return_images: bool = False,
+) -> AsyncGenerator[PerplexitySearchEvent, None]:
+    """
+    使用 OpenRouter 提供的 Perplexity API 服務
+    structured_output: 只有在 stream 為 False 時有效
+    """
+    # 確保模型是 Perplexity 的模型
+    if not model.startswith("perplexity/"):
+        model = "perplexity/sonar-small-online"
+
+    api_key = os.getenv("OPENROUTER_API_KEY")
+    if not api_key:
+        raise ValueError("OPENROUTER_API_KEY environment variable not set")
+
+    headers = {
+        "Authorization": f"Bearer {api_key}",
+        "Content-Type": "application/json",
+        "HTTP-Referer": "https://openrouter.ai/api/v1",  # OpenRouter 需要提供來源
+        "X-Title": "BotRun Flow Lang",  # 可選的應用名稱
+    }
+
+    messages = deepcopy(messages_for_llm)
+    if len(messages) > 0 and messages[-1]["role"] == "user":
+        messages.pop()
+    if user_prompt_prefix:
+        xml_input_content = f"<使用者提問>{input_content}</使用者提問>"
+        messages.append(
+            {"role": "user", "content": user_prompt_prefix + "\n\n" + xml_input_content}
+        )
+    else:
+        messages.append({"role": "user", "content": input_content})
+
+    filtered_domain_filter = []
+    for domain in domain_filter:
+        if domain and is_valid_domain(domain):
+            filtered_domain_filter.append(domain)
+
+    payload = {
+        "model": model,
+        "messages": messages,
+        "temperature": 0.5,
+        "stream": stream,
+        # OpenRouter 可能不支持 search_domain_filter 參數，如果有問題可以移除
+        "search_domain_filter": filtered_domain_filter,
+        "stream_usage": True,
+        "return_images": return_images,
+        # "reasoning_effort": "high",
+    }
+
+    try:
+        input_token = 0
+        output_token = 0
+        async with aiohttp.ClientSession() as session:
+            async with session.post(
+                OPENROUTER_API_URL, headers=headers, json=payload
+            ) as response:
+                if response.status != 200:
+                    error_text = await response.text()
+                    raise ValueError(f"OpenRouter API error: {error_text}")
+
+                if not stream:
+                    # 非串流模式的處理
+                    response_data = await response.json()
+                    content = response_data["choices"][0]["message"]["content"]
+                    content = remove_citation_number_from_content(content)
+                    if not structured_output:
+                        yield PerplexitySearchEvent(chunk=content, raw_json=response_data)
+
+                    # 處理引用 (如果 OpenRouter 返回引用)
+                    citations = response_data.get("citations", [])
+                    final_citations = [
+                        citation
+                        for citation in citations
+                        if should_include_citation(citation, domain_filter)
+                    ]
+                    images = response_data.get("images", [])
+
+                    if final_citations:
+                        references = f"\n\n參考來源：\n"
+                        for citation in final_citations:
+                            references += f"- {citation}\n"
+                        if not structured_output:
+                            yield PerplexitySearchEvent(chunk=references)
+
+                    if structured_output:
+                        yield PerplexitySearchEvent(
+                            chunk=json.dumps(
+                                {
+                                    "content": content,
+                                    "citations": final_citations,
+                                    "images": images,
+                                }
+                            ),
+                            raw_json=response_data,
+                        )
+                else:
+                    # 串流模式的處理
+                    full_response = ""
+                    final_citations = []
+                    async for line in response.content:
+                        if line:
+                            line = line.decode("utf-8").strip()
+                            if line.startswith("data: "):
+                                line = line[6:]  # Remove 'data: ' prefix
+                                if line == "[DONE]":
+                                    break
+
+                                try:
+                                    chunk_data = json.loads(line)
+                                    response_data = chunk_data
+
+                                    if (
+                                        chunk_data["choices"][0]
+                                        .get("delta", {})
+                                        .get("content")
+                                    ):
+                                        content = chunk_data["choices"][0]["delta"][
+                                            "content"
+                                        ]
+                                        full_response += content
+                                        yield PerplexitySearchEvent(
+                                            chunk=content,
+                                            raw_json=chunk_data,
+                                        )
+                                    if not final_citations and chunk_data.get(
+                                        "citations", []
+                                    ):
+                                        citations = chunk_data.get("citations", [])
+                                        final_citations = [
+                                            citation
+                                            for citation in citations
