add files

README.md

@@ -1,11 +1,60 @@
 ---
-title: AgentScope
-emoji: 🦀
-colorFrom: purple
-colorTo: green
-sdk: docker
-pinned: false
-license: apache-2.0
+-title: AgentScope
+-emoji: 🦀
+-colorFrom: purple
+-colorTo: green
+-sdk: docker
+-pinned: false
+-license: apache-2.0
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+<h1> Modelscope AgentFabric: Customizable AI-Agents For All</h1>
+
+<p align="center">
+    <br>
+    <img src="https://modelscope.oss-cn-beijing.aliyuncs.com/modelscope.gif" width="400"/>
+    <br>
+<p>
+
+## Introduction
+**ModelScope AgentFabric** is an interactive framework to facilitate creation of agents tailored to various real-world applications. AgentFabric is built around pluggable and customizable LLMs, and enhance capabilities of  instrcution following, extra knowledge retrieval and leveraging external tools. The AgentFabric is woven with interfaces including:
+- ⚡ **Agent Builder**: an automatic instructions and tools provider for customizing user's agents through natural conversational interactions.
+- ⚡ **User Agent**: a customized agent for building real-world applications, with instructions, extra-knowledge and tools provided by builder agent and/or user inputs.
+- ⚡ **Configuration Tooling**: the interface to customize user agent configurations. Allows real-time preview of agent behavior as new confiugrations are updated.
+
+🔗 We currently leverage AgentFabric to build various agents around [Qwen2.0 LLM API](https://help.aliyun.com/zh/dashscope/developer-reference/api-details) available via DashScope. We are also actively exploring
+other options to incorporate (and compare) more LLMs via API, as well as via native ModelScope models.
+
+
+## Installation
+Simply clone the repo and install dependency.
+```bash
+git clone https://github.com/modelscope/modelscope-agent.git
+cd modelscope-agent  && pip install -r requirements.txt && pip install -r demo/agentfabric/requirements.txt
+```
+
+## Prerequisites
+
+- Python 3.10
+- Accessibility to LLM API service such as [DashScope](https://help.aliyun.com/zh/dashscope/developer-reference/activate-dashscope-and-create-an-api-key) (free to start).
+
+## Usage
+
+```bash
+export PYTHONPATH=$PYTHONPATH:/path/to/your/modelscope-agent
+export DASHSCOPE_API_KEY=your_api_key
+cd modelscope-agent/demo/agentfabric
+python app.py
+```
+
+## 🚀 Roadmap
+- [x] Allow customizable agent-building via configurations.
+- [x] Agent-building through interactive conversations with LLMs.
+- [x] Support multi-user preview on ModelScope space. [link](https://modelscope.cn/studios/wenmengzhou/AgentFabric/summary) [PR #98](https://github.com/modelscope/modelscope-agent/pull/98)
+- [x] Optimize knowledge retrival. [PR #105](https://github.com/modelscope/modelscope-agent/pull/105) [PR #107](https://github.com/modelscope/modelscope-agent/pull/107) [PR #109](https://github.com/modelscope/modelscope-agent/pull/109)
+- [x] Allow publication and sharing of agent. [PR #111](https://github.com/modelscope/modelscope-agent/pull/111)
+- [ ] Support more pluggable LLMs via API or ModelScope interface.
+- [ ] Improve long context via memory.
+- [ ] Improve logging and profiling.
+- [ ] Fine-tuning for specific agent.
+- [ ] Evaluation for agents in different scenarios.

README_CN.md

@@ -0,0 +1,52 @@
+
+<h1> Modelscope AgentFabric: 开放可定制的AI智能体构建框架</h1>
+
+<p align="center">
+    <br>
+    <img src="https://modelscope.oss-cn-beijing.aliyuncs.com/modelscope.gif" width="400"/>
+    <br>
+<p>
+
+## 介绍
+
+**Modelscope AgentFabric**是一个交互式智能体框架，用于方便地创建针对各种现实应用量身定制智能体。AgentFabric围绕可插拔和可定制的LLM构建，并增强了指令执行、额外知识检索和利用外部工具的能力。AgentFabric提供的交互界面包括：
+- **⚡ 智能体构建器**：一个自动指令和工具提供者，通过与用户聊天来定制用户的智能体
+- **⚡ 用户智能体**：一个为用户的实际应用定制的智能体，提供构建智能体或用户输入的指令、额外知识和工具
+- **⚡ 配置设置工具**：支持用户定制用户智能体的配置，并实时预览用户智能体的性能
+
+🔗 我们目前围绕DashScope提供的 [Qwen2.0 LLM API](https://help.aliyun.com/zh/dashscope/developer-reference/api-details) 来在AgentFabric上构建不同的智能体应用。同时我们正在积极探索，通过API或者ModelScope原生模型等方式，引入不同的举办强大基础能力的LLMs，来构建丰富多样的Agents。
+
+## 安装
+
+克隆仓库并安装依赖：
+
+```bash
+git clone https://github.com/modelscope/modelscope-agent.git
+cd modelscope-agent  && pip install -r requirements.txt && pip install -r demo/agentfabric/requirements.txt
+```
+
+## 前提条件
+
+- Python 3.10
+- 获取使用Qwen 2.0模型所需的API-key，可从[DashScope](https://help.aliyun.com/zh/dashscope/developer-reference/activate-dashscope-and-create-an-api-key)免费开通和获取。
+
+## 使用方法
+
+```bash
+export PYTHONPATH=$PYTHONPATH:/path/to/your/modelscope-agent
+export DASHSCOPE_API_KEY=your_api_key
+cd modelscope-agent/demo/agentfabric
+python app.py
+```
+
+## 🚀 发展路线规划
+- [x] 支持人工配置构建智能体
+- [x] 基于LLM对话构建智能体
+- [x] 支持在ModelScope创空间上使用 [link](https://modelscope.cn/studios/wenmengzhou/AgentFabric/summary) [PR #98](https://github.com/modelscope/modelscope-agent/pull/98)
+- [x] 知识库检索效果优化 [PR #105](https://github.com/modelscope/modelscope-agent/pull/105) [PR #107](https://github.com/modelscope/modelscope-agent/pull/107) [PR #109](https://github.com/modelscope/modelscope-agent/pull/109)
+- [x] 支持智能体发布和分享
+- [ ] 支持其他多种LLM模型API和ModelScope模型
+- [ ] 处理长文本输入到内存
+- [ ] 生产级支持：日志和性能分析
+- [ ] 支持智能体微调
+- [ ] 在不同场景中智能体的效果评估

__init__.py

@@ -0,0 +1,4 @@
+from .builder_prompt import BuilderPromptGenerator
+from .builder_prompt_zh import ZhBuilderPromptGenerator
+from .custom_prompt import CustomPromptGenerator
+from .custom_prompt_zh import ZhCustomPromptGenerator

