jarvis-ai-assistant 0.1.175__tar.gz → 0.1.176__tar.gz

{jarvis_ai_assistant-0.1.175/src/jarvis_ai_assistant.egg-info → jarvis_ai_assistant-0.1.176}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: jarvis-ai-assistant
-Version: 0.1.175
+Version: 0.1.176
 Summary: Jarvis: An AI assistant that uses tools to interact with the system
 Home-page: https://github.com/skyfireitdiy/Jarvis
 Author: skyfire
@@ -213,7 +213,8 @@ OPENAI_API_BASE=https://api.openai.com/v1  # 可选，默认为官方API地址
 | `JARVIS_CONFIRM_BEFORE_APPLY_PATCH` | true | 应用补丁前是否需要确认 |
 | `JARVIS_MAX_TOOL_CALL_COUNT` | 20 | 最大连续工具调用次数 |
 | `JARVIS_AUTO_UPDATE` | true | 是否自动更新Jarvis（仅在以git仓库方式安装时有效） |
-| `JARVIS_MAX_BIG_CONTENT_SIZE` | 10485760 | 最大大内容大小 |
+| `JARVIS_MAX_BIG_CONTENT_SIZE` | 96000 | 最大大内容大小 |
+| `JARVIS_PRETTY_OUTPUT` | false | 是否启用PrettyOutput |
 
 所有配置编写到`~/.jarvis/env`文件中即可生效。
 

{jarvis_ai_assistant-0.1.175 → jarvis_ai_assistant-0.1.176}/README.md

@@ -143,7 +143,8 @@ OPENAI_API_BASE=https://api.openai.com/v1  # 可选，默认为官方API地址
 | `JARVIS_CONFIRM_BEFORE_APPLY_PATCH` | true | 应用补丁前是否需要确认 |
 | `JARVIS_MAX_TOOL_CALL_COUNT` | 20 | 最大连续工具调用次数 |
 | `JARVIS_AUTO_UPDATE` | true | 是否自动更新Jarvis（仅在以git仓库方式安装时有效） |
-| `JARVIS_MAX_BIG_CONTENT_SIZE` | 10485760 | 最大大内容大小 |
+| `JARVIS_MAX_BIG_CONTENT_SIZE` | 96000 | 最大大内容大小 |
+| `JARVIS_PRETTY_OUTPUT` | false | 是否启用PrettyOutput |
 
 所有配置编写到`~/.jarvis/env`文件中即可生效。
 

{jarvis_ai_assistant-0.1.175 → jarvis_ai_assistant-0.1.176}/pyproject.toml

@@ -8,7 +8,7 @@ default = true
 
 [project]
 name = "jarvis-ai-assistant"
-version = "0.1.175"
+version = "0.1.176"
 description = "Jarvis: An AI assistant that uses tools to interact with the system"
 readme = "README.md"
 authors = [{ name = "skyfire", email = "skyfireitdiy@hotmail.com" }]

{jarvis_ai_assistant-0.1.175 → jarvis_ai_assistant-0.1.176}/setup.py

@@ -3,7 +3,7 @@ from setuptools import setup, find_packages
 
 setup(
     name="jarvis-ai-assistant",
-    version="0.1.175",
+    version="0.1.176",
     author="skyfire",
     author_email="skyfireitdiy@hotmail.com",
     description="An AI assistant that uses various tools to interact with the system",

{jarvis_ai_assistant-0.1.175 → jarvis_ai_assistant-0.1.176}/src/jarvis/__init__.py

@@ -1,4 +1,4 @@
 # -*- coding: utf-8 -*-
 """Jarvis AI Assistant"""
 
-__version__ = "0.1.175"
+__version__ = "0.1.176"

{jarvis_ai_assistant-0.1.175 → jarvis_ai_assistant-0.1.176}/src/jarvis/jarvis_agent/__init__.py

@@ -337,33 +337,20 @@ class Agent:
         action_handlers = '\n'.join([f'- {handler.name()}' for handler in self.output_handler])
 
