docs(tutorial): add content about Reasoning, MCP, and Long-Term Memory

docs/.vitepress/config.mts

@@ -63,7 +63,15 @@ export default defineConfig({
                     },
                     {
                         text: 'Agent',
-                        items: [{ text: 'Friday', link: '/agent/friday' }],
+                        items: [
+                            { text: 'Friday', link: '/agent/friday' },
+                            { text: 'Reasoning', link: '/agent/think' },
+                            { text: 'MCP', link: '/agent/MCP' },
+                            {
+                                text: 'Long-Term Memory',
+                                link: '/agent/LongTermMemory',
+                            },
+                        ],
                     },
                 ],
             },
@@ -117,6 +125,12 @@ export default defineConfig({
                         text: '智能体',
                         items: [
                             { text: 'Friday', link: '/zh_CN/agent/friday' },
+                            { text: '思考模式', link: '/zh_CN/agent/think' },
+                            { text: 'MCP', link: '/zh_CN/agent/MCP' },
+                            {
+                                text: '长期记忆',
+                                link: '/zh_CN/agent/LongTermMemory',
+                            },
                         ],
                     },
                 ],

docs/tutorial/en/agent/LongTermMemory.md

@@ -0,0 +1,14 @@
+# Long-Term Memory
+
+Friday supports storing memories in a local database to enable long-term memory.
+
+## How to Set Up Long-Term Memory
+
+In the settings interface, click to enable long-term memory, and fill in the Embedding model and vector database related settings
+![Long-Term Memory Settings Interface](../../zh_CN/agent/assets/memory_config.png)
+
+## Database Selection
+
+- The system has default configuration for Qdrant. Simply choose whether to save to disk, fill in the memory storage address, and you can use long-term memory.
+- If you need to use other vector databases to store memories, please select the corresponding database provider and fill in the required parameters. Friday implements long-term memory based on the mem0 library. You can find the corresponding file in the configs/vector_stores folder of the mem0 library to view the specific parameters required for each database.
+  ![Long-Term Memory Other Providers](../../zh_CN/agent/assets/memory_provider.png)

docs/tutorial/en/agent/MCP.md

@@ -0,0 +1,20 @@
+# MCP Server Management
+
+Friday supports dynamic management of MCP servers:
+
+- Add and remove MCP servers
+- Enable and disable specific MCP servers at any time
+- Support both remote and local MCP server types
+
+## Adding MCP Servers
+
+1. Click the MCP button on the right side of the Friday chat interface to open the MCP server management interface
+   ![MCP Configuration Button](../../zh_CN/agent/assets/mcp_botton.png)
+
+2. Write the MCP server configuration and click Save
+   ![MCP Configuration Interface](../../zh_CN/agent/assets/mcp_config.png)
+
+## Enabling and Disabling MCP Servers
+
+During usage, you can enable or disable specific MCP servers through this switch
+![MCP Switch Button](../../zh_CN/agent/assets/mcp_switch.png)

docs/tutorial/en/agent/friday.md

@@ -43,7 +43,7 @@ We are continuously integrating more features into Friday. The current and upcom
 - [x] Support meta tool
 - [x] Support realtime steering/interruption
 - [x] Support state/Session management
+- [x] Support long-term memory
+- [x] Support user to add MCP server dynamically
 - [ ] 🚧 Support planning and plan visualization
-- [ ] 🚧 Support long-term memory
 - [ ] Support anthropic agent skill
-- [ ] Support user to add MCP server dynamically

docs/tutorial/en/agent/think.md

@@ -0,0 +1,53 @@
+# Reasoning Mode
+
+Friday supports enabling model reasoning mode and displaying the model's reasoning process.
+
+![Reasoning Mode Display](../../zh_CN/agent/assets/think_show.png)
+
+## How to Enable Reasoning Mode
+
+If the model supports reasoning mode, Friday enables it by passing parameters when calling the model API. The parameters supported by different model providers are as follows:
+
+| Model Provider                  | Parameter Name     | Available Values                                    | Description                                         |
+| ------------------------------- | ------------------ | --------------------------------------------------- | --------------------------------------------------- |
+| **OpenAI** (o1/o3)              | `reasoning_effort` | `low` / `medium` / `high`                           | Controls thinking depth, default is medium          |
+| **Anthropic** (Claude 3.7+)     | `thinking`         | `{"type": "enabled", "budget_tokens": 1024-128000}` | Extended thinking mode, requires token budget       |
+| **Gemini** (3.0+)               | `thinking_config`  | `{"thinking_level": "low/medium/high"}`             | Controls reasoning level, high activates Deep Think |
+| **Ollama** (DeepSeek R1/Qwen 3) | `think`            | `true` / `false`                                    | Enable/disable reasoning mode                       |
+| **DashScope** (Qwen)            | `enable_thinking`  | `true` / `false`                                    | Only supports streaming output                      |
+
+## Configuration Examples
+
+All reasoning mode parameters are passed through **`generate_kwargs`**:
+
+### DashScope (Alibaba Cloud)
+
+![DashScope Parameter Configuration](../../zh_CN/agent/assets/dashscope.png)
+
+> **Note**: For non-streaming calls, you must set `enable_thinking=False`, otherwise an error will occur
+
+### OpenAI
+
+![OpenAI Parameter Configuration](../../zh_CN/agent/assets/openai.png)
+
+### Anthropic (Claude)
+
+![Anthropic Parameter Configuration](../../zh_CN/agent/assets/anthropic.png)
+
+### Google Gemini
+
+![Gemini Parameter Configuration](../../zh_CN/agent/assets/gemini.png)
+
+> **Note**: Gemini 2.5 uses `thinking_budget` parameter (integer value), while Gemini 3+ uses `thinking_level`
+
+### Ollama
+
+![Ollama Parameter Configuration](../../zh_CN/agent/assets/ollama.png)
+
+## Parameter Selection Recommendations
+
+| Task Type                                                                      | Recommended Level | Description                         |
+| ------------------------------------------------------------------------------ | ----------------- | ----------------------------------- |
+| Simple Tasks (translation, classification, data extraction)                    | `low`             | Fast response, low cost             |
+| Daily Development (code generation, content writing, debugging)                | `medium`          | Balance performance and cost        |
+| Complex Reasoning (mathematical proofs, scientific analysis, algorithm design) | `high`            | Deep reasoning, high-quality output |