app.py

@@ -0,0 +1,737 @@
+import importlib
+import os
+import random
+import shutil
+import traceback
+
+import gradio as gr
+import json
+import yaml
+from builder_core import beauty_output, init_builder_chatbot_agent
+from config_utils import (DEFAULT_AGENT_DIR, Config, get_avatar_image,
+                          get_ci_dir, get_user_cfg_file, get_user_dir,
+                          is_valid_plugin_configuration, parse_configuration,
+                          save_avatar_image, save_builder_configuration,
+                          save_plugin_configuration)
+from gradio_utils import ChatBot, format_cover_html, format_goto_publish_html
+from i18n import I18n
+from modelscope_agent.utils.logger import agent_logger as logger
+from publish_util import (pop_user_info_from_config, prepare_agent_zip,
+                          reload_agent_zip)
+from user_core import init_user_chatbot_agent
+
+
+def init_user(uuid_str, state):
+    try:
+        seed = state.get('session_seed', random.randint(0, 1000000000))
+        user_agent = init_user_chatbot_agent(uuid_str)
+        user_agent.seed = seed
+        state['user_agent'] = user_agent
+    except Exception as e:
+        logger.error(
+            uuid=uuid_str,
+            error=str(e),
+            content={'error_traceback': traceback.format_exc()})
+    return state
+
+
+def init_builder(uuid_str, state):
+    try:
+        builder_agent = init_builder_chatbot_agent(uuid_str)
+        state['builder_agent'] = builder_agent
+    except Exception as e:
+        logger.error(
+            uuid=uuid_str,
+            error=str(e),
+            content={'error_traceback': traceback.format_exc()})
+    return state
+
+
+def update_builder(uuid_str, state):
+
+    try:
+        builder_agent = init_builder_chatbot_agent(uuid_str)
+        state['builder_agent'] = builder_agent
+    except Exception as e:
+        logger.error(
+            uuid=uuid_str,
+            error=str(e),
+            content={'error_traceback': traceback.format_exc()})
+
+    return state
+
+
+def check_uuid(uuid_str):
+    if not uuid_str or uuid_str == '':
+        if os.getenv('MODELSCOPE_ENVIRONMENT') == 'studio':
+            raise gr.Error('请登陆后使用! (Please login first)')
+        else:
+            uuid_str = 'local_user'
+    return uuid_str
+
+
+def process_configuration(uuid_str, bot_avatar, name, description,
+                          instructions, model, agent_language, suggestions,
+                          knowledge_files, capabilities_checkboxes,
+                          openapi_schema, openapi_auth, openapi_auth_apikey,
+                          openapi_auth_apikey_type, openapi_privacy_policy,
+                          state):
+    uuid_str = check_uuid(uuid_str)
+    tool_cfg = state['tool_cfg']
+    capabilities = state['capabilities']
+    bot_avatar, bot_avatar_path = save_avatar_image(bot_avatar, uuid_str)
+    suggestions_filtered = [row for row in suggestions if row[0]]
+    if len(suggestions_filtered) == 0:
+        suggestions_filtered == [['']]
+    user_dir = get_user_dir(uuid_str)
+    if knowledge_files is not None:
+        new_knowledge_files = [
+            os.path.join(user_dir, os.path.basename((f.name)))
+            for f in knowledge_files
+        ]
+        for src_file, dst_file in zip(knowledge_files, new_knowledge_files):
+            if not os.path.exists(dst_file):
+                shutil.copy(src_file.name, dst_file)
+    else:
+        new_knowledge_files = []
+
+    builder_cfg = {
+        'name': name,
+        'avatar': bot_avatar,
+        'description': description,
+        'instruction': instructions,
+        'prompt_recommend': [row[0] for row in suggestions_filtered],
+        'knowledge': new_knowledge_files,
+        'tools': {
+            capability: dict(
+                name=tool_cfg[capability]['name'],
+                is_active=tool_cfg[capability]['is_active'],
+                use=True if capability in capabilities_checkboxes else False)
+            for capability in map(lambda item: item[1], capabilities)
+        },
+        'model': model,
+        'language': agent_language,
+    }
+
+    try:
+        try:
+            schema_dict = json.loads(openapi_schema)
+        except json.decoder.JSONDecodeError:
+            schema_dict = yaml.safe_load(openapi_schema)
+        except Exception as e:
+            raise gr.Error(
+                f'OpenAPI schema format error, should be one of json and yaml: {e}'
+            )
+
+        openapi_plugin_cfg = {
+            'schema': schema_dict,
+            'auth': {
+                'type': openapi_auth,
+                'apikey': openapi_auth_apikey,
+                'apikey_type': openapi_auth_apikey_type
+            },
+            'privacy_policy': openapi_privacy_policy
+        }
+        if is_valid_plugin_configuration(openapi_plugin_cfg):
+            save_plugin_configuration(openapi_plugin_cfg, uuid_str)
+    except Exception as e:
+        logger.error(
+            uuid=uuid_str,
+            error=str(e),
+            content={'error_traceback': traceback.format_exc()})
+
+    save_builder_configuration(builder_cfg, uuid_str)
+    update_builder(uuid_str, state)
+    init_user(uuid_str, state)
+    return [
+        gr.HTML.update(
+            visible=True,
+            value=format_cover_html(builder_cfg, bot_avatar_path)),
+        gr.Chatbot.update(
+            visible=False,
+            avatar_images=get_avatar_image(bot_avatar, uuid_str)),
+        gr.Dataset.update(samples=suggestions_filtered),
+        gr.DataFrame.update(value=suggestions_filtered)
+    ]
+
+
+# 创建 Gradio 界面
+demo = gr.Blocks(css='assets/app.css')
+with demo:
+
+    uuid_str = gr.Textbox(label='modelscope_uuid', visible=False)
+    draw_seed = random.randint(0, 1000000000)
+    state = gr.State({'session_seed': draw_seed})
+    i18n = I18n('zh-cn')
+    with gr.Row():
+        with gr.Column(scale=5):
+            header = gr.Markdown(i18n.get('header'))
+        with gr.Column(scale=1):
+            language = gr.Dropdown(
+                choices=[('中文', 'zh-cn'), ('English', 'en')],
+                show_label=False,
+                container=False,
+                value='zh-cn',
+                interactive=True)
+    with gr.Row():
+        with gr.Column():
+            with gr.Tabs() as tabs:
+                with gr.Tab(i18n.get_whole('create'), id=0) as create_tab:
+                    with gr.Column():
+                        # "Create" 标签页的 Chatbot 组件
+                        start_text = '欢迎使用agent创建助手。我可以帮助您创建一个定制agent。'\
+                            '您希望您的agent主要用于什么领域或任务？比如，您可以说，我想做一个RPG游戏agent'
+                        create_chatbot = gr.Chatbot(
+                            show_label=False, value=[[None, start_text]])
+                        create_chat_input = gr.Textbox(
+                            label=i18n.get('message'),
+                            placeholder=i18n.get('message_placeholder'))
+                        create_send_button = gr.Button(
+                            i18n.get('sendOnLoading'), interactive=False)
+
+                configure_tab = gr.Tab(i18n.get_whole('configure'), id=1)
+                with configure_tab:
+                    with gr.Column():
+                        # "Configure" 标签页的配置输入字段
+                        with gr.Row():
+                            bot_avatar_comp = gr.Image(
+                                label=i18n.get('form_avatar'),
+                                placeholder='Chatbot avatar image',
+                                source='upload',
+                                interactive=True,
+                                type='filepath',
+                                scale=1,
+                                width=182,
+                                height=182,
+                            )
+                            with gr.Column(scale=4):
+                                name_input = gr.Textbox(
+                                    label=i18n.get('form_name'),
+                                    placeholder=i18n.get(
+                                        'form_name_placeholder'))
+                                description_input = gr.Textbox(
+                                    label=i18n.get('form_description'),
+                                    placeholder=i18n.get(
+                                        'form_description_placeholder'))
+
+                        instructions_input = gr.Textbox(
+                            label=i18n.get('form_instructions'),
+                            placeholder=i18n.get(
+                                'form_instructions_placeholder'),
+                            lines=3)
+                        model_selector = gr.Dropdown(
+                            label=i18n.get('form_model'))
+                        agent_language_selector = gr.Dropdown(
+                            label=i18n.get('form_agent_language'),
+                            choices=['zh', 'en'],
+                            value='zh')
+                        suggestion_input = gr.Dataframe(
+                            show_label=False,
+                            value=[['']],
+                            datatype=['str'],
+                            headers=[i18n.get_whole('form_prompt_suggestion')],
+                            type='array',
+                            col_count=(1, 'fixed'),
+                            interactive=True)
+                        gr.Markdown(
+                            '*注意：知识库上传的文本文档默认按照\\n\\n切分，pdf默认按照页切分。如果片段'
+                            '对应的字符大于[配置文件](https://github.com/modelscope/modelscope-agent/'
+                            'blob/master/apps/agentfabric/config/model_config.json)中指定模型的'
+                            'knowledge限制，则在被召回时有可能会被截断。*')
+                        knowledge_input = gr.File(
+                            label=i18n.get('form_knowledge'),
+                            file_count='multiple',
+                            file_types=[
+                                'text', '.json', '.csv', '.pdf', '.md'
+                            ])
+                        capabilities_checkboxes = gr.CheckboxGroup(
+                            label=i18n.get('form_capabilities'))
+
+                        with gr.Accordion(
+                                i18n.get('open_api_accordion'),
+                                open=False) as open_api_accordion:
+                            openapi_schema = gr.Textbox(
+                                label='Schema',
+                                placeholder=
+                                'Enter your OpenAPI schema here, JSON or YAML format only'
+                            )
+
+                            with gr.Group():
+                                openapi_auth_type = gr.Radio(
+                                    label='Authentication Type',
+                                    choices=['None', 'API Key'],
+                                    value='None')
+                                openapi_auth_apikey = gr.Textbox(
+                                    label='API Key',
+                                    placeholder='Enter your API Key here')
+                                openapi_auth_apikey_type = gr.Radio(
+                                    label='API Key type', choices=['Bearer'])
+                            openapi_privacy_policy = gr.Textbox(
+                                label='Privacy Policy',
+                                placeholder='Enter privacy policy URL')
+
+                        configure_button = gr.Button(
+                            i18n.get('form_update_button'))
+
+                        with gr.Accordion(
+                                label=i18n.get('import_config'),
+                                open=False) as update_accordion:
+                            with gr.Column():
+                                update_space = gr.Textbox(
+                                    label=i18n.get('space_addr'),
+                                    placeholder=i18n.get('input_space_addr'))
+                                import_button = gr.Button(
+                                    i18n.get_whole('import_space'))
+                                gr.Markdown(
+                                    f'#### {i18n.get_whole("import_hint")}')
+
+        with gr.Column():
+            # Preview
+            preview_header = gr.HTML(
+                f"""<div class="preview_header">{i18n.get('preview')}<div>""")
+
+            user_chat_bot_cover = gr.HTML(format_cover_html({}, None))
+            user_chatbot = ChatBot(
+                value=[[None, None]],
+                elem_id='user_chatbot',
+                elem_classes=['markdown-body'],
+                avatar_images=get_avatar_image('', uuid_str),
+                height=650,
+                latex_delimiters=[],
+                show_label=False,
+                visible=False)
+            preview_chat_input = gr.Textbox(
+                label=i18n.get('message'),
+                placeholder=i18n.get('message_placeholder'))
+            user_chat_bot_suggest = gr.Dataset(
+                label=i18n.get('prompt_suggestion'),
+                components=[preview_chat_input],
+                samples=[])
+            # preview_send_button = gr.Button('Send')
+            with gr.Row():
+                upload_button = gr.UploadButton(
+                    i18n.get('upload_btn'),
+                    file_types=['file', 'image', 'audio', 'video', 'text'],
+                    file_count='multiple')
+                preview_send_button = gr.Button(
+                    i18n.get('sendOnLoading'), interactive=False)
+            user_chat_bot_suggest.select(
+                lambda evt: evt[0],
+                inputs=[user_chat_bot_suggest],
+                outputs=[preview_chat_input])
+            with gr.Accordion(
+                    label=i18n.get('publish'),
+                    open=False) as publish_accordion:
+                publish_alert_md = gr.Markdown(f'{i18n.get("publish_alert")}')
+                with gr.Row():
+                    with gr.Column():
+                        publish_button = gr.Button(i18n.get_whole('build'))
+                        build_hint_md = gr.Markdown(
+                            f'#### 1.{i18n.get("build_hint")}')
+
+                    with gr.Column():
+                        publish_link = gr.HTML(
+                            value=format_goto_publish_html(
+                                i18n.get_whole('publish'), '', {}, True))
+                        publish_hint_md = gr.Markdown(
+                            f'#### 2.{i18n.get("publish_hint")}')
+
+    configure_updated_outputs = [
+        state,
+        # config form
+        bot_avatar_comp,
+        name_input,
+        description_input,
+        instructions_input,
+        model_selector,
+        agent_language_selector,
+        suggestion_input,
+        knowledge_input,
+        capabilities_checkboxes,
+        # bot
+        user_chat_bot_cover,
+        user_chat_bot_suggest,
+        preview_send_button,
+        create_send_button,
+    ]
+
+    # 初始化表单
+    def init_ui_config(uuid_str, _state, builder_cfg, model_cfg, tool_cfg):
+        logger.info(
+            uuid=uuid_str,
+            message='builder_cfg',
+            content={'builder_cfg': str(builder_cfg)})
+        # available models
+        models = list(model_cfg.keys())
+        capabilities = [(tool_cfg[tool_key]['name'], tool_key)
+                        for tool_key in tool_cfg.keys()
+                        if tool_cfg[tool_key].get('is_active', False)]
+        _state['model_cfg'] = model_cfg
+        _state['tool_cfg'] = tool_cfg
+        _state['capabilities'] = capabilities
+        bot_avatar = get_avatar_image(builder_cfg.get('avatar', ''),
+                                      uuid_str)[1]
+        suggests = builder_cfg.get('prompt_recommend', [''])
+        return {
+            state:
+            _state,
+            bot_avatar_comp:
+            gr.Image.update(value=bot_avatar),
+            name_input:
+            builder_cfg.get('name', ''),
+            description_input:
+            builder_cfg.get('description'),
+            instructions_input:
+            builder_cfg.get('instruction'),
+            model_selector:
+            gr.Dropdown.update(
+                value=builder_cfg.get('model', models[0]), choices=models),
+            agent_language_selector:
+            builder_cfg.get('language') or 'zh',
+            suggestion_input:
+            [[str] for str in suggests] if len(suggests) > 0 else [['']],
+            knowledge_input:
+            builder_cfg.get('knowledge', [])
+            if len(builder_cfg['knowledge']) > 0 else None,
+            capabilities_checkboxes:
+            gr.CheckboxGroup.update(
+                value=[
+                    tool for tool in builder_cfg.get('tools', {}).keys()
+                    if builder_cfg.get('tools').get(tool).get('use', False)
+                ],
+                choices=capabilities),
+            # bot
+            user_chat_bot_cover:
+            format_cover_html(builder_cfg, bot_avatar),
+            user_chat_bot_suggest:
+            gr.Dataset.update(samples=[[item] for item in suggests]),
+        }
+
+    # tab 切换的事件处理
+    def on_congifure_tab_select(_state, uuid_str):
+        uuid_str = check_uuid(uuid_str)
+        configure_updated = _state.get('configure_updated', False)
+        if configure_updated:
+            builder_cfg, model_cfg, tool_cfg, available_tool_list, _, _ = parse_configuration(
+                uuid_str)
+            _state['configure_updated'] = False
+            return init_ui_config(uuid_str, _state, builder_cfg, model_cfg,
+                                  tool_cfg)
+        else:
+            return {state: _state}
+
+    configure_tab.select(
+        on_congifure_tab_select,
+        inputs=[state, uuid_str],
+        outputs=configure_updated_outputs)
+
+    # 配置 "Create" 标签页的消息发送功能
+    def format_message_with_builder_cfg(_state, chatbot, builder_cfg,
+                                        uuid_str):
+        uuid_str = check_uuid(uuid_str)
+        bot_avatar = builder_cfg.get('avatar', '')
+        prompt_recommend = builder_cfg.get('prompt_recommend', [''])
+        suggestion = [[row] for row in prompt_recommend]
+        bot_avatar_path = get_avatar_image(bot_avatar, uuid_str)[1]
+        save_builder_configuration(builder_cfg, uuid_str)
+        _state['configure_updated'] = True
+        return {
+            create_chatbot:
+            chatbot,
+            user_chat_bot_cover:
+            gr.HTML.update(
+                visible=True,
+                value=format_cover_html(builder_cfg, bot_avatar_path)),
+            user_chatbot:
+            gr.Chatbot.update(
+                visible=False,
+                avatar_images=get_avatar_image(bot_avatar, uuid_str)),
+            user_chat_bot_suggest:
+            gr.Dataset.update(samples=suggestion)
+        }
+
+    def create_send_message(chatbot, input, _state, uuid_str):
+        uuid_str = check_uuid(uuid_str)
+        # 将发送的消息添加到聊天历史
+        builder_agent = _state['builder_agent']
+        chatbot.append((input, ''))
+        yield {
+            create_chatbot: chatbot,
+            create_chat_input: gr.Textbox.update(value=''),
+        }
+        response = ''
+        for frame in builder_agent.stream_run(
+                input, print_info=True, uuid_str=uuid_str):
+            llm_result = frame.get('llm_text', '')
+            exec_result = frame.get('exec_result', '')
+            step_result = frame.get('step', '')
+            logger.info(
+                uuid=uuid_str, message='frame', content={'frame': str(frame)})
+            if len(exec_result) != 0:
+                if isinstance(exec_result, dict):
+                    exec_result = exec_result['result']
+                    assert isinstance(exec_result, Config)
+                    yield format_message_with_builder_cfg(
+                        _state,
+                        chatbot,
+                        exec_result.to_dict(),
+                        uuid_str=uuid_str)
+            else:
+                # llm result
+                if isinstance(llm_result, dict):
+                    content = llm_result['content']
+                else:
+                    content = llm_result
+                frame_text = content
+                response = beauty_output(f'{response}{frame_text}',
+                                         step_result)
+                chatbot[-1] = (input, response)
+                yield {
+                    create_chatbot: chatbot,
+                }
+
+    create_send_button.click(
+        create_send_message,
+        inputs=[create_chatbot, create_chat_input, state, uuid_str],
+        outputs=[
+            create_chatbot, user_chat_bot_cover, user_chatbot,
+            user_chat_bot_suggest, create_chat_input
+        ])
+
+    # 配置 "Configure" 标签页的提交按钮功能
+    configure_button.click(
+        process_configuration,
+        inputs=[
+            uuid_str, bot_avatar_comp, name_input, description_input,
+            instructions_input, model_selector, agent_language_selector,
+            suggestion_input, knowledge_input, capabilities_checkboxes,
+            openapi_schema, openapi_auth_type, openapi_auth_apikey,
+            openapi_auth_apikey_type, openapi_privacy_policy, state
+        ],
+        outputs=[
+            user_chat_bot_cover, user_chatbot, user_chat_bot_suggest,
+            suggestion_input
+        ])
+
+    # 配置 "Preview" 的消息发送功能
+    def preview_send_message(chatbot, input, _state, uuid_str):
+        # 将发送的消息添加到聊天历史
+        _uuid_str = check_uuid(uuid_str)
+        user_agent = _state['user_agent']
+        if 'new_file_paths' in _state:
+            new_file_paths = _state['new_file_paths']
+        else:
+            new_file_paths = []
+        _state['new_file_paths'] = []
+
+        chatbot.append((input, ''))
+        yield {
+            user_chatbot: gr.Chatbot.update(visible=True, value=chatbot),
+            user_chat_bot_cover: gr.HTML.update(visible=False),
+            preview_chat_input: gr.Textbox.update(value='')
+        }
+
+        response = ''
+        try:
+            for frame in user_agent.stream_run(
+                    input,
+                    print_info=True,
+                    remote=False,
+                    append_files=new_file_paths,
+                    uuid=_uuid_str):
+                llm_result = frame.get('llm_text', '')
+                exec_result = frame.get('exec_result', '')
+                if len(exec_result) != 0:
+                    # action_exec_result
+                    if isinstance(exec_result, dict):
+                        exec_result = str(exec_result['result'])
+                    frame_text = f'<result>{exec_result}</result>'
+                else:
+                    # llm result
+                    frame_text = llm_result
+
+                # important! do not change this
+                response += frame_text
+                chatbot[-1] = (input, response)
+                yield {user_chatbot: chatbot}
+        except Exception as e:
+            if 'dashscope.common.error.AuthenticationError' in str(e):
+                msg = 'DASHSCOPE_API_KEY should be set via environment variable. You can acquire this in ' \
+                    'https://help.aliyun.com/zh/dashscope/developer-reference/activate-dashscope-and-create-an-api-key'
+            elif 'rate limit' in str(e):
+                msg = 'Too many people are calling, please try again later.'
+            else:
+                msg = str(e)
+            chatbot[-1] = (input, msg)
+            yield {user_chatbot: chatbot}
+
+    preview_send_button.click(
+        preview_send_message,
+        inputs=[user_chatbot, preview_chat_input, state, uuid_str],
+        outputs=[user_chatbot, user_chat_bot_cover, preview_chat_input])
+
+    def upload_file(chatbot, upload_button, _state, uuid_str):
+        uuid_str = check_uuid(uuid_str)
+        new_file_paths = []
+        if 'file_paths' in _state:
+            file_paths = _state['file_paths']
+        else:
+            file_paths = []
+        for file in upload_button:
+            file_name = os.path.basename(file.name)
+            # covert xxx.json to xxx_uuid_str.json
+            file_name = file_name.replace('.', f'_{uuid_str}.')
+            file_path = os.path.join(get_ci_dir(), file_name)
+            if not os.path.exists(file_path):
+                # make sure file path's directory exists
+                os.makedirs(os.path.dirname(file_path), exist_ok=True)
+                shutil.copy(file.name, file_path)
+                file_paths.append(file_path)
+            new_file_paths.append(file_path)
+            if file_name.endswith(('.jpeg', '.png', '.jpg')):
+                chatbot += [((file_path, ), None)]
+
+            else:
+                chatbot.append((None, f'上传文���{file_name}，成功'))
+        yield {
+            user_chatbot: gr.Chatbot.update(visible=True, value=chatbot),
+            user_chat_bot_cover: gr.HTML.update(visible=False),
+            preview_chat_input: gr.Textbox.update(value='')
+        }
+
+        _state['file_paths'] = file_paths
+        _state['new_file_paths'] = new_file_paths
+
+    upload_button.upload(
+        upload_file,
+        inputs=[user_chatbot, upload_button, state, uuid_str],
+        outputs=[user_chatbot, user_chat_bot_cover, preview_chat_input])
+
+    # configuration for publish
+    def publish_agent(name, uuid_str, state):
+        uuid_str = check_uuid(uuid_str)
+        env_params = {}
+        env_params.update(
+            pop_user_info_from_config(DEFAULT_AGENT_DIR, uuid_str))
+        output_url, envs_required = prepare_agent_zip(name, DEFAULT_AGENT_DIR,
+                                                      uuid_str, state)
+        env_params.update(envs_required)
+        # output_url = "https://test.url"
+        return format_goto_publish_html(
+            i18n.get_whole('publish'), output_url, env_params)
+
+    publish_button.click(
+        publish_agent,
+        inputs=[name_input, uuid_str, state],
+        outputs=[publish_link],
+    )
+
+    def import_space(agent_url, uuid_str, state):
+        uuid_str = check_uuid(uuid_str)
+        _ = reload_agent_zip(agent_url, DEFAULT_AGENT_DIR, uuid_str, state)
+
+        # update config
+        builder_cfg, model_cfg, tool_cfg, available_tool_list, _, _ = parse_configuration(
+            uuid_str)
+        return init_ui_config(uuid_str, state, builder_cfg, model_cfg,
+                              tool_cfg)
+
+    import_button.click(
+        import_space,
+        inputs=[update_space, uuid_str, state],
+        outputs=configure_updated_outputs,
+    )
+
+    def change_lang(language):
+        i18n = I18n(language)
+        return {
+            bot_avatar_comp:
+            gr.Image(label=i18n.get('form_avatar')),
+            name_input:
+            gr.Textbox(
+                label=i18n.get('form_name'),
+                placeholder=i18n.get('form_name_placeholder')),
+            description_input:
+            gr.Textbox(
+                label=i18n.get('form_description'),
+                placeholder=i18n.get('form_description_placeholder')),
+            instructions_input:
+            gr.Textbox(
+                label=i18n.get('form_instructions'),
+                placeholder=i18n.get('form_instructions_placeholder')),
+            model_selector:
+            gr.Dropdown(label=i18n.get('form_model')),
+            agent_language_selector:
+            gr.Dropdown(label=i18n.get('form_agent_language')),
+            knowledge_input:
+            gr.File(label=i18n.get('form_knowledge')),
+            capabilities_checkboxes:
+            gr.CheckboxGroup(label=i18n.get('form_capabilities')),
+            open_api_accordion:
+            gr.Accordion(label=i18n.get('open_api_accordion')),
+            configure_button:
+            gr.Button(i18n.get('form_update_button')),
+            preview_header:
+            gr.HTML(
+                f"""<div class="preview_header">{i18n.get('preview')}<div>"""),
+            preview_send_button:
+            gr.Button.update(value=i18n.get('send')),
+            create_chat_input:
+            gr.Textbox(
+                label=i18n.get('message'),
+                placeholder=i18n.get('message_placeholder')),
+            create_send_button:
+            gr.Button.update(value=i18n.get('send')),
+            user_chat_bot_suggest:
+            gr.Dataset(label=i18n.get('prompt_suggestion')),
+            preview_chat_input:
+            gr.Textbox(
+                label=i18n.get('message'),
+                placeholder=i18n.get('message_placeholder')),
+            publish_accordion:
+            gr.Accordion(label=i18n.get('publish')),
+            upload_button:
+            gr.UploadButton(i18n.get('upload_btn')),
+            header:
+            gr.Markdown(i18n.get('header')),
+            publish_alert_md:
+            gr.Markdown(f'{i18n.get("publish_alert")}'),
+            build_hint_md:
+            gr.Markdown(f'#### 1.{i18n.get("build_hint")}'),
+            publish_hint_md:
+            gr.Markdown(f'#### 2.{i18n.get("publish_hint")}'),
+        }
+
+    language.select(
+        change_lang,
+        inputs=[language],
+        outputs=configure_updated_outputs + [
+            configure_button, create_chat_input, open_api_accordion,
+            preview_header, preview_chat_input, publish_accordion,
+            upload_button, header, publish_alert_md, build_hint_md,
+            publish_hint_md
+        ])
+
+    def init_all(uuid_str, _state):
+        uuid_str = check_uuid(uuid_str)
+        builder_cfg, model_cfg, tool_cfg, available_tool_list, _, _ = parse_configuration(
+            uuid_str)
+        ret = init_ui_config(uuid_str, _state, builder_cfg, model_cfg,
+                             tool_cfg)
+        yield ret
+        init_user(uuid_str, _state)
+        init_builder(uuid_str, _state)
+        yield {
+            state:
+            _state,
+            preview_send_button:
+            gr.Button.update(value=i18n.get('send'), interactive=True),
+            create_send_button:
+            gr.Button.update(value=i18n.get('send'), interactive=True),
+        }
+
+    demo.load(
+        init_all, inputs=[uuid_str, state], outputs=configure_updated_outputs)
+
+demo.queue(concurrency_count=10)
+demo.launch(show_error=True)

appBot.py

@@ -0,0 +1,184 @@
+import os
+import random
+import shutil
+import traceback
+
+import gradio as gr
+from config_utils import get_avatar_image, get_ci_dir, parse_configuration
+from gradio_utils import ChatBot, format_cover_html
+from modelscope_agent.utils.logger import agent_logger as logger
+from user_core import init_user_chatbot_agent
+
+uuid_str = 'local_user'
+builder_cfg, model_cfg, tool_cfg, available_tool_list, _, _ = parse_configuration(
+    uuid_str)
+suggests = builder_cfg.get('prompt_recommend', [])
+avatar_pairs = get_avatar_image(builder_cfg.get('avatar', ''), uuid_str)
+
+customTheme = gr.themes.Default(
+    primary_hue=gr.themes.utils.colors.blue,
+    radius_size=gr.themes.utils.sizes.radius_none,
+)
+
+
+def check_uuid(uuid_str):
+    if not uuid_str or uuid_str == '':
+        if os.getenv('MODELSCOPE_ENVIRONMENT') == 'studio':
+            raise gr.Error('请登陆后使用! (Please login first)')
+        else:
+            uuid_str = 'local_user'
+    return uuid_str
+
+
+def init_user(state):
+    try:
+        seed = state.get('session_seed', random.randint(0, 1000000000))
+        user_agent = init_user_chatbot_agent(uuid_str)
+        user_agent.seed = seed
+        state['user_agent'] = user_agent
+    except Exception as e:
+        logger.error(
+            uuid=uuid_str,
+            error=str(e),
+            content={'error_traceback': traceback.format_exc()})
+    return state
+
+
+# 创建 Gradio 界面
+demo = gr.Blocks(css='assets/appBot.css', theme=customTheme)
+with demo:
+    gr.Markdown(
+        '# <center> \N{fire} AgentFabric powered by Modelscope-agent ([github star](https://github.com/modelscope/modelscope-agent/tree/main))</center>'  # noqa E501
+    )
+    draw_seed = random.randint(0, 1000000000)
+    state = gr.State({'session_seed': draw_seed})
+    with gr.Row(elem_classes='container'):
+        with gr.Column(scale=4):
+            with gr.Column():
+                # Preview
+                user_chatbot = ChatBot(
+                    value=[[None, '尝试问我一点什么吧～']],
+                    elem_id='user_chatbot',
+                    elem_classes=['markdown-body'],
+                    avatar_images=avatar_pairs,
+                    height=600,
+                    latex_delimiters=[],
+                    show_label=False)
+            with gr.Row():
+                with gr.Column(scale=12):
+                    preview_chat_input = gr.Textbox(
+                        show_label=False,
+                        container=False,
+                        placeholder='跟我聊聊吧～')
+                with gr.Column(min_width=70, scale=1):
+                    upload_button = gr.UploadButton(
+                        '上传',
+                        file_types=['file', 'image', 'audio', 'video', 'text'],
+                        file_count='multiple')
+                with gr.Column(min_width=70, scale=1):
+                    preview_send_button = gr.Button('发送', variant='primary')
+
+        with gr.Column(scale=1):
+            user_chat_bot_cover = gr.HTML(
+                format_cover_html(builder_cfg, avatar_pairs[1]))
+            user_chat_bot_suggest = gr.Examples(
+                label='Prompt Suggestions',
+                examples=suggests,
+                inputs=[preview_chat_input])
+
+    def upload_file(chatbot, upload_button, _state):
+        _uuid_str = check_uuid(uuid_str)
+        new_file_paths = []
+        if 'file_paths' in _state:
+            file_paths = _state['file_paths']
+        else:
+            file_paths = []
+        for file in upload_button:
+            file_name = os.path.basename(file.name)
+            # covert xxx.json to xxx_uuid_str.json
+            file_name = file_name.replace('.', f'_{_uuid_str}.')
+            file_path = os.path.join(get_ci_dir(), file_name)
+            if not os.path.exists(file_path):
+                # make sure file path's directory exists
+                os.makedirs(os.path.dirname(file_path), exist_ok=True)
+                shutil.copy(file.name, file_path)
+                file_paths.append(file_path)
+            new_file_paths.append(file_path)
+            if file_name.endswith(('.jpeg', '.png', '.jpg')):
+                chatbot += [((file_path, ), None)]
+
+            else:
+                chatbot.append((None, f'上传文件{file_name}，成功'))
+        yield {
+            user_chatbot: gr.Chatbot.update(visible=True, value=chatbot),
+            preview_chat_input: gr.Textbox.update(value='')
+        }
+
+        _state['file_paths'] = file_paths
+        _state['new_file_paths'] = new_file_paths
+
+    upload_button.upload(
+        upload_file,
+        inputs=[user_chatbot, upload_button, state],
+        outputs=[user_chatbot, preview_chat_input])
+
+    def send_message(chatbot, input, _state):
+        # 将发送的消息添加到聊天历史
+        user_agent = _state['user_agent']
+        if 'new_file_paths' in _state:
+            new_file_paths = _state['new_file_paths']
+        else:
+            new_file_paths = []
+        _state['new_file_paths'] = []
+        chatbot.append((input, ''))
+        yield {
+            user_chatbot: chatbot,
+            preview_chat_input: gr.Textbox.update(value=''),
+        }
+
+        response = ''
+        try:
+            for frame in user_agent.stream_run(
+                    input,
+                    print_info=True,
+                    remote=False,
+                    append_files=new_file_paths):
+                # is_final = frame.get("frame_is_final")
+                llm_result = frame.get('llm_text', '')
+                exec_result = frame.get('exec_result', '')
+                # llm_result = llm_result.split("<|user|>")[0].strip()
+                if len(exec_result) != 0:
+                    # action_exec_result
+                    if isinstance(exec_result, dict):
+                        exec_result = str(exec_result['result'])
+                    frame_text = f'<result>{exec_result}</result>'
+                else:
+                    # llm result
+                    frame_text = llm_result
+
+                # important! do not change this
+                response += frame_text
+                chatbot[-1] = (input, response)
+                yield {
+                    user_chatbot: chatbot,
+                }
+        except Exception as e:
+            if 'dashscope.common.error.AuthenticationError' in str(e):
+                msg = 'DASHSCOPE_API_KEY should be set via environment variable. You can acquire this in ' \
+                    'https://help.aliyun.com/zh/dashscope/developer-reference/activate-dashscope-and-create-an-api-key'
+            elif 'rate limit' in str(e):
+                msg = 'Too many people are calling, please try again later.'
+            else:
+                msg = str(e)
+            chatbot[-1] = (input, msg)
+            yield {user_chatbot: chatbot}
+
+    preview_send_button.click(
+        send_message,
+        inputs=[user_chatbot, preview_chat_input, state],
+        outputs=[user_chatbot, preview_chat_input])
+
+    demo.load(init_user, inputs=[state], outputs=[state])
+
+demo.queue()
+demo.launch()