         # 任务完成提示
-        complete_prompt = f"3. 输出{ot('!!!COMPLETE!!!')}" if need_complete and self.auto_complete else ""
+        complete_prompt = f"- 输出{ot('!!!COMPLETE!!!')}" if need_complete and self.auto_complete else ""
 
         addon_prompt = f"""
-<addon>
-<instructions>
-**系统指令：**
-- 每次响应必须且只能包含一个操作
-- 严格遵循操作调用格式
-- 必须包含参数和说明
-- 操作结束需等待结果
-- 如果判断任务已经完成，不必输出操作
+请判断是否已经完成任务，如果已经完成：
+- 说明完成原因，不需要再有新的操作
+{complete_prompt}
+如果没有完成，请进行下一步操作：
+- 仅包含一个操作
 - 如果信息不明确，请请求用户补充
 - 如果执行过程中连续失败5次，请使用ask_user询问用户操作
-</instructions>
-
-<actions>
-**可用操作列表：**
+- 操作列表：
 {action_handlers}
-</actions>
 
-<completion>
-如果任务已完成，请：
-1. 说明完成原因
-2. 保持输出格式规范
-{complete_prompt}
-</completion>
-</addon>
+请继续。
 """
 
         return addon_prompt
@@ -401,8 +388,6 @@ class Agent:
         if self.conversation_length > self.max_token_count:
             message = self._summarize_and_clear_history() + "\n\n" + message
             self.conversation_length += get_context_token_count(message)
-
-        print("🤖 模型思考：")
         return self.model.chat_until_success(message)   # type: ignore
 
 
@@ -547,7 +532,7 @@ class Agent:
 1. 首先检查现有工具或方法论是否已经可以完成该任务，如果可以，直接说明即可，无需生成新内容
 2. 如果现有工具/方法论不足，评估当前任务是否可以通过编写新工具来自动化解决
 3. 如果可以通过工具解决，请设计并提供工具代码
-4. 如果无法通过编写通用工具完成，评估是否可以总结为通用方法论
+4. 如果无法通过编写通用工具完成，评估当前的执行流程是否可以总结为通用方法论
 5. 如果以上都不可行，给出详细理由
 
 请根据分析结果采取相应行动：说明现有工具/方法论、创建新工具、生成新方法论或说明原因。
@@ -570,6 +555,7 @@ class Agent:
 2. 方法论应该具备足够的通用性，可应用于同类问题
 3. 特别注意用户在执行过程中提供的修正、反馈和改进建议
 4. 如果用户明确指出了某个解决步骤的优化方向，这应该被纳入方法论
+5. 方法论要严格按照实际的执行流程来总结，不要遗漏或增加任何步骤
 </evaluation_criteria>
 
 <tool_requirements>
@@ -751,19 +737,18 @@ arguments:
             self.prompt = f"{user_input}"
 
             if self.first:
-                msg = user_input
-                for handler in self.input_handler:
-                    msg, _ = handler(msg, self)
                 
                 # 先尝试上传方法轮
                 platform = self.model if hasattr(self.model, 'upload_files') else None
                 if platform and upload_methodology(platform):
-                    methodology_prompt = f"{user_input}\n\n方法论已上传到平台，请参考平台上的方法论内容"
+                    self.prompt = f"{user_input}\n\n方法论已上传到平台，请参考平台上的方法论内容"
                 else:
+                    msg = user_input
+                    for handler in self.input_handler:
+                        msg, _ = handler(msg, self)
                     # 上传失败则回退到本地加载
-                    methodology_prompt = f"{user_input}\n\n以下是历史类似问题的执行经验，可参考：\n{load_methodology(msg, self.get_tool_registry())}"
+                    self.prompt = f"{user_input}\n\n以下是历史类似问题的执行经验，可参考：\n{load_methodology(msg, self.get_tool_registry())}"
                 
-                self.prompt = methodology_prompt
                 self.first = False
 
             self.conversation_length = get_context_token_count(self.prompt)

{jarvis_ai_assistant-0.1.175 → jarvis_ai_assistant-0.1.176}/src/jarvis/jarvis_agent/jarvis.py

@@ -18,7 +18,6 @@ from jarvis.jarvis_agent import (
 )
 from jarvis.jarvis_tools.registry import ToolRegistry
 from jarvis.jarvis_utils.utils import init_env
-from jarvis.jarvis_agent.file_input_handler import file_input_handler
 from jarvis.jarvis_agent.shell_input_handler import shell_input_handler
 from jarvis.jarvis_agent.builtin_input_handler import builtin_input_handler
 
@@ -122,7 +121,10 @@ def main() -> None:
             system_prompt=origin_agent_system_prompt,
             platform=args.platform,
             model_name=args.model,
-            input_handler=[file_input_handler, shell_input_handler, builtin_input_handler],
+            input_handler=[
+                shell_input_handler, 
+                builtin_input_handler
+            ],
             output_handler=[ToolRegistry()],
             need_summary=False
         )