+                                            if should_include_citation(
+                                                citation, domain_filter
+                                            )
+                                        ]
+
+                                except json.JSONDecodeError:
+                                    continue
+
+                    # 只在有符合條件的 citations 時才產生參考文獻
+                    if final_citations:
+                        references = f"\n\n參考來源：\n"
+                        for citation in final_citations:
+                            references += f"- {citation}\n"
+                        yield PerplexitySearchEvent(chunk=references)
+
+        if response_data.get("usage"):
+            logging.info(
+                f"perplexity_search_openrouter============> input_token: {response_data['usage'].get('prompt_tokens', 0) + response_data['usage'].get('citation_tokens', 0)}, output_token: {response_data['usage'].get('completion_tokens', 0)}",
+            )
+    except Exception as e:
+        import traceback
+
+        traceback.print_exc()
+        print(e)
+
+
+async def respond_with_perplexity_search(
+    input_content,
+    user_prompt_prefix,
+    messages_for_llm,
+    domain_filter: list[str],
+    stream: bool = False,
+    model: str = "sonar-reasoning-pro",
+    structured_output: bool = False,
+    return_images: bool = False,
+) -> AsyncGenerator[PerplexitySearchEvent, None]:
+    """
+    structured_output: 只有在 stream 為 False 時有效
+    return_images: 是否返回圖片，但是 openrouter 不支援
+    """
+    # 檢查是否使用 OpenRouter
+    is_use_openrouter = os.getenv("OPENROUTER_API_KEY") and os.getenv(
+        "OPENROUTER_BASE_URL"
+    )
+    if return_images:
+        # if os.getenv("PPLX_API_KEY", "") == "":
+        #     raise ValueError(
+        #         "PPLX_API_KEY environment variable not set, return_images needs PPLX_API_KEY"
+        #     )
+        # Openrouter 尚不支援 return_images
+        is_use_openrouter = False
+
+    if is_use_openrouter:
+        # 若使用 OpenRouter，轉換模型名稱並呼叫 OpenRouter 版本的函數
+        openrouter_model = "perplexity/sonar-reasoning-pro"
+        if model == "sonar-reasoning-pro":
+            openrouter_model = "perplexity/sonar-reasoning-pro"
+        elif model == "sonar-reasoning":
+            openrouter_model = "perplexity/sonar-reasoning"
+        elif model == "sonar-pro":
+            openrouter_model = "perplexity/sonar-pro"
+        elif model == "sonar":
+            openrouter_model = "perplexity/sonar"
+
+        async for event in respond_with_perplexity_search_openrouter(
+            input_content,
+            user_prompt_prefix,
+            messages_for_llm,
+            domain_filter,
+            stream,
+            openrouter_model,
+            structured_output,
+        ):
+            yield event
+        return
+
+    # 以下是原有的邏輯
+    if model not in ["sonar-reasoning-pro", "sonar-reasoning", "sonar-pro", "sonar"]:
+        model = "sonar-reasoning-pro"
+    api_key = os.getenv("PPLX_API_KEY")
+    if not api_key:
+        raise ValueError("PPLX_API_KEY environment variable not set")
+
+    headers = {
+        "Authorization": f"Bearer {api_key}",
+        "Content-Type": "application/json",
+    }
+    messages = deepcopy(messages_for_llm)
+    if len(messages) > 0 and messages[-1]["role"] == "user":
+        messages.pop()
+    if user_prompt_prefix:
+        xml_input_content = f"<使用者提問>{input_content}</使用者提問>"
+        messages.append(
+            {"role": "user", "content": user_prompt_prefix + "\n\n" + xml_input_content}
+        )
+    else:
+        messages.append({"role": "user", "content": input_content})
+    filtered_domain_filter = []
+
+    for domain in domain_filter:
+        if domain and is_valid_domain(domain):
+            filtered_domain_filter.append(domain)
+
+    payload = {
+        "model": model,
+        "messages": messages,
+        "temperature": 0.5,
+        "stream": stream,
+        "search_domain_filter": filtered_domain_filter,
+        "stream_usage": True,
+        "return_images": return_images,
+
+    }
+    try:
+        input_token = 0
+        output_token = 0
+        async with aiohttp.ClientSession() as session:
+            async with session.post(
+                PERPLEXITY_API_URL, headers=headers, json=payload
+            ) as response:
+                if response.status != 200:
+                    error_text = await response.text()
+                    raise ValueError(f"Perplexity API error: {error_text}")
+
+                if not stream:
+                    # 非串流模式的處理
+                    response_data = await response.json()
+                    content = response_data["choices"][0]["message"]["content"]
+                    content = remove_citation_number_from_content(content)
+                    if not structured_output:
+                        yield PerplexitySearchEvent(chunk=content, raw_json=response_data)
+
+                    # 處理引用
+                    citations = response_data.get("citations", [])
+                    final_citations = [
+                        citation
+                        for citation in citations
+                        if should_include_citation(citation, domain_filter)
+                    ]
+                    images = response_data.get("images", [])
+
+                    if final_citations:
+                        references = f"\n\n參考來源：\n"
+                        for citation in final_citations:
+                            references += f"- {citation}\n"
+                        if not structured_output:
+                            yield PerplexitySearchEvent(chunk=references)