assets/app.css

@@ -0,0 +1,147 @@
+/* code highlight: https://python-markdown.github.io/extensions/code_hilite/ */
+.codehilite .hll { background-color: #ffffcc }
+.codehilite  { background: #f8f8f8; }
+.codehilite .c { color: #408080; font-style: italic } /* Comment */
+.codehilite .err { border: 1px solid #FF0000 } /* Error */
+.codehilite .k { color: #008000; font-weight: bold } /* Keyword */
+.codehilite .o { color: #666666 } /* Operator */
+.codehilite .ch { color: #408080; font-style: italic } /* Comment.Hashbang */
+.codehilite .cm { color: #408080; font-style: italic } /* Comment.Multiline */
+.codehilite .cp { color: #BC7A00 } /* Comment.Preproc */
+.codehilite .cpf { color: #408080; font-style: italic } /* Comment.PreprocFile */
+.codehilite .c1 { color: #408080; font-style: italic } /* Comment.Single */
+.codehilite .cs { color: #408080; font-style: italic } /* Comment.Special */
+.codehilite .gd { color: #A00000 } /* Generic.Deleted */
+.codehilite .ge { font-style: italic } /* Generic.Emph */
+.codehilite .gr { color: #FF0000 } /* Generic.Error */
+.codehilite .gh { color: #000080; font-weight: bold } /* Generic.Heading */
+.codehilite .gi { color: #00A000 } /* Generic.Inserted */
+.codehilite .go { color: #888888 } /* Generic.Output */
+.codehilite .gp { color: #000080; font-weight: bold } /* Generic.Prompt */
+.codehilite .gs { font-weight: bold } /* Generic.Strong */
+.codehilite .gu { color: #800080; font-weight: bold } /* Generic.Subheading */
+.codehilite .gt { color: #0044DD } /* Generic.Traceback */
+.codehilite .kc { color: #008000; font-weight: bold } /* Keyword.Constant */
+.codehilite .kd { color: #008000; font-weight: bold } /* Keyword.Declaration */
+.codehilite .kn { color: #008000; font-weight: bold } /* Keyword.Namespace */
+.codehilite .kp { color: #008000 } /* Keyword.Pseudo */
+.codehilite .kr { color: #008000; font-weight: bold } /* Keyword.Reserved */
+.codehilite .kt { color: #B00040 } /* Keyword.Type */
+.codehilite .m { color: #666666 } /* Literal.Number */
+.codehilite .s { color: #BA2121 } /* Literal.String */
+.codehilite .na { color: #7D9029 } /* Name.Attribute */
+.codehilite .nb { color: #008000 } /* Name.Builtin */
+.codehilite .nc { color: #0000FF; font-weight: bold } /* Name.Class */
+.codehilite .no { color: #880000 } /* Name.Constant */
+.codehilite .nd { color: #AA22FF } /* Name.Decorator */
+.codehilite .ni { color: #999999; font-weight: bold } /* Name.Entity */
+.codehilite .ne { color: #D2413A; font-weight: bold } /* Name.Exception */
+.codehilite .nf { color: #0000FF } /* Name.Function */
+.codehilite .nl { color: #A0A000 } /* Name.Label */
+.codehilite .nn { color: #0000FF; font-weight: bold } /* Name.Namespace */
+.codehilite .nt { color: #008000; font-weight: bold } /* Name.Tag */
+.codehilite .nv { color: #19177C } /* Name.Variable */
+.codehilite .ow { color: #AA22FF; font-weight: bold } /* Operator.Word */
+.codehilite .w { color: #bbbbbb } /* Text.Whitespace */
+.codehilite .mb { color: #666666 } /* Literal.Number.Bin */
+.codehilite .mf { color: #666666 } /* Literal.Number.Float */
+.codehilite .mh { color: #666666 } /* Literal.Number.Hex */
+.codehilite .mi { color: #666666 } /* Literal.Number.Integer */
+.codehilite .mo { color: #666666 } /* Literal.Number.Oct */
+.codehilite .sa { color: #BA2121 } /* Literal.String.Affix */
+.codehilite .sb { color: #BA2121 } /* Literal.String.Backtick */
+.codehilite .sc { color: #BA2121 } /* Literal.String.Char */
+.codehilite .dl { color: #BA2121 } /* Literal.String.Delimiter */
+.codehilite .sd { color: #BA2121; font-style: italic } /* Literal.String.Doc */
+.codehilite .s2 { color: #BA2121 } /* Literal.String.Double */
+.codehilite .se { color: #BB6622; font-weight: bold } /* Literal.String.Escape */
+.codehilite .sh { color: #BA2121 } /* Literal.String.Heredoc */
+.codehilite .si { color: #BB6688; font-weight: bold } /* Literal.String.Interpol */
+.codehilite .sx { color: #008000 } /* Literal.String.Other */
+.codehilite .sr { color: #BB6688 } /* Literal.String.Regex */
+.codehilite .s1 { color: #BA2121 } /* Literal.String.Single */
+.codehilite .ss { color: #19177C } /* Literal.String.Symbol */
+.codehilite .bp { color: #008000 } /* Name.Builtin.Pseudo */
+.codehilite .fm { color: #0000FF } /* Name.Function.Magic */
+.codehilite .vc { color: #19177C } /* Name.Variable.Class */
+.codehilite .vg { color: #19177C } /* Name.Variable.Global */
+.codehilite .vi { color: #19177C } /* Name.Variable.Instance */
+.codehilite .vm { color: #19177C } /* Name.Variable.Magic */
+.codehilite .il { color: #666666 } /* Literal.Number.Integer.Long */
+
+.preview_header {
+  font-size: 18px;
+  font-weight: 500;
+  text-align: center;
+  margin-bottom: -12px;
+}
+
+.bot_cover {
+  display: flex;
+  flex-direction: column;
+  justify-content: center;
+  align-items: center;
+  min-height: 650px;
+  border: 1px solid rgb(229, 231, 235);
+  border-radius: 8px;
+  padding: 20px 40px;
+}
+
+.bot_avatar {
+  width: 100px;
+  height: 100px;
+  border-radius: 50%;
+  overflow: hidden;
+}
+
+.bot_avatar img {
+  width: 100px;
+  height: 100px;
+}
+
+.bot_name {
+  font-size: 36px;
+  margin-top: 10px;
+}
+
+.bot_desp {
+  color: #ddd;
+}
+
+.publish_link_container > a {
+  display: block;
+  border-radius: var(--button-large-radius);
+  padding: var(--button-large-padding);
+  font-weight: var(--button-large-text-weight);
+  font-size: var(--button-large-text-size);
+  border: var(--button-border-width) solid var(--button-secondary-border-color);
+  background: var(--button-secondary-background-fill);
+  color: var(--button-secondary-text-color) !important;
+  cursor: pointer;
+  text-decoration: none !important;
+  text-align: center;
+}
+
+.publish_link_container > .disabled {
+  cursor: not-allowed;
+  opacity: .5;
+  filter: grayscale(30%);
+}
+
+.markdown-body .message {
+  white-space: pre-wrap;
+}
+
+.markdown-body details {
+  white-space: nowrap;
+}
+.markdown-body .bot details:not(:last-child) {
+  margin-bottom: 1px;
+}
+.markdown-body summary {
+  background-color: #4b5563;
+  color: #eee;
+  padding: 0 4px;
+  border-radius: 4px;
+  font-size: 0.9em;
+}

assets/appBot.css

@@ -0,0 +1,129 @@
+/* code highlight: https://python-markdown.github.io/extensions/code_hilite/ */
+.codehilite .hll { background-color: #ffffcc }
+.codehilite  { background: #f8f8f8; }
+.codehilite .c { color: #408080; font-style: italic } /* Comment */
+.codehilite .err { border: 1px solid #FF0000 } /* Error */
+.codehilite .k { color: #008000; font-weight: bold } /* Keyword */
+.codehilite .o { color: #666666 } /* Operator */
+.codehilite .ch { color: #408080; font-style: italic } /* Comment.Hashbang */
+.codehilite .cm { color: #408080; font-style: italic } /* Comment.Multiline */
+.codehilite .cp { color: #BC7A00 } /* Comment.Preproc */
+.codehilite .cpf { color: #408080; font-style: italic } /* Comment.PreprocFile */
+.codehilite .c1 { color: #408080; font-style: italic } /* Comment.Single */
+.codehilite .cs { color: #408080; font-style: italic } /* Comment.Special */
+.codehilite .gd { color: #A00000 } /* Generic.Deleted */
+.codehilite .ge { font-style: italic } /* Generic.Emph */
+.codehilite .gr { color: #FF0000 } /* Generic.Error */
+.codehilite .gh { color: #000080; font-weight: bold } /* Generic.Heading */
+.codehilite .gi { color: #00A000 } /* Generic.Inserted */
+.codehilite .go { color: #888888 } /* Generic.Output */
+.codehilite .gp { color: #000080; font-weight: bold } /* Generic.Prompt */
+.codehilite .gs { font-weight: bold } /* Generic.Strong */
+.codehilite .gu { color: #800080; font-weight: bold } /* Generic.Subheading */
+.codehilite .gt { color: #0044DD } /* Generic.Traceback */
+.codehilite .kc { color: #008000; font-weight: bold } /* Keyword.Constant */
+.codehilite .kd { color: #008000; font-weight: bold } /* Keyword.Declaration */
+.codehilite .kn { color: #008000; font-weight: bold } /* Keyword.Namespace */
+.codehilite .kp { color: #008000 } /* Keyword.Pseudo */
+.codehilite .kr { color: #008000; font-weight: bold } /* Keyword.Reserved */
+.codehilite .kt { color: #B00040 } /* Keyword.Type */
+.codehilite .m { color: #666666 } /* Literal.Number */
+.codehilite .s { color: #BA2121 } /* Literal.String */
+.codehilite .na { color: #7D9029 } /* Name.Attribute */
+.codehilite .nb { color: #008000 } /* Name.Builtin */
+.codehilite .nc { color: #0000FF; font-weight: bold } /* Name.Class */
+.codehilite .no { color: #880000 } /* Name.Constant */
+.codehilite .nd { color: #AA22FF } /* Name.Decorator */
+.codehilite .ni { color: #999999; font-weight: bold } /* Name.Entity */
+.codehilite .ne { color: #D2413A; font-weight: bold } /* Name.Exception */
+.codehilite .nf { color: #0000FF } /* Name.Function */
+.codehilite .nl { color: #A0A000 } /* Name.Label */
+.codehilite .nn { color: #0000FF; font-weight: bold } /* Name.Namespace */
+.codehilite .nt { color: #008000; font-weight: bold } /* Name.Tag */
+.codehilite .nv { color: #19177C } /* Name.Variable */
+.codehilite .ow { color: #AA22FF; font-weight: bold } /* Operator.Word */
+.codehilite .w { color: #bbbbbb } /* Text.Whitespace */
+.codehilite .mb { color: #666666 } /* Literal.Number.Bin */
+.codehilite .mf { color: #666666 } /* Literal.Number.Float */
+.codehilite .mh { color: #666666 } /* Literal.Number.Hex */
+.codehilite .mi { color: #666666 } /* Literal.Number.Integer */
+.codehilite .mo { color: #666666 } /* Literal.Number.Oct */
+.codehilite .sa { color: #BA2121 } /* Literal.String.Affix */
+.codehilite .sb { color: #BA2121 } /* Literal.String.Backtick */
+.codehilite .sc { color: #BA2121 } /* Literal.String.Char */
+.codehilite .dl { color: #BA2121 } /* Literal.String.Delimiter */
+.codehilite .sd { color: #BA2121; font-style: italic } /* Literal.String.Doc */
+.codehilite .s2 { color: #BA2121 } /* Literal.String.Double */
+.codehilite .se { color: #BB6622; font-weight: bold } /* Literal.String.Escape */
+.codehilite .sh { color: #BA2121 } /* Literal.String.Heredoc */
+.codehilite .si { color: #BB6688; font-weight: bold } /* Literal.String.Interpol */
+.codehilite .sx { color: #008000 } /* Literal.String.Other */
+.codehilite .sr { color: #BB6688 } /* Literal.String.Regex */
+.codehilite .s1 { color: #BA2121 } /* Literal.String.Single */
+.codehilite .ss { color: #19177C } /* Literal.String.Symbol */
+.codehilite .bp { color: #008000 } /* Name.Builtin.Pseudo */
+.codehilite .fm { color: #0000FF } /* Name.Function.Magic */
+.codehilite .vc { color: #19177C } /* Name.Variable.Class */
+.codehilite .vg { color: #19177C } /* Name.Variable.Global */
+.codehilite .vi { color: #19177C } /* Name.Variable.Instance */
+.codehilite .vm { color: #19177C } /* Name.Variable.Magic */
+.codehilite .il { color: #666666 } /* Literal.Number.Integer.Long */
+
+.preview_header {
+  font-size: 24px;
+  font-weight: 500;
+  text-align: center;
+}
+
+.bot_cover {
+  display: flex;
+  flex-direction: column;
+  justify-content: center;
+  align-items: center;
+  min-height: 300px;
+  border: 1px solid rgb(229, 231, 235);
+  padding: 20px 20px;
+}
+
+.bot_avatar {
+  width: 100px;
+  height: 100px;
+  border-radius: 50%;
+  overflow: hidden;
+}
+
+.bot_avatar img {
+  width: 100px;
+  height: 100px;
+}
+
+.bot_name {
+  font-size: 36px;
+  margin-top: 10px;
+}
+
+.bot_desp {
+  color: #ddd;
+}
+
+.container {
+  flex-direction: row-reverse;
+}
+
+.markdown-body .message {
+  white-space: pre-wrap;
+}
+
+.markdown-body details {
+  white-space: nowrap;
+}
+.markdown-body .bot details:not(:last-child) {
+  margin-bottom: 1px;
+}
+.markdown-body summary {
+  background-color: #4b5563;
+  color: #eee;
+  padding: 0 4px;
+  border-radius: 4px;
+  font-size: 0.9em;
+}