{jarvis_ai_assistant-0.1.175 → jarvis_ai_assistant-0.1.176}/src/jarvis/jarvis_code_agent/code_agent.py

@@ -16,7 +16,6 @@ from yaspin import yaspin  # type: ignore
 
 from jarvis.jarvis_agent import Agent
 from jarvis.jarvis_agent.builtin_input_handler import builtin_input_handler
-from jarvis.jarvis_agent.file_input_handler import file_input_handler
 from jarvis.jarvis_agent.shell_input_handler import shell_input_handler
 from jarvis.jarvis_platform.registry import PlatformRegistry
 from jarvis.jarvis_git_utils.git_commiter import GitCommitTool
@@ -109,7 +108,6 @@ class CodeAgent:
             platform=platform_instance,
             input_handler=[
                 shell_input_handler,
-                file_input_handler,
                 builtin_input_handler
             ],
             need_summary=need_summary
@@ -374,8 +372,8 @@ class CodeAgent:
                     final_ret += f"# 应用补丁:\n```diff\n{diff}\n```"
 
                     # 修改后的提示逻辑
-                    addon_prompt = "如果对应语言有静态检查工具，请使用静态检查工具检查代码，如果本次修改引入了警告和错误，请根据警告和错误信息修复代码\n"
-                    addon_prompt += "在引入警告和错误都被修复的前提下，如果用户的需求未完成，请继续生成补丁，如果已经完成，请终止，不要实现任何超出用户需求外的内容\n"
+                    addon_prompt = "如果对应语言有静态检查工具，请使用静态检查工具检查修改的代码，如果本次修改引入了警告和错误，请根据警告和错误信息修复代码\n"
+                    addon_prompt += "在引入警告和错误都被修复的前提下，如果用户的需求未完成，请继续修改代码，如果已经完成，请终止，不要实现任何超出用户需求外的内容\n"
                     addon_prompt += "如果有任何信息不明确，调用工具获取信息\n"
                     addon_prompt += "每次响应必须且只能包含一个操作\n"
 

{jarvis_ai_assistant-0.1.175 → jarvis_ai_assistant-0.1.176}/src/jarvis/jarvis_platform/base.py

@@ -1,15 +1,19 @@
 # -*- coding: utf-8 -*-
 from abc import ABC, abstractmethod
 import re
-from typing import List, Tuple
+from typing import Generator, List, Tuple
 
 from yaspin import yaspin
-from jarvis.jarvis_utils.config import get_max_input_token_count
+
+from jarvis.jarvis_utils.config import get_max_input_token_count, get_pretty_output
 from jarvis.jarvis_utils.embedding import split_text_into_chunks
 from jarvis.jarvis_utils.output import OutputType, PrettyOutput
 from jarvis.jarvis_utils.utils import get_context_token_count, is_context_overflow, while_success, while_true
 from jarvis.jarvis_utils.tag import ot, ct
-
+from rich.live import Live
+from rich.text import Text
+from rich.panel import Panel
+from rich import box
 
 class BasePlatform(ABC):
     """Base class for large language models"""
@@ -33,7 +37,7 @@ class BasePlatform(ABC):
         self.delete_chat()
 
     @abstractmethod
-    def chat(self, message: str) -> str:
+    def chat(self, message: str) -> Generator[str, None, None]:
         """Execute conversation"""
         raise NotImplementedError("chat is not implemented")
     
@@ -53,8 +57,6 @@ class BasePlatform(ABC):
             return "错误：输入内容超过最大限制"
 