docs/tutorial/zh_CN/agent/LongTermMemory.md

@@ -0,0 +1,14 @@
+# 长期记忆
+
+Friday支持将记忆存储于本地数据库以实现长期记忆。
+
+## 如何设置长期记忆
+
+在设置界面点击启用长期记忆，填写Embedding模型以及向量数据库相关设置
+![长期记忆 设置界面](assets/memory_config.png)
+
+## 数据库选择
+
+- 系统对Qdrant进行了默认配置，选择是否保存磁盘，填写记忆存储地址即可使用长期记忆。
+- 如需使用其他向量数据库存储记忆，请选择对应的数据库提供者，并填写相应参数进行使用。Friday基于mem0库实现长期记忆，可以于mem0库中的configs/vector_stores文件夹找到对应文件查看具体数据库所需参数。
+  ![长期记忆 其他提供者](assets/memory_provider.png)

docs/tutorial/zh_CN/agent/MCP.md

@@ -0,0 +1,20 @@
+# MCP 服务器管理
+
+Friday支持用户动态管理 MCP 服务器
+
+- 可以添加及删除 MCP 服务器
+- 可以随时开启和关闭指定 MCP 服务器
+- 支持远程和本地两种类型 MCP 服务器
+
+## 添加 MCP 服务器
+
+1.在Friday对话界面有边框点击MCP按钮打开MCP服务器管理界面
+![MCP 配置按钮](assets/mcp_botton.png)
+
+2.写 MCP 服务器配置，点击保存即可
+![MCP 配置界面](assets/mcp_config.png)
+
+## 开启及关闭 MCP 服务器
+
+在使用过程中，可以通过此开关开启及关闭指定 MCP 服务器
+![MCP 开关按钮](assets/mcp_switch.png)

docs/tutorial/zh_CN/agent/friday.md

@@ -42,7 +42,7 @@ packages/
 - [x] 支持元工具（meta tool）
 - [x] 支持实时操控/中断
 - [x] 支持状态/会话管理
+- [x] 支持长期记忆
+- [x] 支持用户动态添加 MCP 服务器
 - [ ] 🚧 支持规划和规划可视化
-- [ ] 🚧 支持长期记忆
 - [ ] 支持 Anthropic 智能体技能
-- [ ] 支持用户动态添加 MCP 服务器

docs/tutorial/zh_CN/agent/think.md

@@ -0,0 +1,53 @@
+# 思考模式
+
+Friday 支持开启模型思考模式，并将模型思考内容进行展示。
+
+![思考模式展示](assets/think_show.png)
+
+## 如何开启思考模式
+
+如果模型支持思考模式，Friday 通过在调用模型 API 时传入参数以开启模型思考模式。不同模型提供者所支持的参数如下：
+
+| 模型提供者                      | 参数名             | 可选值                                              | 说明                               |
+| ------------------------------- | ------------------ | --------------------------------------------------- | ---------------------------------- |
+| **OpenAI** (o1/o3)              | `reasoning_effort` | `low` / `medium` / `high`                           | 控制思考深度，默认 medium          |
+| **Anthropic** (Claude 3.7+)     | `thinking`         | `{"type": "enabled", "budget_tokens": 1024-128000}` | 扩展思考模式，需指定 token 预算    |
+| **Gemini** (3.0+)               | `thinking_config`  | `{"thinking_level": "low/medium/high"}`             | 控制思考级别，high 激活 Deep Think |
+| **Ollama** (DeepSeek R1/Qwen 3) | `think`            | `true` / `false`                                    | 开启/关闭思考模式                  |
+| **DashScope** (通义千问)        | `enable_thinking`  | `true` / `false`                                    | 仅支持流式输出                     |
+
+## 配置示例
+
+所有思考模式相关参数都通过 **`generate_kwargs`** 传递：
+
+### DashScope (阿里云)
+
+![dashscope 参数配置](assets/dashscope.png)
+
+> **注意**: 非流式调用时必须设置 `enable_thinking=False`，否则会报错
+
+### OpenAI
+
+![openai 参数配置](assets/openai.png)
+
+### Anthropic (Claude)
+
+![anthropic 参数配置](assets/anthropic.png)
+
+### Google Gemini
+
+![gemini 参数配置](assets/gemini.png)
+
+> **注意**: Gemini 2.5 版本使用 `thinking_budget` 参数（整数值），Gemini 3+ 版本使用 `thinking_level`
+
+### Ollama
+
+![ollama 参数配置](assets/ollama.png)
+
+## 参数选择建议
+
+| 任务类型                                 | 推荐等级 | 说明                 |
+| ---------------------------------------- | -------- | -------------------- |
+| 简单任务（翻译、分类、数据提取）         | `low`    | 快速响应，低成本     |
+| 日常开发（代码生成、内容写作、调试）     | `medium` | 平衡性能与成本       |
+| 复杂推理（数学证明、科学分析、算法设计） | `high`   | 深度推理，高质量输出 |