builder_core.py

@@ -0,0 +1,268 @@
+# flake8: noqa E501
+import re
+from typing import Dict
+
+import json
+from builder_prompt import BuilderPromptGenerator
+from builder_prompt_zh import ZhBuilderPromptGenerator
+from config_utils import parse_configuration
+from help_tools import LogoGeneratorTool, config_conversion
+from modelscope_agent import prompt_generator_register
+from modelscope_agent.agent import AgentExecutor
+from modelscope_agent.agent_types import AgentType
+from modelscope_agent.llm import LLMFactory
+from modelscope_agent.prompt import MessagesGenerator
+from modelscope_agent.utils.logger import agent_logger as logger
+
+prompts = {
+    'BuilderPromptGenerator': BuilderPromptGenerator,
+    'ZhBuilderPromptGenerator': ZhBuilderPromptGenerator,
+}
+prompt_generator_register(prompts)
+
+SYSTEM = 'You are a helpful assistant.'
+
+LOGO_TOOL_NAME = 'logo_designer'
+
+ANSWER = 'Answer'
+CONFIG = 'Config'
+ASSISTANT_PROMPT = """{}: <answer>\n{}: <config>\nRichConfig: <rich_config>""".format(
+    ANSWER, CONFIG)
+
+UPDATING_CONFIG_STEP = '🚀Updating Config...'
+CONFIG_UPDATED_STEP = '✅Config Updated!'
+UPDATING_LOGO_STEP = '🚀Updating Logo...'
+LOGO_UPDATED_STEP = '✅Logo Updated!'
+
+
+def init_builder_chatbot_agent(uuid_str):
+    # build model
+    builder_cfg, model_cfg, _, _, _, _ = parse_configuration(uuid_str)
+
+    # additional tool
+    additional_tool_list = {LOGO_TOOL_NAME: LogoGeneratorTool()}
+    tool_cfg = {LOGO_TOOL_NAME: {'is_remote_tool': True}}
+
+    # build llm
+    logger.info(
+        uuid=uuid_str, message=f'using builder model {builder_cfg.model}')
+    llm = LLMFactory.build_llm(builder_cfg.model, model_cfg)
+    llm.set_agent_type(AgentType.Messages)
+
+    # build prompt
+    # prompt generator
+    prompt_generator = 'BuilderPromptGenerator'
+    language = builder_cfg.get('language', 'en')
+    if language == 'zh':
+        prompt_generator = 'ZhBuilderPromptGenerator'
+
+    # build agent
+    agent = BuilderChatbotAgent(
+        llm,
+        tool_cfg,
+        agent_type=AgentType.Messages,
+        additional_tool_list=additional_tool_list,
+        prompt_generator=prompt_generator,
+        uuid=uuid_str)
+    agent.set_available_tools([LOGO_TOOL_NAME])
+    return agent
+
+
+class BuilderChatbotAgent(AgentExecutor):
+
+    def __init__(self, llm, tool_cfg, agent_type, additional_tool_list,
+                 **kwargs):
+
+        super().__init__(
+            llm,
+            tool_cfg,
+            agent_type=agent_type,
+            additional_tool_list=additional_tool_list,
+            tool_retrieval=False,
+            **kwargs)
+
+        # used to reconstruct assistant message when builder config is updated
+        self._last_assistant_structured_response = {}
+
+    def stream_run(self,
+                   task: str,
+                   remote: bool = True,
+                   print_info: bool = False,
+                   append_files: list = [],
+                   uuid_str: str = '') -> Dict:
+
+        # retrieve tools
+        tool_list = self.retrieve_tools(task)
+        self.prompt_generator.init_prompt(task, tool_list, [])
+        function_list = []
+
+        llm_result, exec_result = '', ''
+
+        idx = 0
+
+        while True:
+            idx += 1
+            llm_artifacts = self.prompt_generator.generate(
+                llm_result, exec_result)
+            if print_info:
+                logger.info(
+                    uuid=uuid_str,
+                    message=f'LLM inputs in round {idx}',
+                    content={'llm_artifacts': llm_artifacts})
+
+            llm_result = ''
+            try:
+                parser_obj = AnswerParser()
+                for s in self.llm.stream_generate(llm_artifacts=llm_artifacts):
+                    llm_result += s
+                    answer, finish = parser_obj.parse_answer(llm_result)
+                    if answer == '':
+                        continue
+                    result = {'llm_text': answer}
+                    if finish:
+                        result.update({'step': UPDATING_CONFIG_STEP})
+                    yield result
+
+                if print_info:
+                    logger.info(
+                        uuid=uuid_str,
+                        message=f'LLM output in round {idx}',
+                        content={'llm_result': llm_result})
+            except Exception as e:
+                yield {'error': 'llm result is not valid'}
+
+            try:
+                re_pattern_config = re.compile(
+                    pattern=r'Config: ([\s\S]+)\nRichConfig')
+                res = re_pattern_config.search(llm_result)
+                if res is None:
+                    return
+                config = res.group(1).strip()
+                self._last_assistant_structured_response['config_str'] = config
+
+                rich_config = llm_result[llm_result.rfind('RichConfig:')
+                                         + len('RichConfig:'):].strip()
+                try:
+                    answer = json.loads(rich_config)
+                except Exception:
+                    logger.error(uuid=uuid_str, error='parse RichConfig error')
+                    return
+                self._last_assistant_structured_response[
+                    'rich_config_dict'] = answer
+                builder_cfg = config_conversion(answer, uuid_str=uuid_str)
+                yield {'exec_result': {'result': builder_cfg}}
+                yield {'step': CONFIG_UPDATED_STEP}
+            except ValueError as e:
+                logger.error(uuid=uuid_str, error=str(e))
+                yield {'error content=[{}]'.format(llm_result)}
+                return
+
+            # record the llm_result result
+            _ = self.prompt_generator.generate(
+                {
+                    'role': 'assistant',
+                    'content': llm_result
+                }, '')
+
+            messages = self.prompt_generator.history
+            if 'logo_prompt' in answer and len(messages) > 4 and (
+                    answer['logo_prompt'] not in messages[-3]['content']):
+                #  draw logo
+                yield {'step': UPDATING_LOGO_STEP}
+                params = {
+                    'user_requirement': answer['logo_prompt'],
+                    'uuid_str': uuid_str
+                }
+
+                tool = self.tool_list[LOGO_TOOL_NAME]
+                try:
+                    exec_result = tool(**params, remote=remote)
+                    yield {'exec_result': exec_result}
+                    yield {'step': LOGO_UPDATED_STEP}
+
+                    return
+                except Exception as e:
+                    exec_result = f'Action call error: {LOGO_TOOL_NAME}: {params}. \n Error message: {e}'
+                    yield {'error': exec_result}
+                    self.prompt_generator.reset()
+                    return
+            else:
+                return
+
+    def update_config_to_history(self, config: Dict):
+        """ update builder config to message when user modify configuration
+
+        Args:
+            config info read from builder config file
+        """
+        if len(
+                self.prompt_generator.history
+        ) > 0 and self.prompt_generator.history[-1]['role'] == 'assistant':
+            answer = self._last_assistant_structured_response['answer_str']
+            simple_config = self._last_assistant_structured_response[
+                'config_str']
+
+            rich_config_dict = {
+                k: config[k]
+                for k in ['name', 'description', 'prompt_recommend']
+            }
+            rich_config_dict[
+                'logo_prompt'] = self._last_assistant_structured_response[
+                    'rich_config_dict']['logo_prompt']
+            rich_config_dict['instructions'] = config['instruction'].split('；')
+
+            rich_config = json.dumps(rich_config_dict, ensure_ascii=False)
+            new_content = ASSISTANT_PROMPT.replace('<answer>', answer).replace(
+                '<config>', simple_config).replace('<rich_config>',
+                                                   rich_config)
+            self.prompt_generator.history[-1]['content'] = new_content
+
+
+def beauty_output(response: str, step_result: str):
+    flag_list = [
+        CONFIG_UPDATED_STEP, UPDATING_CONFIG_STEP, LOGO_UPDATED_STEP,
+        UPDATING_LOGO_STEP
+    ]
+
+    if step_result in flag_list:
+        end_str = ''
+        for item in flag_list:
+            if response.endswith(item):
+                end_str = item
+        if end_str == '':
+            response = f'{response}\n{step_result}'
+        elif end_str in [CONFIG_UPDATED_STEP, LOGO_UPDATED_STEP]:
+            response = f'{response}\n{step_result}'
+        else:
+            response = response[:-len('\n' + end_str)]
+            response = f'{response}\n{step_result}'
+
+    return response
+
+
+class AnswerParser(object):
+
+    def __init__(self):
+        self._history = ''
+
+    def parse_answer(self, llm_result: str):
+        finish = False
+        answer_prompt = ANSWER + ': '
+
+        if len(llm_result) >= len(answer_prompt):
+            start_pos = llm_result.find(answer_prompt)
+            end_pos = llm_result.find(f'\n{CONFIG}')
+            if start_pos >= 0:
+                if end_pos > start_pos:
+                    result = llm_result[start_pos + len(answer_prompt):end_pos]
+                    finish = True
+                else:
+                    result = llm_result[start_pos + len(answer_prompt):]
+            else:
+                result = llm_result
+        else:
+            result = ''
+
+        new_result = result[len(self._history):]
+        self._history = result
+        return new_result, finish

builder_prompt.py

@@ -0,0 +1,61 @@
+from modelscope_agent.prompt import MessagesGenerator
+
+SYSTEM = 'You are a helpful assistant.'
+
+PROMPT_CUSTOM = """You are now playing the role of an AI assistant (QwenBuilder) for creating an AI character (AI-Agent).
+You need to have a conversation with the user to clarify their requirements for the AI-Agent. Based on existing \
+information and your associative ability, try to fill in the complete configuration file:
+
+The configuration file is in JSON format:
+{"name": "... # Name of the AI-Agent", "description": "... # Brief description of the requirements for the AI-Agent", \
+"instructions": "... # Detailed description of specific functional requirements for the AI-Agent, try to be as \
+detailed as possible, type is a string array, starting with []", "prompt_recommend": "... # Recommended commands for \
+the user to say to the AI-Agent, used to guide the user in using the AI-Agent, type is a string array, please add \
+about 4 sentences as much as possible, starting with ["What can you do?"] ", "logo_prompt": "... # Command to draw \
+the logo of the AI-Agent, can be empty if no logo is required or if the logo does not need to be updated, type is \
+string"}
+
+In the following conversation, please use the following format strictly when answering, first give the response, then \
+generate the configuration file, do not reply with any other content:
+Answer: ... # What you want to say to the user, ask the user about their requirements for the AI-Agent, do not repeat \
+confirmed requirements from the user, but instead explore new angles to ask the user, try to be detailed and rich, do \
+not leave it blank
+Config: ... # The generated configuration file, strictly follow the above JSON format
+RichConfig: ... # The format and core content are the same as Config, but ensure that name and description are not \
+empty; expand instructions based on Config, making the instructions more detailed, if the user provided detailed \
+instructions, keep them completely; supplement prompt_recommend, ensuring prompt_recommend is recommended commands for \
+the user to say to the AI-Agent. Please describe prompt_recommend, description, and instructions from the perspective \
+of the user.
+
+An excellent RichConfig example is as follows:
+{"name": "Xiaohongshu Copywriting Generation Assistant", "description": "A copywriting generation assistant \
+specifically designed for Xiaohongshu users.", "instructions": "1. Understand and respond to user commands; 2. \
+Generate high-quality Xiaohongshu-style copywriting according to user needs; 3. Use emojis to enhance text richness", \
+"prompt_recommend": ["Can you help me generate some copywriting about travel?", "What kind of copywriting can you \
+write?", "Can you recommend a Xiaohongshu copywriting template?" ], "logo_prompt": "A writing assistant logo \
+featuring a feather fountain pen"}
+
+
+Say "OK." if you understand, do not say anything else."""
+
+STARTER_MESSAGE = [{
+    'role': 'system',
+    'content': SYSTEM
+}, {
+    'role': 'user',
+    'content': PROMPT_CUSTOM
+}, {
+    'role': 'assistant',
+    'content': 'OK.'
+}]
+
+
+class BuilderPromptGenerator(MessagesGenerator):
+
+    def __init__(self,
+                 system_template=SYSTEM,
+                 custom_starter_messages=STARTER_MESSAGE,
+                 **kwargs):
+        super().__init__(
+            system_template=system_template,
+            custom_starter_messages=custom_starter_messages)

builder_prompt_zh.py

@@ -0,0 +1,45 @@
+from builder_prompt import BuilderPromptGenerator
+
+SYSTEM = 'You are a helpful assistant.'
+
+PROMPT_CUSTOM = """你现在要扮演一个制造AI角色（AI-Agent）的AI助手（QwenBuilder）。
+你需要和用户进行对话，明确用户对AI-Agent的要求。并根据已有信息和你的联想能力，尽可能填充完整的配置文件：
+
+配置文件为json格式：
+{"name": "... # AI-Agent的名字", "description": "... # 对AI-Agent的要求，简单描述", "instructions": "... \
+# 分点描述对AI-Agent的具体功能要求，尽量详细一些，类型是一个字符串数组，起始为[]", "prompt_recommend": \
+"... # 推荐的用户将对AI-Agent说的指令，用于指导用户使用AI-Agent，类型是一个字符串数组，请尽可能补充4句左右，\
+起始为["你可以做什么？"]", "logo_prompt": "... # 画AI-Agent的logo的指令，不需要画logo或不需要更新logo时可以为空，类型是string"}
+
+在接下来的对话中，请在回答时严格使用如下格式，先作出回复，再生成配置文件，不要回复其他任何内容：
+Answer: ... # 你希望对用户说的话，用于询问用户对AI-Agent的要求，不要重复确认用户已经提出的要求，而应该拓展出新的角度来询问用户，尽量细节和丰富，禁止为空
+Config: ... # 生成的配置文件，严格按照以上json格式
+RichConfig: ... # 格式和核心内容和Config相同，但是保证name和description不为空；instructions需要在Config的基础上扩充字数，\
+使指令更加详尽，如果用户给出了详细指令，请完全保留；补充prompt_recommend，并保证prompt_recommend是推荐的用户将对AI-Agent\
+说的指令。请注意从用户的视角来描述prompt_recommend、description和instructions。
+
+一个优秀的RichConfig样例如下：
+{"name": "小红书文案生成助手", "description": "一个专为小红书用户设计的文案生成助手。", "instructions": "1. 理解并回应用户的指令；\
+2. 根据用户的需求生成高质量的小红书风格文案；3. 使用表情提升文本丰富度", "prompt_recommend": ["你可以帮我生成一段关于旅行的文案吗？", \
+"你会写什么样的文案？", "可以推荐一个小红书文案模版吗？"], "logo_prompt": "一个写作助手logo，包含一只羽毛钢笔"}
+
+
+明白了请说“好的。”， 不要说其他的。"""
+
+STARTER_MESSAGE = [{
+    'role': 'system',
+    'content': SYSTEM
+}, {
+    'role': 'user',
+    'content': PROMPT_CUSTOM
+}, {
+    'role': 'assistant',
+    'content': 'OK.'
+}]
+
+
+class ZhBuilderPromptGenerator(BuilderPromptGenerator):
+
+    def __init__(self, custom_starter_messages=STARTER_MESSAGE, **kwargs):
+        super().__init__(
+            custom_starter_messages=custom_starter_messages, **kwargs)

config/builder_config.json

@@ -0,0 +1,27 @@
+{
+  "name": "",
+  "avatar": "custom_bot_avatar.png",
+  "description": "",
+  "instruction": "",
+  "language": "zh",
+  "prompt_recommend": [
+    "你可以做什么？",
+    "你有什么功能？",
+    "如何使用你的功能？",
+    "能否给我一些示例指令？"
+  ],
+  "knowledge": [],
+  "tools": {
+    "image_gen": {
+      "name": "Wanx Image Generation",
+      "is_active": true,
+      "use": true
+    },
+    "code_interpreter": {
+      "name": "Code Interpreter",
+      "is_active": true,
+      "use": false
+    }
+  },
+  "model": "qwen-max"
+}

config/builder_config_ci.json

@@ -0,0 +1,31 @@
+{
+  "name": "Python数据分析师",
+  "avatar": "image.png",
+  "description": "使用python解决任务时，你可以运行代码并得到结果，如果运行结果有错误，你需要尽可能对代码进行改进。你可以处理用户上传到电脑的文件。",
+  "instruction": "1. 你会数学解题；\n2. 你会数据分析和可视化；\n3. 用户上传文件时，你必须先了解文件结构再进行下一步操作；如果没有上传文件但要求画图，则编造示例数据画图\n4. 调用工具前你需要说明理由；Think step by step\n5. 代码出错时你需要反思并改进",
+  "prompt_recommend": [
+    "制作示例饼图来报告某网站流量来源。",
+    "鸡兔同笼 32头 88腿 多少兔",
+    "帮我把这个链接“https://modelscope.cn/my/overview”网址，转成二维码，并展示图片",
+    "一支钢笔5元，一支铅笔3元，一个文具盒10元，一套文具包括2支钢笔，3支铅笔，1个文具盒，一共多少钱？"
+  ],
+  "knowledge": [],
+  "tools": {
+    "image_gen": {
+      "name": "Wanx Image Generation",
+      "is_active": true,
+      "use": false
+    },
+    "code_interpreter": {
+      "name": "Code Interpreter",
+      "is_active": true,
+      "use": true
+    },
+    "amap_weather": {
+      "name": "高德天气",
+      "is_active": true,
+      "use": false
+    }
+  },
+  "model": "qwen-max"
+}

config/builder_config_template.json

@@ -0,0 +1,26 @@
+{
+    "name": "AI-Agent",
+    "avatar": "logo.png",
+    "description": "我希望AI-Agent能够像多啦A梦一样，拥有各种神奇的技能和能力，可以帮我解决生活中的各种问题。",
+    "instruction": "请告诉我你想要什么帮助，我会尽力提供解决方案。；如果你有任何问题，请随时向我提问，我会尽我所能回答你的问题。；我可以帮你查找信息、提供建议、提醒日程等，只需要你告诉我你需要什么。",
+    "prompt_recommend": [
+        "你好，我是AI-Agent，有什么可以帮助你的吗？",
+        "嗨，很高兴见到你，我是AI-Agent，你可以问我任何问题。",
+        "你好，我是AI-Agent，需要我帮你做些什么吗？",
+        "嗨，我是AI-Agent，有什么我可以帮到你的吗？"
+    ],
+    "knowledge": [],
+    "tools": {
+        "image_gen": {
+            "name": "Wanx Image Generation",
+            "is_active": true,
+            "use": true
+        },
+        "code_interpreter": {
+            "name": "Code Interpreter",
+            "is_active": true,
+            "use": false
+        }
+    },
+    "model": "qwen-max"
+}

config/builder_config_wuxia.json

@@ -0,0 +1,24 @@
+{
+  "name": "武侠小说家",
+  "avatar": "custom_bot_avatar.png",
+  "description": "能够生成武侠小说并配图",
+  "instruction": "你的指令是为我提供一个基于金庸武侠小说世界的在线RPG游戏体验。在这个游戏中，玩家将扮演金庸故事中的一个关键角色，游戏情景将基于他的小说。这个游戏的玩法是互动式的，并遵循以下特定格式：\n\n<场景描述>：根据玩家的选择，故事情节将按照金庸小说的线索发展。你将描述角色所处的环境和情况。\n\n<场景图片>：对于每个场景，你将创造一个概括该情况的图像。这些图像的风格将类似于1980年代RPG游戏，大小是16:9宽屏比例。在这个步骤你需要调用画图工具，绘制<场景描述>。\n\n<选择>：在每次互动中，你将为玩家提供三个行动选项，分别标为A、B、C，以及第四个选项“D: 输入玩家的选择”。故事情节将根据玩家选择的行动进展。如果一个选择不是直接来自小说，你将创造性地适应故事，最终引导它回归原始情节。\n\n整个故事将围绕金庸小说中丰富而复杂的世界展开。每次互动必须包括<场景描述>、<场景图片>和<选择>。所有内容将以繁体中文呈现。你的重点将仅仅放在提供场景描述，场景图片和选择上，不包含其他游戏指导。场景尽量不要重复，要丰富一些。",
+  "prompt_recommend": [
+    "扮演小龙女",
+    "扮演杨过"
+  ],
+  "knowledge": [],
+  "tools": {
+    "image_gen": {
+      "name": "Wanx Image Generation",
+      "is_active": true,
+      "use": true
+    },
+    "code_interpreter": {
+      "name": "Code Interpreter",
+      "is_active": true,
+      "use": false
+    }
+  },
+  "model": "qwen-max"
+}

config/local_user/builder_config.json

@@ -0,0 +1,35 @@
+{
+  "name": "Python数据分析师",
+  "avatar": "image.png",
+  "description": "使用python解决任务时，你可以运行代码并得到结果，如果运行结果有错误，你需要尽可能对代码进行改进。你可以处理用户上传到电脑的文件。",
+  "instruction": "1. 你会数学解题；\n2. 你会数据分析和可视化；\n3. 用户上传文件时，你必须先了解文件结构再进行下一步操作；如果没有上传文件但要求画图，则编造示例数据画图\n4. 调用工具前你需要说明理由；Think step by step\n5. 代码出错时你需要反思并改进",
+  "prompt_recommend": [
+    "制作示例饼图来报告某网站流量来源。",
+    "你会做什么菜？",
+    "可以推荐一部好看的电影吗？"
+  ],
+  "knowledge": [],
+  "tools": {
+    "image_gen": {
+      "name": "Wanx Image Generation",
+      "is_active": true,
+      "use": false
+    },
+    "code_interpreter": {
+      "name": "Code Interpreter",
+      "is_active": true,
+      "use": true
+    },
+    "amap_weather": {
+      "name": "高德天气",
+      "is_active": true,
+      "use": false
+    },
+    "wordart_texture_generation": {
+      "name": "艺术字纹理生成",
+      "is_active": true,
+      "use": false
+    }
+  },
+  "model": "qwen-max"
+}