+
+                    if structured_output:
+                        yield PerplexitySearchEvent(
+                            chunk=json.dumps(
+                                {
+                                    "content": content,
+                                    "citations": final_citations,
+                                    "images": images,
+                                }
+                            ),
+                            raw_json=response_data,
+                        )
+
+                # 串流模式的處理
+                full_response = ""
+                final_citations = []
+                async for line in response.content:
+                    if line:
+                        line = line.decode("utf-8").strip()
+                        if line.startswith("data: "):
+                            line = line[6:]  # Remove 'data: ' prefix
+                            if line == "[DONE]":
+                                break
+
+                            try:
+                                chunk_data = json.loads(line)
+                                response_data = chunk_data
+                                # print(chunk_data)
+                                if (
+                                    chunk_data["choices"][0]
+                                    .get("delta", {})
+                                    .get("content")
+                                ):
+                                    content = chunk_data["choices"][0]["delta"][
+                                        "content"
+                                    ]
+                                    full_response += content
+                                    yield PerplexitySearchEvent(
+                                        chunk=content,
+                                        raw_json=chunk_data,
+                                    )
+                                if not final_citations and chunk_data.get(
+                                    "citations", []
+                                ):
+                                    # 發現 perplexity 不會都有 finish_reason 為 stop 的狀況，但是 citations 會有
+                                    # if chunk_data["choices"][0]["finish_reason"] == "stop":
+                                    citations = chunk_data.get("citations", [])
+                                    final_citations = [
+                                        citation
+                                        for citation in citations
+                                        if should_include_citation(
+                                            citation, domain_filter
+                                        )
+                                    ]
+
+                            except json.JSONDecodeError:
+                                continue
+
+                # 只在有符合條件的 citations 時才產生參考文獻
+                if final_citations:
+                    references = f"\n\n參考來源：\n"
+                    for citation in final_citations:
+                        references += f"- {citation}\n"
+                    yield PerplexitySearchEvent(chunk=references)
+        # 安全地存取 usage 資訊，避免鍵不存在的錯誤
+        if response_data and "usage" in response_data:
+            usage = response_data["usage"]
+            prompt_tokens = usage.get("prompt_tokens", 0)
+            citation_tokens = usage.get("citation_tokens", 0)
+            completion_tokens = usage.get("completion_tokens", 0)
+            logging.info(
+                f"perplexity_search============> input_token: {prompt_tokens + citation_tokens}, output_token: {completion_tokens}",
+            )
+        else:
+            logging.info("perplexity_search============> usage information not available")
+    except Exception as e:
+        import traceback
+
+        traceback.print_exc()
+        print(e)
+
+
+def remove_citation_number_from_content(content: str) -> str:
+    """
+    移除文字裡的 [1]、[2]、[3] 等數字
+    """
+    return re.sub(r"\[[0-9]+\]", "", content)
+    # answer_message = await cl.Message(content="").send()
+    # full_response = ""
+    # for response in responses:
+    #     if response.candidates[0].finish_reason != Candidate.FinishReason.STOP:
+    #         # await answer_message.stream_token(response.text)
+    #         yield GeminiGroundingEvent(chunk=response.text)
+    #         full_response += response.text
+    #         if response.candidates[0].grounding_metadata:
+    #             if len(response.candidates[0].grounding_metadata.grounding_chunks) > 0:
+    #                 references = f"\n\n{tr('Sources:')}\n"
+    #                 for grounding_chunk in response.candidates[
+    #                     0
+    #                 ].grounding_metadata.grounding_chunks:
+    #                     references += f"- [{grounding_chunk.web.title}]({grounding_chunk.web.uri})\n"
+    #                 # await answer_message.stream_token(references)
+    #                 yield GeminiGroundingEvent(chunk=references)
+    #     else:
+    #         if response.candidates[0].grounding_metadata:
+    #             if len(response.candidates[0].grounding_metadata.grounding_chunks) > 0:
+    #                 references = f"\n\n{tr('Sources:')}\n"
+    #                 for grounding_chunk in response.candidates[
+    #                     0
+    #                 ].grounding_metadata.grounding_chunks:
+    #                     references += f"- [{grounding_chunk.web.title}]({grounding_chunk.web.uri})\n"
+    #                 # await answer_message.stream_token(references)
+    #                 yield GeminiGroundingEvent(chunk=references)