         if input_token_count > get_max_input_token_count():
-            current_suppress_output = self.suppress_output
-            self.set_suppress_output(True)
             max_chunk_size = get_max_input_token_count() - 1024  # 留出一些余量
             min_chunk_size = max_chunk_size // 2  # 最小块大小设为最大块大小的一半
             inputs = split_text_into_chunks(message, max_chunk_size, min_chunk_size)
@@ -70,32 +72,48 @@ class BasePlatform(ABC):
                     while_true(lambda: while_success(lambda: self.chat(f"<part_content>{input}</part_content>请返回已收到"), 5), 5)
                 spinner.text = "提交完成"
                 spinner.ok("✅")
-            self.set_suppress_output(current_suppress_output)
-            response = while_true(lambda: while_success(lambda: self.chat("内容已经全部提供完毕，请继续"), 5), 5)
-            
+            response = while_true(lambda: while_success(lambda: self._chat("内容已经全部提供完毕，请继续"), 5), 5)
         else:
-            response = self.chat(message)
-
-        end_time = time.time()
-        duration = end_time - start_time
-        char_count = len(response)
-
-        # Calculate token count and tokens per second
-        try:
-            token_count = get_context_token_count(response)
-            tokens_per_second = token_count / duration if duration > 0 else 0
-        except Exception as e:
-            PrettyOutput.print(f"Tokenization failed: {str(e)}", OutputType.WARNING)
-            token_count = 0
-            tokens_per_second = 0
-
-        # Print statistics
-        if not self.suppress_output:
-            PrettyOutput.print(
-                f"对话完成 - 耗时: {duration:.2f}秒, 输入字符数: {len(message)}, 输入Token数量: {input_token_count}, 输出字符数: {char_count}, 输出Token数量: {token_count}, 每秒Token数量: {tokens_per_second:.2f}",
-                OutputType.INFO,
+            response = ""
+
+            text_content = Text()
+            panel = Panel(
+                text_content, 
+                title=f"[bold cyan]{self.name()}[/bold cyan]", 
+                subtitle="[dim]思考中...[/dim]", 
+                border_style="bright_blue",
+                box=box.ROUNDED
             )
 
+            if not self.suppress_output:
+                if get_pretty_output():
+                    with Live(panel, refresh_per_second=10, transient=False) as live:
+                        for s in self.chat(message):
+                            response += s
+                            text_content.append(s, style="bright_white")
+                            panel.subtitle = "[yellow]正在回答...[/yellow]"
+                            live.update(panel)
+                        end_time = time.time()
+                        duration = end_time - start_time
+                        char_count = len(response)
+                        # Calculate token count and tokens per second
+                        try:
+                            token_count = get_context_token_count(response)
+                            tokens_per_second = token_count / duration if duration > 0 else 0
+                        except Exception as e:
+                            PrettyOutput.print(f"Tokenization failed: {str(e)}", OutputType.WARNING)
+                            token_count = 0
+                            tokens_per_second = 0
+                        panel.subtitle = f"[bold green]✓ 对话完成耗时: {duration:.2f}秒, 输入字符数: {len(message)}, 输入Token数量: {input_token_count}, 输出字符数: {char_count}, 输出Token数量: {token_count}, 每秒Token数量: {tokens_per_second:.2f}[/bold green]"
+                        live.update(panel)
+                else:
+                    for s in self.chat(message):
+                        print(s, end="", flush=True)
+                        response += s
+                    print()
+            else:
+                for s in self.chat(message):
+                    response += s
         # Keep original think tag handling
         response = re.sub(ot("think")+r'.*?'+ct("think"), '', response, flags=re.DOTALL)
         return response

{jarvis_ai_assistant-0.1.175 → jarvis_ai_assistant-0.1.176}/src/jarvis/jarvis_platform/human.py

@@ -1,5 +1,5 @@
 # -*- coding: utf-8 -*-
-from typing import Dict, List, Tuple
+from typing import Generator, List, Tuple
 import random
 import string
 from jarvis.jarvis_platform.base import BasePlatform
@@ -34,7 +34,7 @@ class HumanPlatform(BasePlatform):
         else:
             PrettyOutput.print(f"错误：不支持的模型: {model_name}", OutputType.ERROR)
 
-    def chat(self, message: str) -> str:
+    def chat(self, message: str) -> Generator[str, None, None]:
         """发送消息并获取人类响应"""
         if not self.conversation_id:
             self.conversation_id = ''.join(random.choices(string.ascii_letters + string.digits, k=8))
@@ -49,7 +49,8 @@ class HumanPlatform(BasePlatform):
             prompt = f"{message} {session_info}\n\n请回复:"
         
         response = get_multiline_input(prompt)
-        return response
+        yield response
+        return None
 
     def upload_files(self, file_list: List[str]) -> bool:
         """文件上传功能，人类平台不需要实际处理"""

{jarvis_ai_assistant-0.1.175 → jarvis_ai_assistant-0.1.176}/src/jarvis/jarvis_platform/kimi.py

@@ -1,5 +1,5 @@
 # -*- coding: utf-8 -*-
-from typing import Dict, List, Tuple
+from typing import Dict, Generator, List, Tuple
 import requests # type: ignore
 import json
 import os
@@ -241,7 +241,7 @@ class KimiModel(BasePlatform):
         return True
 