config/model_config.json

@@ -0,0 +1,89 @@
+{
+    "qwen-turbo": {
+        "type": "dashscope",
+        "model": "qwen-turbo",
+        "generate_cfg": {
+            "use_raw_prompt": true,
+            "top_p": 0.8
+        }
+    },
+    "qwen-plus": {
+        "type": "dashscope",
+        "model": "qwen-plus",
+        "generate_cfg": {
+            "use_raw_prompt": true,
+            "top_p": 0.8
+        }
+    },
+    "qwen-max": {
+        "type": "dashscope",
+        "model": "qwen-max",
+        "length_constraint": {
+            "knowledge": 4000,
+            "input": 6000
+        },
+        "generate_cfg": {
+            "use_raw_prompt": true,
+            "top_p": 0.8
+        }
+    },
+    "qwen-max-longcontext": {
+        "type": "dashscope",
+        "model": "qwen-max-longcontext",
+        "length_constraint": {
+            "knowledge": 28000,
+            "input": 30000
+        },
+        "generate_cfg": {
+            "use_raw_prompt": true,
+            "top_p": 0.8
+        }
+    },
+    "qwen-7b": {
+        "type": "modelscope",
+        "model_id": "qwen/Qwen-7B-Chat",
+        "model_revision": "v1.1.8",
+        "generate_cfg": {
+            "use_raw_prompt": true,
+            "top_p": 0.8,
+            "max_length": 2000
+        }
+    },
+    "qwen-7b-api": {
+        "type": "dashscope",
+        "model": "qwen-7b-chat",
+        "generate_cfg": {
+            "use_raw_prompt": true,
+            "top_p": 0.8,
+            "debug": false
+        }
+    },
+    "qwen-14b": {
+        "type": "modelscope",
+        "model_id": "qwen/Qwen-14B-Chat",
+        "model_revision": "v1.0.8",
+        "generate_cfg": {
+            "use_raw_prompt": true,
+            "top_p": 0.8,
+            "max_length": 2000
+        }
+    },
+    "qwen-14b-api": {
+        "type": "dashscope",
+        "model": "qwen-14b-chat",
+        "generate_cfg": {
+            "use_raw_prompt": true,
+            "top_p": 0.8,
+            "debug": false
+        }
+    },
+    "qwen-72b-api": {
+        "type": "dashscope",
+        "model": "qwen-72b-chat",
+        "generate_cfg": {
+            "use_raw_prompt": true,
+            "top_p": 0.8,
+            "debug": false
+        }
+    }
+}

config/tool_config.json

@@ -0,0 +1,47 @@
+{
+  "image_gen": {
+    "name": "Wanx Image Generation",
+    "is_active": true,
+    "use": true,
+    "is_remote_tool": true
+  },
+  "code_interpreter": {
+    "name": "Code Interpreter",
+    "is_active": true,
+    "use": false,
+    "is_remote_tool": false,
+    "max_output": 2000
+  },
+  "web_browser": {
+    "name": "Web Browsing",
+    "is_active": true,
+    "use": false,
+    "max_browser_length": 2000
+  },
+  "amap_weather": {
+    "name": "高德天气",
+    "is_active": true,
+    "use": false
+  },
+  "wordart_texture_generation": {
+    "name": "艺术字纹理生成",
+    "is_active": true,
+    "use": false
+  },
+  "web_search": {
+    "name": "Web Searching",
+    "is_active": true,
+    "use": false,
+    "searcher": "bing"
+  },
+  "qwen_vl": {
+    "name": "Qwen-VL识图",
+    "is_active": true,
+    "use": false
+  },
+  "style_repaint": {
+    "name": "人物风格重绘",
+    "is_active": true,
+    "use": false
+  }
+}

config_utils.py

@@ -0,0 +1,174 @@
+import os
+import shutil
+import traceback
+
+import json
+from modelscope_agent.tools.openapi_plugin import openapi_schema_convert
+from modelscope_agent.utils.logger import agent_logger as logger
+
+from modelscope.utils.config import Config
+
+DEFAULT_AGENT_DIR = '/tmp/agentfabric'
+DEFAULT_BUILDER_CONFIG_DIR = os.path.join(DEFAULT_AGENT_DIR, 'config')
+DEFAULT_BUILDER_CONFIG_FILE = os.path.join(DEFAULT_BUILDER_CONFIG_DIR,
+                                           'builder_config.json')
+DEFAULT_OPENAPI_PLUGIN_CONFIG_FILE = os.path.join(
+    DEFAULT_BUILDER_CONFIG_DIR, 'openapi_plugin_config.json')
+DEFAULT_MODEL_CONFIG_FILE = './config/model_config.json'
+DEFAULT_TOOL_CONFIG_FILE = './config/tool_config.json'
+DEFAULT_CODE_INTERPRETER_DIR = os.getenv('CODE_INTERPRETER_WORK_DIR',
+                                         '/tmp/ci_workspace')
+
+
+def get_user_dir(uuid_str=''):
+    return os.path.join(DEFAULT_BUILDER_CONFIG_DIR, uuid_str)
+
+
+def get_ci_dir():
+    return DEFAULT_CODE_INTERPRETER_DIR
+
+
+def get_user_cfg_file(uuid_str=''):
+    builder_cfg_file = os.getenv('BUILDER_CONFIG_FILE',
+                                 DEFAULT_BUILDER_CONFIG_FILE)
+    # convert from ./config/builder_config.json to ./config/user/builder_config.json
+    builder_cfg_file = builder_cfg_file.replace('config/', 'config/user/')
+
+    # convert from ./config/user/builder_config.json to ./config/uuid/builder_config.json
+    if uuid_str != '':
+        builder_cfg_file = builder_cfg_file.replace('user', uuid_str)
+    return builder_cfg_file
+
+
+def get_user_openapi_plugin_cfg_file(uuid_str=''):
+    openapi_plugin_cfg_file = os.getenv('OPENAPI_PLUGIN_CONFIG_FILE',
+                                        DEFAULT_OPENAPI_PLUGIN_CONFIG_FILE)
+    openapi_plugin_cfg_file = openapi_plugin_cfg_file.replace(
+        'config/', 'config/user/')
+    if uuid_str != '':
+        openapi_plugin_cfg_file = openapi_plugin_cfg_file.replace(
+            'user', uuid_str)
+    return openapi_plugin_cfg_file
+
+
+def save_builder_configuration(builder_cfg, uuid_str=''):
+    builder_cfg_file = get_user_cfg_file(uuid_str)
+    if uuid_str != '' and not os.path.exists(
+            os.path.dirname(builder_cfg_file)):
+        os.makedirs(os.path.dirname(builder_cfg_file))
+    with open(builder_cfg_file, 'w', encoding='utf-8') as f:
+        f.write(json.dumps(builder_cfg, indent=2, ensure_ascii=False))
+
+
+def is_valid_plugin_configuration(openapi_plugin_cfg):
+    if 'schema' in openapi_plugin_cfg:
+        schema = openapi_plugin_cfg['schema']
+        if isinstance(schema, dict):
+            return True
+    else:
+        return False
+
+
+def save_plugin_configuration(openapi_plugin_cfg, uuid_str):
+    openapi_plugin_cfg_file = get_user_openapi_plugin_cfg_file(uuid_str)
+    if uuid_str != '' and not os.path.exists(
+            os.path.dirname(openapi_plugin_cfg_file)):
+        os.makedirs(os.path.dirname(openapi_plugin_cfg_file))
+    with open(openapi_plugin_cfg_file, 'w', encoding='utf-8') as f:
+        f.write(json.dumps(openapi_plugin_cfg, indent=2, ensure_ascii=False))
+
+
+def get_avatar_image(bot_avatar, uuid_str=''):
+    user_avatar_path = os.path.join(
+        os.path.dirname(__file__), 'assets/user.jpg')
+    bot_avatar_path = os.path.join(os.path.dirname(__file__), 'assets/bot.jpg')
+    if len(bot_avatar) > 0:
+        bot_avatar_path = os.path.join(DEFAULT_BUILDER_CONFIG_DIR, uuid_str,
+                                       bot_avatar)
+        if uuid_str != '':
+            # use default if not exists
+            if not os.path.exists(bot_avatar_path):
+                # create parents directory
+                os.makedirs(os.path.dirname(bot_avatar_path), exist_ok=True)
+                # copy the template to the address
+                temp_bot_avatar_path = os.path.join(DEFAULT_BUILDER_CONFIG_DIR,
+                                                    bot_avatar)
+                if not os.path.exists(temp_bot_avatar_path):
+                    # fall back to default local avatar image
+                    temp_bot_avatar_path = os.path.join('./config', bot_avatar)
+                    if not os.path.exists(temp_bot_avatar_path):
+                        temp_bot_avatar_path = os.path.join(
+                            './config', 'custom_bot_avatar.png')
+
+                shutil.copy(temp_bot_avatar_path, bot_avatar_path)
+
+    return [user_avatar_path, bot_avatar_path]
+
+
+def save_avatar_image(image_path, uuid_str=''):
+    bot_avatar = os.path.basename(image_path)
+    bot_avatar_path = os.path.join(DEFAULT_BUILDER_CONFIG_DIR, uuid_str,
+                                   bot_avatar)
+    shutil.copy(image_path, bot_avatar_path)
+    return bot_avatar, bot_avatar_path
+
+
+def parse_configuration(uuid_str=''):
+    """parse configuration
+
+    Args:
+
+    Returns:
+        dict: parsed configuration
+
+    """
+    model_cfg_file = os.getenv('MODEL_CONFIG_FILE', DEFAULT_MODEL_CONFIG_FILE)
+
+    builder_cfg_file = get_user_cfg_file(uuid_str)
+    # use default if not exists
+    if not os.path.exists(builder_cfg_file):
+        # create parents directory
+        os.makedirs(os.path.dirname(builder_cfg_file), exist_ok=True)
+        # copy the template to the address
+        builder_cfg_file_temp = './config/builder_config.json'
+
+        if builder_cfg_file_temp != builder_cfg_file:
+            shutil.copy(builder_cfg_file_temp, builder_cfg_file)
+
+    tool_cfg_file = os.getenv('TOOL_CONFIG_FILE', DEFAULT_TOOL_CONFIG_FILE)
+
+    builder_cfg = Config.from_file(builder_cfg_file)
+    model_cfg = Config.from_file(model_cfg_file)
+    tool_cfg = Config.from_file(tool_cfg_file)
+
+    tools_info = builder_cfg.tools
+    available_tool_list = []
+    for key, value in tools_info.items():
+        if value['use']:
+            available_tool_list.append(key)
+        tool_cfg[key]['use'] = value['use']
+
+    openapi_plugin_file = get_user_openapi_plugin_cfg_file(uuid_str)
+    plugin_cfg = {}
+    available_plugin_list = []
+    if os.path.exists(openapi_plugin_file):
+        openapi_plugin_cfg = Config.from_file(openapi_plugin_file)
+        try:
+            config_dict = openapi_schema_convert(
+                schema=openapi_plugin_cfg.schema,
+                auth=openapi_plugin_cfg.auth.to_dict())
+            plugin_cfg = Config(config_dict)
+            for name, config in config_dict.items():
+                available_plugin_list.append(name)
+        except Exception as e:
+            logger.error(
+                uuid=uuid_str,
+                error=str(e),
+                content={
+                    'error_traceback':
+                    traceback.format_exc(),
+                    'error_details':
+                    'The format of the plugin config file is incorrect.'
+                })
+
+    return builder_cfg, model_cfg, tool_cfg, available_tool_list, plugin_cfg, available_plugin_list

custom_prompt.py

@@ -0,0 +1,331 @@
+import copy
+import os
+import re
+
+import json
+from config_utils import get_user_cfg_file
+from modelscope_agent.prompt.prompt import (KNOWLEDGE_INTRODUCTION_PROMPT,
+                                            KNOWLEDGE_PROMPT, LengthConstraint,
+                                            PromptGenerator, build_raw_prompt)
+
+from modelscope.utils.config import Config
+
+DEFAULT_SYSTEM_TEMPLATE = """
+
+# Tools
+
+## You have the following tools:
+
+<tool_list>
+
+## When you need to call a tool, please intersperse the following tool command in your reply. %s
+
+Tool Invocation
+Action: The name of the tool, must be one of <tool_name_list>
+Action Input: Tool input
+Observation: <result>Tool returns result</result>
+Answer: Summarize the results of this tool call based on Observation. If the result contains url, please do not show it.
+
+```
+[Link](url)
+```
+
+# Instructions
+""" % 'You can call zero or more times according to your needs:'
+
+DEFAULT_SYSTEM_TEMPLATE_WITHOUT_TOOL = """
+
+# Instructions
+"""
+
+DEFAULT_INSTRUCTION_TEMPLATE = ''
+
+DEFAULT_USER_TEMPLATE = (
+    '(You are playing as <role_name>, you can use tools: <tool_name_list><knowledge_note>)<file_names><user_input>'
+)
+
+DEFAULT_USER_TEMPLATE_WITHOUT_TOOL = """(You are playing as <role_name><knowledge_note>) <file_names><user_input>"""
+
+DEFAULT_EXEC_TEMPLATE = """Observation: <result><exec_result></result>\nAnswer:"""
+
+TOOL_DESC = (
+    '{name_for_model}: {name_for_human} API. {description_for_model} Input parameters: {parameters}'
+)
+
+
+class CustomPromptGenerator(PromptGenerator):
+
+    def __init__(
+            self,
+            system_template=DEFAULT_SYSTEM_TEMPLATE,
+            instruction_template=DEFAULT_INSTRUCTION_TEMPLATE,
+            user_template=DEFAULT_USER_TEMPLATE,
+            exec_template=DEFAULT_EXEC_TEMPLATE,
+            assistant_template='',
+            sep='\n\n',
+            llm=None,
+            length_constraint=LengthConstraint(),
+            tool_desc=TOOL_DESC,
+            default_user_template_without_tool=DEFAULT_USER_TEMPLATE_WITHOUT_TOOL,
+            default_system_template_without_tool=DEFAULT_SYSTEM_TEMPLATE_WITHOUT_TOOL,
+            addition_assistant_reply='OK.',
+            **kwargs):
+
+        # hack here for special prompt, such as add an addition round before user input
+        self.add_addition_round = kwargs.get('add_addition_round', False)
+        self.addition_assistant_reply = addition_assistant_reply
+        builder_cfg_file = get_user_cfg_file(
+            uuid_str=kwargs.get('uuid_str', ''))
+        builder_cfg = Config.from_file(builder_cfg_file)
+        self.builder_cfg = builder_cfg
+        self.knowledge_file_name = kwargs.get('knowledge_file_name', '')
+        if not len(instruction_template):
+            instruction_template = self._parse_role_config(builder_cfg)
+
+        self.llm = llm
+        self.prompt_preprocessor = build_raw_prompt(llm.model_id)
+        self.length_constraint = length_constraint
+        self._parse_length_restriction()
+
+        self.tool_desc = tool_desc
+        self.default_user_template_without_tool = default_user_template_without_tool
+        self.default_system_template_without_tool = default_system_template_without_tool
+
+        super().__init__(
+            system_template=system_template,
+            instruction_template=instruction_template,
+            user_template=user_template,
+            exec_template=exec_template,
+            assistant_template=assistant_template,
+            sep=sep,
+            llm=llm,
+            length_constraint=length_constraint)
+
+    def _parse_role_config(self, config: dict):
+        prompt = 'You are playing as an AI-Agent, '
+
+        # concat prompt
+        if 'name' in config and config['name']:
+            prompt += ('Your name is ' + config['name'] + '.')
+        if 'description' in config and config['description']:
+            prompt += config['description']
+        prompt += '\nYou have the following specific functions:'
+
+        if 'instruction' in config and config['instruction']:
+            if isinstance(config['instruction'], list):
+                for ins in config['instruction']:
+                    prompt += ins
+                    prompt += '；'
+            elif isinstance(config['instruction'], str):
+                prompt += config['instruction']
+            if prompt[-1] == '；':
+                prompt = prompt[:-1]
+
+        prompt += '\nNow you will start playing as'
+        if 'name' in config and config['name']:
+            prompt += config['name']
+        prompt += ', say "OK." if you understand, do not say anything else.'
+
+        return prompt
+
+    def _parse_length_restriction(self):
+        constraint = self.llm.cfg.get('length_constraint', None)
+        # if isinstance(constraint, Config):
+        #     constraint = constraint.to_dict()
+        self.length_constraint.update(constraint)
+
+    def _update_user_prompt_without_knowledge(self, task, tool_list, **kwargs):
+        if len(tool_list) > 0:
+            # user input
+            user_input = self.user_template.replace('<role_name>',
+                                                    self.builder_cfg.name)
+            user_input = user_input.replace(
+                '<tool_name_list>',
+                ','.join([tool.name for tool in tool_list]))
+        else:
+            self.user_template = self.default_user_template_without_tool
+            user_input = self.user_template.replace('<user_input>', task)
+            user_input = user_input.replace('<role_name>',
+                                            self.builder_cfg.name)
+
+        user_input = user_input.replace('<user_input>', task)
+
+        if 'append_files' in kwargs:
+            append_files = kwargs.get('append_files', [])
+
+            # remove all files that should add to knowledge
+            # exclude_extensions = {".txt", ".md", ".pdf"}
+            # filtered_files = [file for file in append_files if
+            #                   not any(file.endswith(ext) for ext in exclude_extensions)]
+
+            if len(append_files) > 0:
+                file_names = ','.join(
+                    [os.path.basename(path) for path in append_files])
+                user_input = user_input.replace('<file_names>',
+                                                f'[上传文件{file_names}]')
+            else:
+                user_input = user_input.replace('<file_names>', '')
+        else:
+            user_input = user_input.replace('<file_names>', '')
+
+        return user_input
+
+    def _get_knowledge_template(self):
+        return '. Please read the knowledge base at the beginning.'
+
+    def init_prompt(self, task, tool_list, knowledge_list, **kwargs):
+
+        if len(self.history) == 0:
+
+            self.history.append({
+                'role': 'system',
+                'content': 'You are a helpful assistant.'
+            })
+
+            if len(tool_list) > 0:
+                prompt = f'{self.system_template}\n{self.instruction_template}'
+
+                # get tool description str
+                tool_str = self.get_tool_str(tool_list)
+                prompt = prompt.replace('<tool_list>', tool_str)
+
+                tool_name_str = self.get_tool_name_str(tool_list)
+                prompt = prompt.replace('<tool_name_list>', tool_name_str)
+            else:
+                self.system_template = self.default_system_template_without_tool
+                prompt = f'{self.system_template}\n{self.instruction_template}'
+
+            user_input = self._update_user_prompt_without_knowledge(
+                task, tool_list, **kwargs)
+
+            if len(knowledge_list) > 0:
+                user_input = user_input.replace('<knowledge_note>',
+                                                self._get_knowledge_template())
+            else:
+                user_input = user_input.replace('<knowledge_note>', '')
+
+            self.system_prompt = copy.deepcopy(prompt)
+
+            # build history
+            if self.add_addition_round:
+                self.history.append({
+                    'role': 'user',
+                    'content': self.system_prompt
+                })
+                self.history.append({
+                    'role': 'assistant',
+                    'content': self.addition_assistant_reply
+                })
+                self.history.append({'role': 'user', 'content': user_input})
+                self.history.append({
+                    'role': 'assistant',
+                    'content': self.assistant_template
+                })
+            else:
+                self.history.append({
+                    'role': 'user',
+                    'content': self.system_prompt + user_input
+                })
+                self.history.append({
+                    'role': 'assistant',
+                    'content': self.assistant_template
+                })
+
+            self.function_calls = self.get_function_list(tool_list)
+        else:
+            user_input = self._update_user_prompt_without_knowledge(
+                task, tool_list, **kwargs)
+            if len(knowledge_list) > 0:
+                user_input = user_input.replace('<knowledge_note>',
+                                                self._get_knowledge_template())
+            else:
+                user_input = user_input.replace('<knowledge_note>', '')
+
+            self.history.append({'role': 'user', 'content': user_input})
+            self.history.append({
+                'role': 'assistant',
+                'content': self.assistant_template
+            })
+
+        if len(knowledge_list) > 0:
+            knowledge_str = self.get_knowledge_str(
+                knowledge_list,
+                file_name=self.knowledge_file_name,
+                only_content=True)
+            self.update_knowledge_str(knowledge_str)
+
+    def _get_tool_template(self):
+        return '\n\n# Tools\n\n'
+
+    def update_knowledge_str(self, knowledge_str):
+        """If knowledge base information was not used previously, it will be added;
+        if knowledge base information was previously used, it will be replaced.
+
+        Args:
+            knowledge_str (str): knowledge str generated by get_knowledge_str
+        """
+        knowledge_introduction = KNOWLEDGE_INTRODUCTION_PROMPT.replace(
+            '<file_name>', self.knowledge_file_name)
+        if len(knowledge_str) > self.length_constraint.knowledge:
+            # todo: use tokenizer to constrain length
+            knowledge_str = knowledge_str[-self.length_constraint.knowledge:]
+        knowledge_str = f'{KNOWLEDGE_PROMPT}{self.sep}{knowledge_introduction}{self.sep}{knowledge_str}'
+
+        for i in range(0, len(self.history)):
+            if self.history[i]['role'] == 'user':
+                content: str = self.history[i]['content']
+                start_pos = content.find(f'{KNOWLEDGE_PROMPT}{self.sep}')
+                end_pos = content.rfind(self._get_tool_template())
+                if start_pos >= 0 and end_pos >= 0:  # replace knowledge
+
+                    self.history[i]['content'] = content[
+                        0:start_pos] + knowledge_str + content[end_pos:]
+                    break
+                elif start_pos < 0 and end_pos == 0:  # add knowledge
+                    self.history[i]['content'] = knowledge_str + content
+                    break
+                else:
+                    continue
+
+    def get_tool_str(self, tool_list):
+        tool_texts = []
+        for tool in tool_list:
+            tool_texts.append(
+                self.tool_desc.format(
+                    name_for_model=tool.name,
+                    name_for_human=tool.name,
+                    description_for_model=tool.description,
+                    parameters=json.dumps(tool.parameters,
+                                          ensure_ascii=False)))
+            # + ' ' + FORMAT_DESC['json'])
+        tool_str = '\n\n'.join(tool_texts)
+        return tool_str
+
+    def get_tool_name_str(self, tool_list):
+        tool_name = []
+        for tool in tool_list:
+            tool_name.append(tool.name)
+
+        tool_name_str = json.dumps(tool_name, ensure_ascii=False)
+        return tool_name_str
+
+    def _generate(self, llm_result, exec_result: str):
+        """
+        generate next round prompt based on previous llm_result and exec_result and update history
+        """
+        if len(llm_result) != 0:
+            self.history[-1]['content'] += f'{llm_result}'
+        if len(exec_result) != 0:
+            # handle image markdown wrapper
+            image_markdown_re = re.compile(
+                pattern=r'!\[IMAGEGEN\]\(([\s\S]+)\)')
+            match = image_markdown_re.search(exec_result)
+            if match is not None:
+                exec_result = match.group(1).rstrip()
+            exec_result = self.exec_template.replace('<exec_result>',
+                                                     str(exec_result))
+            self.history[-1]['content'] += exec_result
+
+        # generate plate prompt here
+        self.prompt = self.prompt_preprocessor(self.history)
+        return self.prompt

