add: context window size usage description

frontend/public/tutorial-en.md

@@ -19,6 +19,7 @@ This tutorial will walk you through how to use a **Graph** structure to organize
   - [Entry Node](#entry-nodes)
 - [5. Build Review and Revise Loop](#5-build-review-and-revise-loop)
   - [Creating a Second Editor Node](#creating-a-second-editing-node)
+  - [Set the Context Window](#set-the-context-window)
   - [Human Node: Introducing Human Participation](#human-node-introducing-human-in-the-Loop)
 - [More Details](#more-things)
 ---
@@ -171,6 +172,28 @@ You are an editor skilled at integrating and polishing an article.
 Based on the article and the revision suggestions, refine and revise the article. Output only the revised article.
 ```
 
+---
+### Set the Context Window
+
+To give the Agent contextual memory, set this on **Editor 2**:
+
+* `Context Window Size = 7`
+
+This keeps the newest 7 messages (including node input and output messages).
+
+#### What is a Context Window?
+
+In DevAll, the **context window** is a **node-level retention policy**. After each run, the node tries to clean up its own input queue of Messages, keeping only what satisfies the retention rules to control context size. This does not affect the current run; it only affects inputs visible in later runs.
+
+**Rules** (configured via `Context Window Size` on the node):
+* `0`: clear all context, keeping only messages marked by `Keep Message Input = True` on edges.
+* `-1`: keep all messages.
+* `> 0`: keep the newest N messages (kept messages still count toward the limit).
+* When `Context Window Size != 0`, the system automatically saves the node's output messages into that node's context for later runs.
+
+**Tips**:
+* For long-lived context, use `Keep Message Input` in Edge config or the Memory module.
+
 ---
 
 ### Human Node: Introducing Human-in-the-Loop
@@ -237,7 +260,7 @@ We provide a set of examples: files named `demo_*.yaml` are feature/module demos
 
 # Node Types Explained
 
-MAC provides various node types, each with specific purposes and configuration options.
+DevAll provides various node types, each with specific purposes and configuration options.
 
 ## Agent Node
 
@@ -252,7 +275,7 @@ Agent Nodes are the core node type, used to call large language models (LLMs) fo
 Agent Nodes can be configured with tools, allowing the model to call external APIs or execute functions. Click Advanced Settings to see the options. Multiple tools can be configured, such as both MCP and Function tools, or multiple MCP tools.
 ![agent_tooling_config.png](media/agent_toolig_config.png)
 
-MAC supports two types of tools:
+DevAll supports two types of tools:
 
 #### 1. Function Tooling (Local Functions)
 
@@ -617,7 +640,7 @@ If none are met, repeat from Step 2.
 
 # Workspace Structure
 
-MAC uses a layered workspace structure to manage files:
+DevAll uses a layered workspace structure to manage files:
 
 ## Directory Structure
 

frontend/public/tutorial-zh.md

@@ -18,6 +18,7 @@
   - [初始节点](#初始节点)
 - [5. 审核与二次编辑流程](#5-build-review-and-revise-loop)
   - [创建二次编辑节点](#创建二次编辑节点)
+  - [设置 Context Window](#设置-context-window)
   - [Human Node：引入人工参与](#human-node引入人工参与)
 - [更多细节](#更多细节)
 
@@ -167,6 +168,28 @@ Graph 本身不承担具体任务的执行，而是负责对各节点的运行
 请根据输入的文章与修改建议，对文章进行润色和修改，直接输出修改后的文章。
 ```
 
+---
+### 设置 Context Window
+
+为了让 Agent 具备上下文记忆，请在 **Editor 2** 节点上设置：
+
+* `Context Window Size = 7`
+
+这表示 Editor 2 执行后会保留最近 7 条消息（包括节点输入和输出消息）。
+
+#### 什么是 Context Window（上下文窗口）？
+
+在 DevAll 中，**Context Window** 是**节点级别的上下文保留策略**。节点每次执行结束后，会尝试对自身输入队列中的 Messages 进行清理，仅保留满足保留规则的消息，以控制上下文规模。这不会影响本次执行，仅影响后续执行可见的输入。
+
+**规则**（对应节点配置里的 `Context Window Size`）：
+* `0`：清空全部上下文，仅保留通过边设置了 `Keep Message Input` 为 `True` 的消息。
+* `-1`：不清理，保留全部消息。
+* `> 0`：保留最新的 N 条消息（保留消息也会占用配额）。
+* 当 `Context Window Size != 0` 时，系统会自动把节点的输出消息写入该节点的上下文，供后续执行使用。
+
+**建议**：
+* 需要长期保留的关键上下文，搭配边上的 `Keep Message Input` 或 Memory 模块管理。
+
 ---
 ### Human Node：引入人工参与
 
@@ -227,7 +250,7 @@ Graph 本身不承担具体任务的执行，而是负责对各节点的运行
 
 # 节点类型详解
 
-MAC 提供多种节点类型，每种都有特定的用途和配置选项。
+DevAll 提供多种节点类型，每种都有特定的用途和配置选项。
 
 ## Agent 节点
 
@@ -242,7 +265,7 @@ Agent 节点是最核心的节点类型，用于调用大语言模型 (LLM) 完
 Agent 节点可以配置工具，让模型调用外部 API 或执行函数。请点击展开 Advanced Settings，即可看到配置项。Tooling 可配置多项，例如可同时配置 MCP 和 Function 工具；可配置多个 MCP 工具等。
 ![agent_tooling_config.png](media/agent_toolig_config.png)
 
-MAC 支持两类工具：
+DevAll 支持两类工具：
 
 #### 1. Function Tooling（本地函数）
 
@@ -610,7 +633,7 @@ def my_processor(data: str, _context: Dict[str, Any]) -> str:
 
 # Workspace 工作区
 
-MAC 使用分层的工作区结构管理文件：
+DevAll 使用分层的工作区结构管理文件：
 
 ## 目录结构