 
-    def chat(self, message: str) -> str:
+    def chat(self, message: str) -> Generator[str, None, None]:
         """Send message and get response"""
         if not self.chat_id:
             if not self._create_chat():
@@ -279,177 +279,29 @@ class KimiModel(BasePlatform):
 
         try:
             response = while_success(lambda: requests.post(url, headers=headers, json=payload, stream=True), sleep_time=5)
-            full_response = ""
-
-            # 收集搜索和引用结果
-            search_results = []
-            ref_sources = []
-
-            # 使用Rich的Live组件来实时展示更新
-            if not self.suppress_output:
-                text_content = Text()
-                panel = Panel(text_content, 
-                            title=f"[bold magenta]{self.model_name}[/bold magenta]", 
-                            subtitle="思考中...", 
-                            border_style="magenta", 
-                            box=box.ROUNDED)
-                
-                with Live(panel, refresh_per_second=3, transient=False) as live:
-                    for line in response.iter_lines():
-                        if not line:
-                            continue
-
-                        line = line.decode('utf-8')
-                        if not line.startswith("data: "):
-                            continue
-
-                        try:
-                            data = json.loads(line[6:])
-                            event = data.get("event")
-
-                            if event == "cmpl":
-                                # 处理补全文本
-                                text = data.get("text", "")
-                                if text:
-                                    full_response += text
-                                    text_content.append(text)
-                                    panel.subtitle = "生成中..."
-                                    live.update(panel)
-
-                            elif event == "search_plus":
-                                # 收集搜索结果
-                                msg = data.get("msg", {})
-                                if msg.get("type") == "get_res":
-                                    search_results.append({
-                                        "date": msg.get("date", ""),
-                                        "site_name": msg.get("site_name", ""),
-                                        "snippet": msg.get("snippet", ""),
-                                        "title": msg.get("title", ""),
-                                        "type": msg.get("type", ""),
-                                        "url": msg.get("url", "")
-                                    })
-                                    panel.subtitle = f"搜索中: 找到 {len(search_results)} 个结果"
-                                    live.update(panel)
-
-                            elif event == "ref_docs":
-                                # 收集引用来源
-                                ref_cards = data.get("ref_cards", [])
-                                for card in ref_cards:
-                                    ref_sources.append({
-                                        "idx_s": card.get("idx_s", ""),
-                                        "idx_z": card.get("idx_z", ""),
-                                        "ref_id": card.get("ref_id", ""),
-                                        "url": card.get("url", ""),
-                                        "title": card.get("title", ""),
-                                        "abstract": card.get("abstract", ""),
-                                        "source": card.get("source_label", ""),
-                                        "rag_segments": card.get("rag_segments", []),
-                                        "origin": card.get("origin", {})
-                                    })
-                                    panel.subtitle = f"分析引用: 找到 {len(ref_sources)} 个来源"
-                                    live.update(panel)
-
-                        except json.JSONDecodeError:
-                            continue
-                    
-                    # 显示对话完成状态
-                    panel.subtitle = "[bold green]回答完成[/bold green]"
-                    live.update(panel)
-            else:
-                # 如果禁止输出，则静默处理
-                for line in response.iter_lines():
-                    if not line:
-                        continue
-
-                    line = line.decode('utf-8')
-                    if not line.startswith("data: "):
-                        continue
-
-                    try:
-                        data = json.loads(line[6:])
-                        event = data.get("event")
-
-                        if event == "cmpl":
-                            # 处理补全文本
-                            text = data.get("text", "")
-                            if text:
-                                full_response += text
-
-                        elif event == "search_plus":
-                            # 收集搜索结果
-                            msg = data.get("msg", {})
-                            if msg.get("type") == "get_res":
-                                search_results.append({
-                                    "date": msg.get("date", ""),
-                                    "site_name": msg.get("site_name", ""),
-                                    "snippet": msg.get("snippet", ""),
-                                    "title": msg.get("title", ""),
-                                    "type": msg.get("type", ""),
-                                    "url": msg.get("url", "")
-                                })
-
-                        elif event == "ref_docs":
-                            # 收集引用来源
-                            ref_cards = data.get("ref_cards", [])
-                            for card in ref_cards:
-                                ref_sources.append({
-                                    "idx_s": card.get("idx_s", ""),
-                                    "idx_z": card.get("idx_z", ""),
-                                    "ref_id": card.get("ref_id", ""),
-                                    "url": card.get("url", ""),
-                                    "title": card.get("title", ""),
-                                    "abstract": card.get("abstract", ""),
-                                    "source": card.get("source_label", ""),
-                                    "rag_segments": card.get("rag_segments", []),
-                                    "origin": card.get("origin", {})
-                                })
-
-                    except json.JSONDecodeError:
-                        continue
-
-            # 显示搜索结果摘要
-            if search_results and not self.suppress_output:
-                output = ["搜索结果:"]
-                for result in search_results:
-                    output.append(f"- {result['title']}")
-                    if result['date']:
-                        output.append(f"  日期: {result['date']}")
-                    output.append(f"  来源: {result['site_name']}")
-                    if result['snippet']:
-                        output.append(f"  摘要: {result['snippet']}")
-                    output.append(f"  链接: {result['url']}")
-                    output.append("")
-                PrettyOutput.print("\n".join(output), OutputType.PROGRESS)
-
-            # 显示引用来源
-            if ref_sources and not self.suppress_output:
-                output = ["引用来源:"]
-                for source in ref_sources:
-                    output.append(f"- [{source['ref_id']}] {source['title']} ({source['source']})")
-                    output.append(f"  链接: {source['url']}")
-                    if source['abstract']:
-                        output.append(f"  摘要: {source['abstract']}")
-
-                    # 显示相关段落
-                    if source['rag_segments']:
-                        output.append("  相关段落:")
-                        for segment in source['rag_segments']:
-                            text = segment.get('text', '').replace('\n', ' ').strip()
-                            if text:
-                                output.append(f"    - {text}")
-
-                    # 显示原文引用
-                    origin = source['origin']
-                    if origin:
-                        text = origin.get('text', '')
-                        if text:
-                            output.append(f"  原文: {text}")
+            # 如果禁止输出，则静默处理
+            for line in response.iter_lines():
+                if not line:
+                    continue
+
+                line = line.decode('utf-8')
+                if not line.startswith("data: "):
+                    continue
+
+                try:
+                    data = json.loads(line[6:])
+                    event = data.get("event")
 