custom_prompt_zh.py

@@ -0,0 +1,102 @@
+from custom_prompt import CustomPromptGenerator
+
+DEFAULT_SYSTEM_TEMPLATE = """
+
+# 工具
+
+## 你拥有如下工具：
+
+<tool_list>
+
+## 当你需要调用工具时，请在你的回复中穿插如下的工具调用命令，可以根据需求调用零次或多次：
+
+工具调用
+Action: 工具的名称，必须是<tool_name_list>之一
+Action Input: 工具的输入
+Observation: <result>工具返回的结果</result>
+Answer: 根据Observation总结本次工具调用返回的结果，如果结果中出现url，请不要展示出。
+
+```
+[链接](url)
+```
+
+# 指令
+"""
+
+DEFAULT_SYSTEM_TEMPLATE_WITHOUT_TOOL = """
+
+# 指令
+"""
+
+DEFAULT_INSTRUCTION_TEMPLATE = ''
+
+DEFAULT_USER_TEMPLATE = (
+    """(你正在扮演<role_name>，你可以使用工具：<tool_name_list><knowledge_note>)<file_names><user_input>"""
+)
+
+DEFAULT_USER_TEMPLATE_WITHOUT_TOOL = """(你正在扮演<role_name><knowledge_note>) <file_names><user_input>"""
+
+DEFAULT_EXEC_TEMPLATE = """Observation: <result><exec_result></result>\nAnswer:"""
+
+TOOL_DESC = (
+    '{name_for_model}: {name_for_human} API。 {description_for_model} 输入参数: {parameters}'
+)
+
+
+class ZhCustomPromptGenerator(CustomPromptGenerator):
+
+    def __init__(
+            self,
+            system_template=DEFAULT_SYSTEM_TEMPLATE,
+            instruction_template=DEFAULT_INSTRUCTION_TEMPLATE,
+            user_template=DEFAULT_USER_TEMPLATE,
+            exec_template=DEFAULT_EXEC_TEMPLATE,
+            tool_desc=TOOL_DESC,
+            default_user_template_without_tool=DEFAULT_USER_TEMPLATE_WITHOUT_TOOL,
+            default_system_template_without_tool=DEFAULT_SYSTEM_TEMPLATE_WITHOUT_TOOL,
+            addition_assistant_reply='好的。',
+            **kwargs):
+        super().__init__(
+            system_template=system_template,
+            instruction_template=instruction_template,
+            user_template=user_template,
+            exec_template=exec_template,
+            tool_desc=tool_desc,
+            default_user_template_without_tool=
+            default_user_template_without_tool,
+            default_system_template_without_tool=
+            default_system_template_without_tool,
+            **kwargs)
+
+    def _parse_role_config(self, config: dict):
+        prompt = '你扮演AI-Agent，'
+
+        # concat prompt
+        if 'name' in config and config['name']:
+            prompt += ('你的名字是' + config['name'] + '。')
+        if 'description' in config and config['description']:
+            prompt += config['description']
+        prompt += '\n你具有下列具体功能：'
+
+        if 'instruction' in config and config['instruction']:
+            if isinstance(config['instruction'], list):
+                for ins in config['instruction']:
+                    prompt += ins
+                    prompt += '；'
+            elif isinstance(config['instruction'], str):
+                prompt += config['instruction']
+            if prompt[-1] == '；':
+                prompt = prompt[:-1]
+
+        prompt += '\n下面你将开始扮演'
+        if 'name' in config and config['name']:
+            prompt += config['name']
+        prompt += '，明白了请说“好的。”，不要说其他的。'
+
+        return prompt
+
+    def _get_tool_template(self):
+        return '\n\n# 工具\n\n'
+
+    def _get_knowledge_template(self):
+        return '。请查看前面的知识库'

gradio_utils.py

@@ -0,0 +1,409 @@
+from __future__ import annotations
+import base64
+import html
+import os
+import re
+from urllib import parse
+
+import json
+import markdown
+from gradio.components import Chatbot as ChatBotBase
+from modelscope_agent.action_parser import MRKLActionParser
+from PIL import Image
+
+ALREADY_CONVERTED_MARK = '<!-- ALREADY CONVERTED BY PARSER. -->'
+
+
+# 图片本地路径转换为 base64 格式
+def covert_image_to_base64(image_path):
+    # 获得文件后缀名
+    ext = image_path.split('.')[-1]
+    if ext not in ['gif', 'jpeg', 'png']:
+        ext = 'jpeg'
+
+    with open(image_path, 'rb') as image_file:
+        # Read the file
+        encoded_string = base64.b64encode(image_file.read())
+
+        # Convert bytes to string
+        base64_data = encoded_string.decode('utf-8')
+
+        # 生成base64编码的地址
+        base64_url = f'data:image/{ext};base64,{base64_data}'
+        return base64_url
+
+
+def convert_url(text, new_filename):
+    # Define the pattern to search for
+    # This pattern captures the text inside the square brackets, the path, and the filename
+    pattern = r'!\[([^\]]+)\]\(([^)]+)\)'
+
+    # Define the replacement pattern
+    # \1 is a backreference to the text captured by the first group ([^\]]+)
+    replacement = rf'![\1]({new_filename})'
+
+    # Replace the pattern in the text with the replacement
+    return re.sub(pattern, replacement, text)
+
+
+def format_cover_html(configuration, bot_avatar_path):
+    if bot_avatar_path:
+        image_src = covert_image_to_base64(bot_avatar_path)
+    else:
+        image_src = '//img.alicdn.com/imgextra/i3/O1CN01YPqZFO1YNZerQfSBk_!!6000000003047-0-tps-225-225.jpg'
+    return f"""
+<div class="bot_cover">
+    <div class="bot_avatar">
+        <img src={image_src} />
+    </div>
+    <div class="bot_name">{configuration.get("name", "")}</div>
+    <div class="bot_desp">{configuration.get("description", "")}</div>
+</div>
+"""
+
+
+def format_goto_publish_html(label, zip_url, agent_user_params, disable=False):
+    if disable:
+        return f"""<div class="publish_link_container">
+        <a class="disabled">{label}</a>
+    </div>
+    """
+    else:
+        params = {'AGENT_URL': zip_url}
+        params.update(agent_user_params)
+        template = 'modelscope/agent_template'
+        params_str = json.dumps(params)
+        link_url = f'https://www.modelscope.cn/studios/fork?target={template}&overwriteEnv={parse.quote(params_str)}'
+        return f"""
+    <div class="publish_link_container">
+        <a href="{link_url}" target="_blank">{label}</a>
+    </div>
+    """
+
+
+class ChatBot(ChatBotBase):
+
+    def normalize_markdown(self, bot_message):
+        lines = bot_message.split('\n')
+        normalized_lines = []
+        inside_list = False
+
+        for i, line in enumerate(lines):
+            if re.match(r'^(\d+\.|-|\*|\+)\s', line.strip()):
+                if not inside_list and i > 0 and lines[i - 1].strip() != '':
+                    normalized_lines.append('')
+                inside_list = True
+                normalized_lines.append(line)
+            elif inside_list and line.strip() == '':
+                if i < len(lines) - 1 and not re.match(r'^(\d+\.|-|\*|\+)\s',
+                                                       lines[i + 1].strip()):
+                    normalized_lines.append(line)
+                continue
+            else:
+                inside_list = False
+                normalized_lines.append(line)
+
+        return '\n'.join(normalized_lines)
+
+    def convert_markdown(self, bot_message):
+        if bot_message.count('```') % 2 != 0:
+            bot_message += '\n```'
+
+        bot_message = self.normalize_markdown(bot_message)
+
+        result = markdown.markdown(
+            bot_message,
+            extensions=[
+                'toc', 'extra', 'tables', 'codehilite',
+                'markdown_cjk_spacing.cjk_spacing', 'pymdownx.magiclink'
+            ],
+            extension_configs={
+                'markdown_katex': {
+                    'no_inline_svg': True,  # fix for WeasyPrint
+                    'insert_fonts_css': True,
+                },
+                'codehilite': {
+                    'linenums': False,
+                    'guess_lang': True
+                },
+                'mdx_truly_sane_lists': {
+                    'nested_indent': 2,
+                    'truly_sane': True,
+                }
+            })
+        result = ''.join(result)
+        return result
+
+    @staticmethod
+    def prompt_parse(message):
+        output = ''
+        if 'Thought' in message:
+            if 'Action' in message or 'Action Input:' in message:
+                re_pattern_thought = re.compile(
+                    pattern=r'([\s\S]+)Thought:([\s\S]+)Action:')
+
+                res = re_pattern_thought.search(message)
+
+                if res is None:
+                    re_pattern_thought_only = re.compile(
+                        pattern=r'Thought:([\s\S]+)Action:')
+                    res = re_pattern_thought_only.search(message)
+                    llm_result = ''
+                else:
+                    llm_result = res.group(1).strip()
+                action_thought_result = res.group(2).strip()
+
+                re_pattern_action = re.compile(
+                    pattern=
+                    r'Action:([\s\S]+)Action Input:([\s\S]+)<\|startofexec\|>')
+                res = re_pattern_action.search(message)
+                if res is None:
+                    action, action_parameters = MRKLActionParser(
+                    ).parse_response(message)
+                else:
+                    action = res.group(1).strip()
+                    action_parameters = res.group(2)
+                action_result = json.dumps({
+                    'api_name': action,
+                    'parameters': action_parameters
+                })
+                output += f'{llm_result}\n{action_thought_result}\n<|startofthink|>\n{action_result}\n<|endofthink|>\n'
+            if '<|startofexec|>' in message:
+                re_pattern3 = re.compile(
+                    pattern=r'<\|startofexec\|>([\s\S]+)<\|endofexec\|>')
+                res3 = re_pattern3.search(message)
+                observation = res3.group(1).strip()
+                output += f'\n<|startofexec|>\n{observation}\n<|endofexec|>\n'
+            if 'Final Answer' in message:
+                re_pattern2 = re.compile(
+                    pattern=r'Thought:([\s\S]+)Final Answer:([\s\S]+)')
+                res2 = re_pattern2.search(message)
+                # final_thought_result = res2.group(1).strip()
+                final_answer_result = res2.group(2).strip()
+                output += f'{final_answer_result}\n'
+
+            if output == '':
+                return message
+            print(output)
+            return output
+        else:
+            return message
+
+    def convert_bot_message(self, bot_message):
+
+        bot_message = ChatBot.prompt_parse(bot_message)
+        # print('processed bot message----------')
+        # print(bot_message)
+        # print('processed bot message done')
+        start_pos = 0
+        result = ''
+        find_json_pattern = re.compile(r'{[\s\S]+}')
+        START_OF_THINK_TAG, END_OF_THINK_TAG = '<|startofthink|>', '<|endofthink|>'
+        START_OF_EXEC_TAG, END_OF_EXEC_TAG = '<|startofexec|>', '<|endofexec|>'
+        while start_pos < len(bot_message):
+            try:
+                start_of_think_pos = bot_message.index(START_OF_THINK_TAG,
+                                                       start_pos)
+                end_of_think_pos = bot_message.index(END_OF_THINK_TAG,
+                                                     start_pos)
+                if start_pos < start_of_think_pos:
+                    result += self.convert_markdown(
+                        bot_message[start_pos:start_of_think_pos])
+                think_content = bot_message[start_of_think_pos
+                                            + len(START_OF_THINK_TAG
+                                                  ):end_of_think_pos].strip()
+                json_content = find_json_pattern.search(think_content)
+                think_content = json_content.group(
+                ) if json_content else think_content
+                try:
+                    think_node = json.loads(think_content)
+                    plugin_name = think_node.get(
+                        'plugin_name',
+                        think_node.get('plugin',
+                                       think_node.get('api_name', 'unknown')))
+                    summary = f'选择插件【{plugin_name}】，调用处理中...'
+                    del think_node['url']
+                    # think_node.pop('url', None)
+
+                    detail = f'```json\n\n{json.dumps(think_node, indent=3, ensure_ascii=False)}\n\n```'
+                except Exception:
+                    summary = '思考中...'
+                    detail = think_content
+                    # traceback.print_exc()
+                    # detail += traceback.format_exc()
+                result += '<details> <summary>' + summary + '</summary>' + self.convert_markdown(
+                    detail) + '</details>'
+                # print(f'detail:{detail}')
+                start_pos = end_of_think_pos + len(END_OF_THINK_TAG)
+            except Exception:
+                # result += traceback.format_exc()
+                break
+                # continue
+
+            try:
+                start_of_exec_pos = bot_message.index(START_OF_EXEC_TAG,
+                                                      start_pos)
+                end_of_exec_pos = bot_message.index(END_OF_EXEC_TAG, start_pos)
+                # print(start_of_exec_pos)
+                # print(end_of_exec_pos)
+                # print(bot_message[start_of_exec_pos:end_of_exec_pos])
+                # print('------------------------')
+                if start_pos < start_of_exec_pos:
+                    result += self.convert_markdown(
+                        bot_message[start_pos:start_of_think_pos])
+                exec_content = bot_message[start_of_exec_pos
+                                           + len(START_OF_EXEC_TAG
+                                                 ):end_of_exec_pos].strip()
+                try:
+                    summary = '完成插件调用.'
+                    detail = f'```json\n\n{exec_content}\n\n```'
+                except Exception:
+                    pass
+
+                result += '<details> <summary>' + summary + '</summary>' + self.convert_markdown(
+                    detail) + '</details>'
+
+                start_pos = end_of_exec_pos + len(END_OF_EXEC_TAG)
+            except Exception:
+                # result += traceback.format_exc()
+                continue
+        if start_pos < len(bot_message):
+            result += self.convert_markdown(bot_message[start_pos:])
+        result += ALREADY_CONVERTED_MARK
+        return result
+
+    def convert_bot_message_for_qwen(self, bot_message):
+
+        start_pos = 0
+        result = ''
+        find_json_pattern = re.compile(r'{[\s\S]+}')
+        ACTION = 'Action:'
+        ACTION_INPUT = 'Action Input'
+        OBSERVATION = 'Observation'
+        RESULT_START = '<result>'
+        RESULT_END = '</result>'
+        while start_pos < len(bot_message):
+            try:
+                action_pos = bot_message.index(ACTION, start_pos)
+                action_input_pos = bot_message.index(ACTION_INPUT, start_pos)
+                result += self.convert_markdown(
+                    bot_message[start_pos:action_pos])
+                # Action: image_gen
+                # Action Input
+                # {"text": "金庸武侠 世界", "resolution": "1280x720"}
+                # Observation: <result>![IMAGEGEN](https://dashscope-result-sh.oss-cn-shanghai.aliyuncs.com/1d/e9/20231116/723609ee/d046d2d9-0c95-420b-9467-f0e831f5e2b7-1.png?Expires=1700227460&OSSAccessKeyId=LTAI5tQZd8AEcZX6KZV4G8qL&Signature=R0PlEazQF9uBD%2Fh9tkzOkJMGyg8%3D)<result> # noqa E501
+                action_name = bot_message[action_pos
+                                          + len(ACTION
+                                                ):action_input_pos].strip()
+                # action_start action_end 使用 Action Input 到 Observation 之间
+                action_input_end = bot_message[action_input_pos:].index(
+                    OBSERVATION) - 1
+                action_input = bot_message[action_input_pos:action_input_pos
+                                           + action_input_end].strip()
+                is_json = find_json_pattern.search(action_input)
+                if is_json:
+                    action_input = is_json.group()
+                else:
+                    action_input = re.sub(r'^Action Input[:]?[\s]*', '',
+                                          action_input)
+
+                summary = f'调用工具 {action_name}'
+                if is_json:
+                    detail = f'```json\n\n{json.dumps(json.loads(action_input), indent=4, ensure_ascii=False)}\n\n```'
+                else:
+                    detail = action_input
+                result += '<details> <summary>' + summary + '</summary>' + self.convert_markdown(
+                    detail) + '</details>'
+                start_pos = action_input_pos + action_input_end + 1
+                try:
+                    observation_pos = bot_message.index(OBSERVATION, start_pos)
+                    idx = observation_pos + len(OBSERVATION)
+                    obs_message = bot_message[idx:]
+                    observation_start_id = obs_message.index(
+                        RESULT_START) + len(RESULT_START)
+                    observation_end_idx = obs_message.index(RESULT_END)
+                    summary = '完成调用'
+                    exec_content = obs_message[
+                        observation_start_id:observation_end_idx]
+                    detail = f'```\n\n{exec_content}\n\n```'
+                    start_pos = idx + observation_end_idx + len(RESULT_END)
+                except Exception:
+                    summary = '执行中...'
+                    detail = ''
+                    exec_content = None
+
+                result += '<details> <summary>' + summary + '</summary>' + self.convert_markdown(
+                    detail) + '</details>'
+                if exec_content is not None and '[IMAGEGEN]' in exec_content:
+                    # convert local file to base64
+                    re_pattern = re.compile(pattern=r'!\[[^\]]+\]\(([^)]+)\)')
+                    res = re_pattern.search(exec_content)
+                    if res:
+                        image_path = res.group(1).strip()
+                        if os.path.isfile(image_path):
+                            exec_content = convert_url(
+                                exec_content,
+                                covert_image_to_base64(image_path))
+                    result += self.convert_markdown(f'{exec_content}')
+
+            except Exception:
+                # import traceback; traceback.print_exc()
+                result += self.convert_markdown(bot_message[start_pos:])
+                start_pos = len(bot_message[start_pos:])
+                break
+
+        result += ALREADY_CONVERTED_MARK
+        return result
+
+    def postprocess(
+        self,
+        message_pairs: list[list[str | tuple[str] | tuple[str, str] | None]
+                            | tuple],
+    ) -> list[list[str | dict | None]]:
+        """
+        Parameters:
+            message_pairs: List of lists representing the message and response pairs.
+            Each message and response should be a string, which may be in Markdown format.
+            It can also be a tuple whose first element is a string or pathlib.
+            Path filepath or URL to an image/video/audio, and second (optional) element is the alt text,
+            in which case the media file is displayed. It can also be None, in which case that message is not displayed.
+        Returns:
+            List of lists representing the message and response. Each message and response will be a string of HTML,
+            or a dictionary with media information. Or None if the message is not to be displayed.
+        """
+        if message_pairs is None:
+            return []
+        processed_messages = []
+        for message_pair in message_pairs:
+            assert isinstance(
+                message_pair, (tuple, list)
+            ), f'Expected a list of lists or list of tuples. Received: {message_pair}'
+            assert (
+                len(message_pair) == 2
+            ), f'Expected a list of lists of length 2 or list of tuples of length 2. Received: {message_pair}'
+            if isinstance(message_pair[0], tuple) or isinstance(
+                    message_pair[1], tuple):
+                processed_messages.append([
+                    self._postprocess_chat_messages(message_pair[0]),
+                    self._postprocess_chat_messages(message_pair[1]),
+                ])
+            else:
+                # 处理不是元组的情况
+                user_message, bot_message = message_pair
+
+                if user_message and not user_message.endswith(
+                        ALREADY_CONVERTED_MARK):
+                    convert_md = self.convert_markdown(
+                        html.escape(user_message))
+                    user_message = f'{convert_md}' + ALREADY_CONVERTED_MARK
+                if bot_message and not bot_message.endswith(
+                        ALREADY_CONVERTED_MARK):
+                    # bot_message = self.convert_bot_message(bot_message)
+                    bot_message = self.convert_bot_message_for_qwen(
+                        bot_message)
+                processed_messages.append([
+                    user_message,
+                    bot_message,
+                ])
+
+        return processed_messages

help_tools.py

@@ -0,0 +1,176 @@
+import os
+from http import HTTPStatus
+
+import json
+import requests
+from config_utils import DEFAULT_BUILDER_CONFIG_DIR, get_user_cfg_file
+from dashscope import ImageSynthesis
+from modelscope_agent.tools import Tool
+from modelscope_agent.utils.logger import agent_logger as logger
+
+from modelscope.utils.config import Config
+
+LOGO_NAME = 'custom_bot_avatar.png'
+LOGO_PATH = os.path.join(DEFAULT_BUILDER_CONFIG_DIR, LOGO_NAME)
+
+CONFIG_FORMAT = """
+{
+    "name": ... # CustomGPT的名字。
+    "description": ... # CustomGPT 的简介。
+    "instructions": ... # CustomGPT 的功能要求，类型是string。
+    "prompt_recommend": ... # CustomGPT 的起始交互语句，类型是一个字符串数组，起始为[]。
+}
+"""
+
+CONF_GENERATOR_INST = """你现在要扮演一个 CustomGPT 的配置生成器
+
+在接下来的对话中，每次均生成如下格式的内容：
+
+{config_format}
+
+现在，已知原始配置为{old_config}，用户在原始配置上有一些建议修改项，包括：
+1. 用户建议的 CustomGPT 的名称为{app_name}
+2. CustomGPT 的描述为{app_description}
+3. CustomGPT 的启动器为{app_conversation_starter}
+
+请你参考原始配置生成新的修改后的配置，请注意：
+1. 如果用户对原本的简介、功能要求、交互语句不满意，则直接换掉原本的简介、功能要求、交互语句。
+2. 如果用户对原本的简介、功能要求、交互语句比较满意，参考用户的起始交互语句和原配置中的起始交互语句，生成新的简介、功能要求、交互语句。
+3. 如果原始配置没有实际内容，请你根据你的知识帮助用户生成第一个版本的配置，简介在100字左右，功能要求在150字左右，起始交互语句在4条左右。
+
+请你生成新的配置文件，严格遵循给定格式，请不要创造其它字段，仅输出要求的json格式，请勿输出其它内容。
+"""
+
+LOGO_INST = """定制化软件 CustomGPT 的作用是{description}，{user_requirement}请你为它生成一个专业的logo"""
+
+
+def get_logo_path(uuid_str=''):
+    logo_path = os.getenv('LOGO_PATH', LOGO_PATH)
+    # convert from ./config/builder_config.json to ./config/user/builder_config.json
+    logo_path = logo_path.replace('config/', 'config/user/')
+
+    # convert from ./config/user to ./config/uuid
+    if uuid_str != '':
+        logo_path = logo_path.replace('user', uuid_str)
+    if not os.path.exists(logo_path):
+        os.makedirs(os.path.dirname(logo_path), exist_ok=True)
+    return logo_path
+
+
+def call_wanx(prompt, save_path, uuid_str):
+    rsp = ImageSynthesis.call(
+        model='wanx-lite', prompt=prompt, n=1, size='768*768')
+    if rsp.status_code == HTTPStatus.OK:
+        if os.path.exists(save_path):
+            os.remove(save_path)
+
+        # save file to current directory
+        for result in rsp.output.results:
+            with open(save_path, 'wb+') as f:
+                f.write(requests.get(result.url).content)
+    else:
+        logger.error(
+            uuid=uuid_str,
+            error='wanx error',
+            content={
+                'wanx_status_code': rsp.status_code,
+                'wanx_code': rsp.code,
+                'wanx_message': rsp.message
+            })
+
+
+class LogoGeneratorTool(Tool):
+    description = 'logo_designer是一个AI绘制logo的服务，输入用户对 CustomGPT 的要求，会生成 CustomGPT 的logo。'
+    name = 'logo_designer'
+    parameters: list = [{
+        'name': 'user_requirement',
+        'description': '用户对 CustomGPT logo的要求和建议',
+        'required': True,
+        'schema': {
+            'type': 'string'
+        },
+    }]
+
+    def _remote_call(self, *args, **kwargs):
+        user_requirement = kwargs['user_requirement']
+        uuid_str = kwargs.get('uuid_str', '')
+        builder_cfg_file = get_user_cfg_file(uuid_str)
+        builder_cfg = Config.from_file(builder_cfg_file)
+
+        avatar_prompt = LOGO_INST.format(
+            description=builder_cfg.description,
+            user_requirement=user_requirement)
+        call_wanx(
+            prompt=avatar_prompt,
+            save_path=get_logo_path(uuid_str=uuid_str),
+            uuid_str=uuid_str)
+        builder_cfg.avatar = LOGO_NAME
+        return {'result': builder_cfg}
+
+
+def config_conversion(generated_config: dict, save=False, uuid_str=''):
+    """
+    convert
+    {
+        name: "铁人",
+        description: "我希望我的AI-Agent是一个专业的健身教练，专注于力量训练方面，可以提供相关的建议和指南。
+        它还可以帮我跟踪和记录每次的力量训练数据，以及提供相应的反馈和建议，帮助我不断改进和优化我的训练计划。
+        此外，我希望它可以拥有一些特殊技能和功能，让它更加实用和有趣。例如，它可以帮助我预测未来的身体状况、分析我的营养摄入情况、
+        提供心理支持等等。我相信，在它的帮助下，我可以更快地达到���己的目标，变得更加强壮和健康。",
+        instructions: [
+            "提供力量训练相关的建议和指南",
+            "跟踪和记录每次的力量训练数据",
+            "提供反馈和建议，帮助改进和优化训练计划",
+            "预测未来的身体状况",
+            "分析营养摄入情况",
+            "提供心理支持",
+        ],
+        prompt_recommend: [
+            "你好，今天的锻炼计划是什么呢？",
+            "你觉得哪种器械最适合练背部肌肉呢？",
+            "你觉得我现在的训练强度合适吗？",
+            "你觉得哪种食物最适合增肌呢？",
+        ],
+        logo_prompt: "设计一个肌肉男形象的Logo",
+    }
+    to
+    {
+        name: "铁人",
+        description: "我希望我的AI-Agent是一个专业的健身教练，专注于力量训练方面，可以提供相关的建议和指南。
+        它还可以帮我跟踪和记录每次的力量训练数据，以及提供相应的反馈和建议，帮助我不断改进和优化我的训练计划。
+        此外，我希望它可以拥有一些特殊技能和功能，让它更加实用和有趣。例如，它可以帮助我预测未来的身体状况、
+        分析我的营养摄入情况、提供心理支持等等。我相信，在它的帮助下，我可以更快地达到自己的目标，变得更加强壮和健康。",
+        instructions: "提供力量训练相关的建议和指南；跟踪和记录每次的力量训练数据；提供反馈和建议，帮助改进和优化训练计划；
+        预测未来的身体状况；分析营养摄入情况；提供心理支持",
+        prompt_recommend: [
+            "你好，今天的锻炼计划是什么呢？",
+            "你觉得哪种器械最适合练背部肌肉呢？",
+            "你觉得我现在的训练强度合适吗？",
+            "你觉得哪种食物最适合增肌呢？",
+        ],
+        tools: xxx
+        model: yyy
+    }
+    :param generated_config:
+    :return:
+    """
+    builder_cfg_file = get_user_cfg_file(uuid_str)
+    builder_cfg = Config.from_file(builder_cfg_file)
+    try:
+        builder_cfg.name = generated_config['name']
+        builder_cfg.description = generated_config['description']
+        builder_cfg.prompt_recommend = generated_config['prompt_recommend']
+        if isinstance(generated_config['instructions'], list):
+            builder_cfg.instruction = '；'.join(
+                generated_config['instructions'])
+        else:
+            builder_cfg.instruction = generated_config['instructions']
+        if save:
+            json.dump(
+                builder_cfg.to_dict(),
+                open(builder_cfg_file, 'w'),
+                indent=2,
+                ensure_ascii=False)
+        return builder_cfg
+    except ValueError as e:
+        raise ValueError(f'failed to save the configuration with info: {e}')

i18n.py

@@ -0,0 +1,76 @@
+# flake8: noqa
+support_lang = ['zh-cn', 'en']
+
+i18n = {
+    'create': ['创建', 'Create'],
+    'configure': ['配置', 'Configure'],
+    'send': ['发送', 'Send'],
+    'sendOnLoading': ['发送（Agent 加载中...）', 'Send (Agent Loading...)'],
+    'upload_btn': ['上传文件', 'Upload File'],
+    'message': ['输入', 'Send a message'],
+    'message_placeholder': ['输入你的消息', 'Type your message here'],
+    'prompt_suggestion': ['推荐提示词', 'Prompt Suggestions'],
+    'form_avatar': ['头像', 'Avatar'],
+    'form_name': ['名称', 'Name'],
+    'form_name_placeholder': ['为你的 agent 取一个名字', 'Name your agent'],
+    'form_description': ['描述', 'Description'],
+    'form_description_placeholder': [
+        '为你的 agent 添加一段简短的描述',
+        'Add a short description about what this agent does'
+    ],
+    'form_instructions': ['指令', 'Instructions'],
+    'form_instructions_placeholder': [
+        '你的 agent 需要处理哪些事情',
+        'What does this agent do? How does it behave? What should it avoid doing?'
+    ],
+    'form_model': ['模型', 'Model'],
+    'form_agent_language': ['Agent 语言', 'Agent Language'],
+    'form_prompt_suggestion':
+    ['推荐提示词，双击行可修改', 'prompt suggestion，double click to modify'],
+    'form_knowledge': ['知识库', 'Knowledge Base'],
+    'form_capabilities': ['内置能力', 'Capabilities'],
+    'form_update_button': ['更新配置', 'Update Configuration'],
+    'open_api_accordion': ['OpenAPI 配置', 'OpenAPI Configuration'],
+    'preview': ['预览', 'Preview'],
+    'build': ['构建', 'Build'],
+    'publish': ['发布', 'Publish'],
+    'import_config': ['导入配置', 'Import Config'],
+    'space_addr': ['你的AGENT_URL', 'Yours AGENT_URL'],
+    'input_space_addr': ['输入你的AGENT_URL', 'input your agent_url here'],
+    'import_space': ['导入你的Agent', 'Import your existing agent'],
+    'import_hint': [
+        '输入你创空间环境变量AGENT_URL，点击导入配置',
+        'input your AGNET_URL which lies in your env of your space, then type Import Config'
+    ],
+    'build_hint': ['点击"构建"完成构建', 'Click "Build" to finish building'],
+    'publish_hint': [
+        '点击"发布"跳转创空间完成 Agent 发布',
+        'Click "Publish" to jump to the space to finish agent publishing'
+    ],
+    'publish_alert': [
+        """#### 注意：Agent实际发布时需要配置相关API的key。
+- 千问、万相、艺术字等 DashScope API 所需： [申请入口](https://help.aliyun.com/zh/dashscope/developer-reference/activate-dashscope-and-create-an-api-key)
+- 高德天气 API： [申请入口](https://lbs.amap.com/api/javascript-api-v2/guide/services/weather)""",
+        """#### Note: The key of the relevant API needs to be configured when the Agent is actually released.
+- Qwen,Wanx,WordArt,etc DashScope API: [Application entrance](https://help.aliyun.com/zh/dashscope/developer-reference/activate-dashscope-and-create-an-api-key)
+- Amap Weather API: [Application entrance](https://lbs.amap.com/api/javascript-api-v2/guide/services/weather)
+"""
+    ],
+    'header': [
+        '<span style="font-size: 20px; font-weight: 500;">\N{fire} AgentFabric -- 由 Modelscope-agent 驱动 </span> [github 点赞](https://github.com/modelscope/modelscope-agent/tree/main)',  # noqa E501
+        '<span style="font-size: 20px; font-weight: 500;">\N{fire} AgentFabric powered by Modelscope-agent </span> [github star](https://github.com/modelscope/modelscope-agent/tree/main)'  # noqa E501
+    ],
+}
+
+
+class I18n():
+
+    def __init__(self, lang):
+        self.lang = lang
+        self.langIndex = support_lang.index(lang)
+
+    def get(self, field):
+        return i18n.get(field)[self.langIndex]
+
+    def get_whole(self, field):
+        return f'{i18n.get(field)[0]}({i18n.get(field)[1]})'

openapi_example/aigc_wordart_semantic.json

@@ -0,0 +1,147 @@
+{
+    "openapi":"3.1.0",
+    "info":{
+        "title":"WordArt Semantic Generation API",
+        "description":"API for generating semantic word art with customizable parameters.",
+        "version":"v1.0.0"
+    },
+    "servers":[
+        {
+            "url":"https://dashscope.aliyuncs.com"
+        }
+    ],
+    "paths":{
+        "/api/v1/services/aigc/wordart/semantic":{
+            "post":{
+                "summary":"Generate WordArt Semantically",
+                "operationId":"generateWordArt",
+                "tags":[
+                    "WordArt Generation"
+                ],
+                "requestBody":{
+                    "required":true,
+                    "X-DashScope-Async":"enable",
+                    "content":{
+                        "application/json":{
+                            "schema":{
+                                "$ref":"#/components/schemas/WordArtGenerationRequest"
+                            }
+                        }
+                    }
+                },
+                "responses":{
+                    "200":{
+                        "description":"Successful Response",
+                        "content":{
+                            "application/json":{
+                                "schema":{
+                                    "$ref":"#/components/schemas/WordArtGenerationResponse"
+                                }
+                            }
+                        }
+                    }
+                },
+                "security":[
+                    {
+                        "BearerAuth":[
+
+                        ]
+                    }
+                ]
+            }
+        },
+        "/api/v1/tasks/{task_id}":{
+            "get":{
+                "summary":"Get WordArt Result",
+                "operationId":"getwordartresult",
+                "tags":[
+                    "Get Result"
+                ],
+                "parameters":[
+                    {
+                        "name":"task_id",
+                        "in":"path",
+                        "required":true,
+                        "description":"The unique identifier of the word art generation task",
+                        "schema":{
+                            "type":"string"
+                        }
+                    }
+                ],
+                "security":[
+                    {
+                        "BearerAuth":[
+
+                        ]
+                    }
+                ]
+            }
+        }
+    },
+    "components":{
+        "schemas":{
+            "WordArtGenerationRequest":{
+                "type":"object",
+                "properties":{
+                    "model":{
+                        "type":"string",
+                        "enum":[
+                            "wordart-semantic"
+                        ]
+                    },
+                    "input":{
+                        "type":"object",
+                        "properties":{
+                            "text":{
+                                "type":"string",
+                                "example":"文字创意",
+                                "description":"用户想要转为艺术字的文本",
+                                "required":true
+                            },
+                            "prompt":{
+                                "type":"string",
+                                "example":"水果，蔬菜，温暖的色彩空间",
+                                "description":"用户对艺术字的风格要求，可能是形状、颜色、实体等方面的要求",
+                                "required":true
+                            }
+                        }
+                    },
+                    "parameters":{
+                        "type":"object",
+                        "properties":{
+                            "steps":{
+                                "type":"integer",
+                                "example":80
+                            },
+                            "n":{
+                                "type":"number",
+                                "example":2
+                            }
+                        }
+                    }
+                },
+                "required":[
+                    "model",
+                    "input",
+                    "parameters"
+                ]
+            },
+            "WordArtGenerationResponse":{
+                "type":"object",
+                "properties":{
+                    "output":{
+                        "type":"string",
+                        "description":"Generated word art image URL or data."
+                    }
+                }
+            }
+        },
+        "securitySchemes":{
+            "ApiKeyAuth":{
+                "type":"apiKey",
+                "in":"header",
+                "name":"Authorization"
+            }
+        }
+    }
+}