-                    output.append("")
+                    if event == "cmpl":
+                        # 处理补全文本
+                        text = data.get("text", "")
+                        if text:
+                            yield text
+                except json.JSONDecodeError:
+                    continue
 
-                PrettyOutput.print("\n".join(output), OutputType.PROGRESS)
 
-            return full_response
+            return None
 
         except Exception as e:
             raise Exception(f"Chat failed: {str(e)}")

{jarvis_ai_assistant-0.1.175 → jarvis_ai_assistant-0.1.176}/src/jarvis/jarvis_platform/openai.py

@@ -1,5 +1,5 @@
 # -*- coding: utf-8 -*-
-from typing import Dict, List, Tuple
+from typing import Dict, Generator, List, Tuple
 import os
 from openai import OpenAI
 from rich.live import Live
@@ -73,7 +73,7 @@ class OpenAIModel(BasePlatform):
         self.system_message = message
         self.messages.append({"role": "system", "content": self.system_message})
 
-    def chat(self, message: str) -> str:
+    def chat(self, message: str) -> Generator[str, None, None]:
         """Execute conversation"""
         try:
 
@@ -87,38 +87,16 @@ class OpenAIModel(BasePlatform):
             ) # type: ignore
 
             full_response = ""
-
-            # 使用Rich的Live组件来实时展示更新
-            if not self.suppress_output:
-                text_content = Text()
-                panel = Panel(text_content, 
-                              title=f"[bold blue]{self.model_name}[/bold blue]", 
-                              subtitle="生成中...", 
-                              border_style="cyan", 
-                              box=box.ROUNDED)
-                
-                with Live(panel, refresh_per_second=3, transient=False) as live:
-                    for chunk in response:
-                        if chunk.choices and chunk.choices[0].delta.content:
-                            text = chunk.choices[0].delta.content
-                            full_response += text
-                            text_content.append(text)
-                            live.update(panel)
-                    
-                    # 显示对话完成状态
-                    panel.subtitle = "[bold green]对话完成[/bold green]"
-                    live.update(panel)
-            else:
-                # 如果禁止输出，则静默处理
-                for chunk in response:
-                    if chunk.choices and chunk.choices[0].delta.content:
-                        text = chunk.choices[0].delta.content
-                        full_response += text
+            for chunk in response:
+                if chunk.choices and chunk.choices[0].delta.content:
+                    text = chunk.choices[0].delta.content
+                    full_response += text
+                    yield text
 
             # Add assistant reply to history
             self.messages.append({"role": "assistant", "content": full_response})
 
-            return full_response
+            return None
 
         except Exception as e:
             PrettyOutput.print(f"对话失败：{str(e)}", OutputType.ERROR)