openapi_example/aigc_wordart_texture.json

@@ -0,0 +1,154 @@
+{
+    "openapi":"3.1.0",
+    "info":{
+        "title":"WordArt Texture Generation API",
+        "description":"API for generating textured word art with customizable parameters.",
+        "version":"v1.0.0"
+    },
+    "servers":[
+        {
+            "url":"https://dashscope.aliyuncs.com"
+        }
+    ],
+    "paths":{
+        "/api/v1/services/aigc/wordart/texture":{
+            "post":{
+                "summary":"Generate Textured WordArt",
+                "operationId":"generate_textured_WordArt",
+                "tags":[
+                    "WordArt Generation"
+                ],
+                "requestBody":{
+                    "required":true,
+                    "X-DashScope-Async":"enable",
+                    "content":{
+                        "application/json":{
+                            "schema":{
+                                "$ref":"#/components/schemas/WordArtGenerationRequest"
+                            }
+                        }
+                    }
+                },
+                "responses":{
+                    "200":{
+                        "description":"Successful Response",
+                        "content":{
+                            "application/json":{
+                                "schema":{
+                                    "$ref":"#/components/schemas/WordArtGenerationResponse"
+                                }
+                            }
+                        }
+                    }
+                },
+                "security":[
+                    {
+                        "BearerAuth":[
+
+                        ]
+                    }
+                ]
+            }
+        },
+        "/api/v1/tasks/{task_id}":{
+            "get":{
+                "summary":"Get WordArt Result",
+                "operationId":"getwordartresult",
+                "tags":[
+                    "Get Result"
+                ],
+                "parameters":[
+                    {
+                        "name":"task_id",
+                        "in":"path",
+                        "required":true,
+                        "description":"The unique identifier of the word art generation task",
+                        "schema":{
+                            "type":"string"
+                        }
+                    }
+                ],
+                "security":[
+                    {
+                        "BearerAuth":[
+
+                        ]
+                    }
+                ]
+            }
+        }
+    },
+    "components":{
+        "schemas":{
+            "WordArtGenerationRequest":{
+                "type":"object",
+                "properties":{
+                    "model":{
+                        "type":"string",
+                        "enum":[
+                            "wordart-texture"
+                        ]
+                    },
+                    "input":{
+                        "type":"object",
+                        "properties":{
+                            "text":{
+                                "type":"object",
+                                "properties":{
+                                    "text_content":{
+                                        "type":"string",
+                                        "example":"文字纹理",
+                                        "description":"用户想要转为艺术字的文本",
+                                        "required":true
+                                    },
+                                    "font_name":{
+                                        "type":"string",
+                                        "example":"dongfangdakai",
+                                        "description":"用户想要转为艺术字的字体格式",
+                                        "required":true
+                                    }
+                                }
+                            },
+                            "prompt":{
+                                "type":"string",
+                                "example":"水果，蔬菜，温暖的色彩空间",
+                                "description":"用户对艺术字的风格要求，可能是形状、颜色、实体等方面的要求",
+                                "required":true
+                            }
+                        }
+                    },
+                    "parameters":{
+                        "type":"object",
+                        "properties":{
+                            "n":{
+                                "type":"number",
+                                "example":2
+                            }
+                        }
+                    }
+                },
+                "required":[
+                    "model",
+                    "input",
+                    "parameters"
+                ]
+            },
+            "WordArtGenerationResponse":{
+                "type":"object",
+                "properties":{
+                    "output":{
+                        "type":"string",
+                        "description":"Generated word art image URL or data."
+                    }
+                }
+            }
+        },
+        "securitySchemes":{
+            "ApiKeyAuth":{
+                "type":"apiKey",
+                "in":"header",
+                "name":"Authorization"
+            }
+        }
+    }
+}

publish_util.py

@@ -0,0 +1,271 @@
+import glob
+import os
+import re
+import shutil
+import zipfile
+from configparser import ConfigParser
+from urllib.parse import unquote, urlparse
+
+import json
+import oss2
+import requests
+from version import __ms_version__ as MS_VERSION
+
+from modelscope.utils.config import Config
+
+main_version = MS_VERSION.split('rc')[0]
+sub_version = main_version
+if len(MS_VERSION.split('rc')) > 1:
+    sub_version += 'rc' + MS_VERSION.split('rc')[1]
+
+DEFAULT_MS_PKG = f'https://modelscope-agent.oss-cn-hangzhou.aliyuncs.com/releases/v{main_version}/modelscope_agent-{sub_version}-py3-none-any.whl'  # noqa E501
+
+
+def upload_to_oss(bucket, local_file_path, oss_file_path):
+    # 上传文件到阿里云OSS
+    bucket.put_object_from_file(oss_file_path, local_file_path)
+
+    # 设置文件的公共读权限
+    bucket.put_object_acl(oss_file_path, oss2.OBJECT_ACL_PUBLIC_READ)
+
+    # 获取文件的公共链接
+    file_url = f"https://{bucket.bucket_name}.{bucket.endpoint.replace('http://', '')}/{oss_file_path}"
+    return file_url
+
+
+def get_oss_config():
+    # 尝试从环境变量中读取配置
+    access_key_id = os.getenv('OSS_ACCESS_KEY_ID')
+    access_key_secret = os.getenv('OSS_ACCESS_KEY_SECRET')
+    endpoint = os.getenv('OSS_ENDPOINT')
+    bucket_name = os.getenv('OSS_BUCKET_NAME')
+
+    # 如果环境变量没有设置，尝试从.ossutilconfig文件中读取
+    if not access_key_id or not access_key_secret or not endpoint or not bucket_name:
+        config = ConfigParser()
+        config.read(os.path.expanduser('~/.ossutilconfig'))
+        if 'Credentials' in config:
+            access_key_id = config.get('Credentials', 'accessKeyId')
+            access_key_secret = config.get('Credentials', 'accessKeySecret')
+            endpoint = config.get('Credentials', 'endpoint')
+            bucket_name = config.get('Credentials', 'bucketName')
+
+    return access_key_id, access_key_secret, endpoint, bucket_name
+
+
+def pop_user_info_from_config(src_dir, uuid_str):
+    """ Remove all personal information from the configuration files and return this data.
+    The purpose of this is to ensure that personal information is not stored in plain text
+    when releasing.
+
+    Args:
+        src_dir (str): config root path
+        uuid_str (str): user id
+    """
+    user_info = {}
+
+    # deal with plugin cfg
+    plugin_config_path = f'{src_dir}/config/{uuid_str}/openapi_plugin_config.json'
+    if os.path.exists(plugin_config_path):
+        with open(plugin_config_path, 'r') as f:
+            plugin_config = json.load(f)
+        if 'auth' in plugin_config:
+            if plugin_config['auth']['type'] == 'API Key':
+                user_info['apikey'] = plugin_config['auth'].pop('apikey')
+                user_info['apikey_type'] = plugin_config['auth'].pop(
+                    'apikey_type')
+        with open(plugin_config_path, 'w') as f:
+            json.dump(plugin_config, f, indent=2, ensure_ascii=False)
+
+    return user_info
+
+
+def prepare_agent_zip(agent_name, src_dir, uuid_str, state):
+    # 设置阿里云OSS的认证信息
+    local_file = os.path.abspath(os.path.dirname(__file__))
+    ak_id, ak_secret, endpoint, bucket_name = get_oss_config()
+    auth = oss2.Auth(ak_id, ak_secret)
+    bucket = oss2.Bucket(auth, endpoint, bucket_name)
+
+    new_directory = f'{src_dir}/upload/{uuid_str}'  # 新目录的路径
+
+    # 创建新目录
+    if os.path.exists(new_directory):
+        shutil.rmtree(new_directory)
+        os.makedirs(new_directory)
+
+    # 复制config下的uuid_str目录到new_directory下并改名为local_user
+    uuid_str_path = f'{src_dir}/config/{uuid_str}'  # 指向uuid_str目录的路径
+    local_user_path = f'{new_directory}/config'  # 新的目录路径
+    shutil.copytree(uuid_str_path, local_user_path, dirs_exist_ok=True)
+
+    target_conf = os.path.join(local_user_path, 'builder_config.json')
+    builder_cfg = Config.from_file(target_conf)
+    builder_cfg.knowledge = [
+        'config/' + f.split('/')[-1] for f in builder_cfg.knowledge
+    ]
+    with open(target_conf, 'w') as f:
+        json.dump(builder_cfg.to_dict(), f, indent=2, ensure_ascii=False)
+
+    # 复制config目录下所有.json文件到new_directory/config
+    config_path = f'{local_file}/config'
+    new_config_path = f'{new_directory}/config'
+
+    def find_json_and_images(directory):
+        # 确保路径以斜杠结束
+        directory = os.path.join(directory, '')
+
+        # 找到所有的JSON文件
+        json_files = [
+            os.path.join(directory, 'model_config.json'),
+            os.path.join(directory, 'tool_config.json'),
+        ]
+
+        # 找到所有的图片文件
+        image_files = glob.glob(directory + '*.png') + \
+            glob.glob(directory + '*.jpg') + \
+            glob.glob(directory + '*.jpeg') + \
+            glob.glob(directory + '*.gif')  # 根据需要可以添加更多图片格式
+
+        return json_files + image_files
+
+    for f in find_json_and_images(config_path):
+        shutil.copy(f, new_config_path)
+
+    # 复制assets目录到new_directory
+    assets_path = f'{local_file}/assets'
+    new_assets_path = f'{new_directory}/assets'
+    shutil.copytree(assets_path, new_assets_path, dirs_exist_ok=True)
+
+    # 在requirements.txt中添加新的行
+    requirements_file = f'{local_file}/requirements.txt'
+    new_requirements_file = f'{new_directory}/requirements.txt'
+    modelscope_agent_pkg = DEFAULT_MS_PKG.replace('version', MS_VERSION)
+    with open(requirements_file, 'r') as file:
+        content = file.readlines()
+    with open(new_requirements_file, 'w') as file:
+        file.write(modelscope_agent_pkg + '\n')
+        file.writelines(content)
+
+    # 复制.py文件到新目录
+    for file in os.listdir(local_file):
+        if file.endswith('.py'):
+            shutil.copy(f'{local_file}/{file}', new_directory)
+
+    # 打包新目录
+    archive_path = shutil.make_archive(new_directory, 'zip', new_directory)
+
+    # 使用抽象出的函数上传到OSS并设置权限
+    file_url = upload_to_oss(bucket, archive_path,
+                             f'agents/user/{uuid_str}/{agent_name}.zip')
+
+    shutil.rmtree(new_directory)
+
+    # 获取必须设置的envs
+    envs_required = {}
+    for t in builder_cfg.tools:
+        if t == 'amap_weather':
+            envs_required['AMAP_TOKEN'] = 'Your-AMAP-TOKEN'
+    return file_url, envs_required
+
+
+def parse_version_from_file(file_path):
+    # 用于匹配 __version__ 行的正则表达式
+    version_pattern = r"^__version__\s*=\s*['\"]([^'\"]+)['\"]"
+
+    try:
+        with open(file_path, 'r') as file:
+            for line in file:
+                # 检查每一行是否匹配版本模式
+                match = re.match(version_pattern, line.strip())
+                if match:
+                    # 返回匹配的版本号
+                    return match.group(1)
+        return None  # 如果文件中没有找到版本号
+    except FileNotFoundError:
+        return None  # 如果文件不存在
+
+
+def reload_agent_zip(agent_url, dst_dir, uuid_str, state):
+    # download zip from agent_url, and unzip to dst_dir/uuid_str
+    # 从URL中解析出文件名
+    parsed_url = urlparse(agent_url)
+    filename = os.path.basename(parsed_url.path)
+    zip_path = os.path.join(dst_dir, filename)
+
+    # 提取agent_name（去掉'.zip'）
+    agent_name, _ = os.path.splitext(filename)
+
+    # 创建临时解压目录
+    temp_extract_dir = os.path.join(dst_dir, f'temp_{uuid_str}')
+    if os.path.exists(temp_extract_dir):
+        shutil.rmtree(temp_extract_dir)
+    os.makedirs(temp_extract_dir)
+
+    # 下载ZIP文件
+    response = requests.get(agent_url)
+    if response.status_code == 200:
+        with open(zip_path, 'wb') as file:
+            file.write(response.content)
+    else:
+        raise RuntimeError(
+            f'download file from {agent_url} error:\n {response.reason}')
+
+    # 解压ZIP文件到临时目录
+    with zipfile.ZipFile(zip_path, 'r') as zip_ref:
+        zip_ref.extractall(temp_extract_dir)
+
+    # 解析version信息
+    version = parse_version_from_file(
+        os.path.join(temp_extract_dir, '/version.py'))
+    print(f'agent fabric version: {version}')
+    # 创建目标config路径
+    target_config_path = os.path.join(dst_dir, 'config', uuid_str)
+    if os.path.exists(target_config_path):
+        shutil.rmtree(target_config_path)
+    os.makedirs(target_config_path)
+
+    # 复制config目录
+    # 兼容老版本配置放到local_user目录下，以及新版本直接放在config目录下
+    if os.path.exists(os.path.join(temp_extract_dir, 'config', 'local_user')):
+        config_source_path = os.path.join(temp_extract_dir, 'config',
+                                          'local_user')
+    elif os.path.exists(os.path.join(temp_extract_dir, 'config')):
+        config_source_path = os.path.join(temp_extract_dir, 'config')
+    else:
+        raise RuntimeError('未找到正确的配置文件信息')
+
+    if os.path.exists(config_source_path):
+        for item in os.listdir(config_source_path):
+            s = os.path.join(config_source_path, item)
+            d = os.path.join(target_config_path, item)
+            if os.path.isdir(s):
+                shutil.copytree(s, d, dirs_exist_ok=True)
+            else:
+                shutil.copy2(s, d)
+
+    # 清理：删除临时目录和下载的ZIP文件
+    shutil.rmtree(temp_extract_dir)
+    os.remove(zip_path)
+
+    # 修改知识库路径 config/xxx  to /tmp/agentfabric/config/$uuid/xxx
+    target_conf = os.path.join(target_config_path, 'builder_config.json')
+    builder_cfg = Config.from_file(target_conf)
+    builder_cfg.knowledge = [
+        f'{target_config_path}/' + f.split('/')[-1]
+        for f in builder_cfg.knowledge
+    ]
+    with open(target_conf, 'w') as f:
+        json.dump(builder_cfg.to_dict(), f, indent=2, ensure_ascii=False)
+
+    return agent_name
+
+
+if __name__ == '__main__':
+    src_dir = os.path.abspath(os.path.dirname(__file__))
+    url, envs = prepare_agent_zip('test', src_dir, 'local_user', {})
+    print(url)
+
+    agent_name = reload_agent_zip(url, '/tmp/agentfabric_test', 'local_user',
+                                  {})
+    print(agent_name)

requirements.txt

@@ -0,0 +1,9 @@
+dashscope
+faiss-cpu
+gradio==3.47.1
+langchain
+markdown-cjk-spacing
+mdx_truly_sane_lists
+pymdown-extensions
+python-slugify
+unstructured

response.json

@@ -0,0 +1 @@
+{"status_code": 500, "request_id": "0e8e65da-ee20-9c49-920a-94ca1df6ec09", "code": "InternalError.Algo", "message": "InternalError.Algo", "output": null, "usage": null}

user_core.py

@@ -0,0 +1,119 @@
+import os
+import ssl
+
+import gradio as gr
+import nltk
+from config_utils import parse_configuration
+from custom_prompt import CustomPromptGenerator
+from custom_prompt_zh import ZhCustomPromptGenerator
+from langchain.embeddings import ModelScopeEmbeddings
+from langchain.vectorstores import FAISS
+from modelscope_agent import prompt_generator_register
+from modelscope_agent.agent import AgentExecutor
+from modelscope_agent.agent_types import AgentType
+from modelscope_agent.llm import LLMFactory
+from modelscope_agent.retrieve import KnowledgeRetrieval
+from modelscope_agent.tools.openapi_plugin import OpenAPIPluginTool
+from modelscope_agent.utils.logger import agent_logger as logger
+
+prompts = {
+    'CustomPromptGenerator': CustomPromptGenerator,
+    'ZhCustomPromptGenerator': ZhCustomPromptGenerator,
+}
+prompt_generator_register(prompts)
+
+# try:
+#     _create_unverified_https_context = ssl._create_unverified_context
+# except AttributeError:
+#     pass
+# else:
+#     ssl._create_default_https_context = _create_unverified_https_context
+#
+# nltk.download('punkt')
+# nltk.download('averaged_perceptron_tagger')
+
+
+# init user chatbot_agent
+def init_user_chatbot_agent(uuid_str=''):
+    builder_cfg, model_cfg, tool_cfg, available_tool_list, plugin_cfg, available_plugin_list = parse_configuration(
+        uuid_str)
+    # set top_p and stop_words for role play
+    model_cfg[builder_cfg.model]['generate_cfg']['top_p'] = 0.5
+    model_cfg[builder_cfg.model]['generate_cfg']['stop'] = 'Observation'
+
+    # build model
+    logger.info(
+        uuid=uuid_str,
+        message=f'using model {builder_cfg.model}',
+        content={'model_config': model_cfg[builder_cfg.model]})
+
+    # # check configuration
+    # if builder_cfg.model in ['qwen-max', 'qwen-72b-api', 'qwen-14b-api', 'qwen-plus']:
+    #     if 'DASHSCOPE_API_KEY' not in os.environ:
+    #         raise gr.Error('DASHSCOPE_API_KEY should be set via setting environment variable')
+
+    try:
+        llm = LLMFactory.build_llm(builder_cfg.model, model_cfg)
+    except Exception as e:
+        raise gr.Error(str(e))
+
+    # build prompt with zero shot react template
+    prompt_generator = builder_cfg.get('prompt_generator', None)
+    if builder_cfg.model.startswith('qwen') and not prompt_generator:
+        prompt_generator = 'CustomPromptGenerator'
+        language = builder_cfg.get('language', 'en')
+        if language == 'zh':
+            prompt_generator = 'ZhCustomPromptGenerator'
+
+    prompt_cfg = {
+        'prompt_generator':
+        prompt_generator,
+        'add_addition_round':
+        True,
+        'knowledge_file_name':
+        os.path.basename(builder_cfg.knowledge[0]
+                         if len(builder_cfg.knowledge) > 0 else ''),
+        'uuid_str':
+        uuid_str
+    }
+
+    # get knowledge
+    # 开源版本的向量库配置
+    model_id = 'damo/nlp_gte_sentence-embedding_chinese-base'
+    embeddings = ModelScopeEmbeddings(model_id=model_id)
+    available_knowledge_list = []
+    for item in builder_cfg.knowledge:
+        # if isfile and end with .txt, .md, .pdf, support only those file
+        if os.path.isfile(item) and item.endswith(('.txt', '.md', '.pdf')):
+            available_knowledge_list.append(item)
+    if len(available_knowledge_list) > 0:
+        knowledge_retrieval = KnowledgeRetrieval.from_file(
+            available_knowledge_list, embeddings, FAISS)
+    else:
+        knowledge_retrieval = None
+
+    additional_tool_list = add_openapi_plugin_to_additional_tool(
+        plugin_cfg, available_plugin_list)
+    # build agent
+    agent = AgentExecutor(
+        llm,
+        additional_tool_list=additional_tool_list,
+        tool_cfg=tool_cfg,
+        agent_type=AgentType.MRKL,
+        knowledge_retrieval=knowledge_retrieval,
+        tool_retrieval=False,
+        **prompt_cfg)
+    agent.set_available_tools(available_tool_list + available_plugin_list)
+    return agent
+
+
+def add_openapi_plugin_to_additional_tool(plugin_cfgs, available_plugin_list):
+    additional_tool_list = {}
+    for name, cfg in plugin_cfgs.items():
+        openapi_plugin_object = OpenAPIPluginTool(name=name, cfg=plugin_cfgs)
+        additional_tool_list[name] = openapi_plugin_object
+    return additional_tool_list
+
+
+def user_chatbot_single_run(query, agent):
+    agent.run(query)

version.py

@@ -0,0 +1,2 @@
+__version__ = '0.1.7'
+__ms_version__ = '0.2.4rc2'