diff --git a/gradle.properties b/gradle.properties
index e69de29..80e5580 100644
--- a/gradle.properties
+++ b/gradle.properties
@@ -0,0 +1,2 @@
+systemProp.org.gradle.internal.http.connectionTimeout=60000
+systemProp.org.gradle.internal.http.socketTimeout=120000
diff --git a/gradle/wrapper/gradle-wrapper.properties b/gradle/wrapper/gradle-wrapper.properties
index b82aa23..e2b3e26 100644
--- a/gradle/wrapper/gradle-wrapper.properties
+++ b/gradle/wrapper/gradle-wrapper.properties
@@ -1,7 +1,7 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
 distributionUrl=https\://services.gradle.org/distributions/gradle-8.7-bin.zip
-networkTimeout=10000
+networkTimeout=60000
 validateDistributionUrl=true
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists
diff --git a/src/main/java/com/ddaodan/MineChatGPT/CommandHandler.java b/src/main/java/com/ddaodan/MineChatGPT/CommandHandler.java
index ff6bd07..079535c 100644
--- a/src/main/java/com/ddaodan/MineChatGPT/CommandHandler.java
+++ b/src/main/java/com/ddaodan/MineChatGPT/CommandHandler.java
@@ -2,22 +2,20 @@ package com.ddaodan.MineChatGPT;
 
 import com.ddaodan.MineChatGPT.service.ApiService;
 import com.ddaodan.MineChatGPT.service.CommandService;
+import com.ddaodan.MineChatGPT.service.RequestCoordinator;
 import com.ddaodan.MineChatGPT.service.UserSessionManager;
 import org.bukkit.command.Command;
 import org.bukkit.command.CommandExecutor;
 import org.bukkit.command.CommandSender;
 public class CommandHandler implements CommandExecutor {
-    private final Main plugin;
     private final ConfigManager configManager;
     private final CommandService commandService;
     private final UserSessionManager sessionManager;
 
-    public CommandHandler(Main plugin, ConfigManager configManager) {
-        this.plugin = plugin;
+    public CommandHandler(ConfigManager configManager, UserSessionManager sessionManager, RequestCoordinator requestCoordinator) {
         this.configManager = configManager;
-        this.sessionManager = new UserSessionManager(configManager);
-        ApiService apiService = new ApiService(configManager);
-        this.commandService = new CommandService(configManager, apiService, sessionManager);
+        this.sessionManager = sessionManager;
+        this.commandService = new CommandService(configManager, requestCoordinator, sessionManager);
     }
 
     @Override
@@ -43,10 +41,12 @@ public class CommandHandler implements CommandExecutor {
                 return commandService.handleClearCommand(sender, userId);
             } else if (subCommand.equalsIgnoreCase("character")) {
                 return commandService.handleCharacterCommand(sender, args, userId);
+            } else if (subCommand.equalsIgnoreCase("stats")) {
+                return commandService.handleStatsCommand(sender, args, userId);
             } else {
                 return commandService.handleAskCommand(sender, args, userId);
             }
         }
         return false;
     }
-}
\ No newline at end of file
+}
diff --git a/src/main/java/com/ddaodan/MineChatGPT/ConfigManager.java b/src/main/java/com/ddaodan/MineChatGPT/ConfigManager.java
index aa3e8ab..c130243 100644
--- a/src/main/java/com/ddaodan/MineChatGPT/ConfigManager.java
+++ b/src/main/java/com/ddaodan/MineChatGPT/ConfigManager.java
@@ -1,17 +1,23 @@
 package com.ddaodan.MineChatGPT;
 
+import com.ddaodan.MineChatGPT.util.ConfigFileUpdater;
 import org.bukkit.ChatColor;
+import org.bukkit.configuration.ConfigurationSection;
 import org.bukkit.configuration.file.FileConfiguration;
 
-import java.util.HashMap;
+import java.util.ArrayList;
+import java.util.Collections;
+import java.util.LinkedHashMap;
 import java.util.List;
 import java.util.Map;
+import java.util.concurrent.atomic.AtomicInteger;
 
 public class ConfigManager {
     private final Main plugin;
     private FileConfiguration config;
     private String currentModel;
     private LanguageManager languageManager;
+    private volatile Map<String, String> charactersCache;
 
     public ConfigManager(Main plugin) {
         this.plugin = plugin;
@@ -19,14 +25,21 @@ public class ConfigManager {
         // 获取语言设置
         String language = config.getString("language", "en");
         this.languageManager = new LanguageManager(plugin, language);
+        this.charactersCache = null;
     }
     public boolean isDebugMode() {
         return config.getBoolean("debug", false);
     }
     public void reloadConfig() {
+        ConfigFileUpdater.UpdateResult updateResult = ConfigFileUpdater.updateIfMissingKeys(plugin, "config.yml");
+        if (updateResult.updated) {
+            plugin.getLogger().info("Config updated: inserted " + updateResult.insertedPaths
+                    + " missing path(s). Backup: " + updateResult.backupFileName);
+        }
         plugin.reloadConfig();
         config = plugin.getConfig();
         currentModel = config.getString("default_model");
+        charactersCache = null;
         
         // 重新加载语言文件
         if (languageManager != null) {
@@ -44,34 +57,56 @@ public class ConfigManager {
     public void setCurrentModel(String model) {
         currentModel = model;
     }
-    private int currentKeyIndex = 0;
+    private final AtomicInteger currentKeyIndex = new AtomicInteger(0);
     
     public String getApiKey() {
-        List<String> keys = config.getStringList("api.keys");
+        List<String> keys = new ArrayList<>();
+        for (String key : config.getStringList("api.keys")) {
+            if (isUsableApiKey(key)) {
+                keys.add(key);
+            }
+        }
         if (keys.isEmpty()) {
-            // 向后兼容：如果没有找到keys列表，尝试使用旧的单一key配置
             String legacyKey = config.getString("api.key");
-            if (legacyKey != null && !legacyKey.isEmpty()) {
+            if (isUsableApiKey(legacyKey)) {
                 return legacyKey;
             }
             return "";
         }
-        
+
         String selectionMethod = config.getString("api.selection_method", "round_robin");
-        
+
         if ("random".equalsIgnoreCase(selectionMethod)) {
-            // 随机选择一个key
             int randomIndex = (int) (Math.random() * keys.size());
             return keys.get(randomIndex);
         } else {
-            // 默认使用轮询方式
-            String key = keys.get(currentKeyIndex);
-            currentKeyIndex = (currentKeyIndex + 1) % keys.size();
-            return key;
+            int index = Math.floorMod(currentKeyIndex.getAndIncrement(), keys.size());
+            return keys.get(index);
         }
     }
+
+    private boolean isUsableApiKey(String key) {
+        if (key == null) {
+            return false;
+        }
+        String trimmed = key.trim();
+        if (trimmed.isEmpty()) {
+            return false;
+        }
+        return !trimmed.startsWith("sk-your_openai_api_key_");
+    }
+
     public String getBaseUrl() {
-        return config.getString("api.base_url");
+        return config.getString("api.base_url", "https://api.openai.com/v1");
+    }
+    public int getApiConnectTimeoutMs() {
+        return config.getInt("api.connect_timeout_ms", 10000);
+    }
+    public int getApiTimeoutMs() {
+        return config.getInt("api.timeout_ms", 30000);
+    }
+    public int getApiThreadPoolSize() {
+        return Math.max(1, config.getInt("api.thread_pool_size", 4));
     }
     public String getDefaultModel() {
         return config.getString("default_model");
@@ -106,18 +141,51 @@ public class ConfigManager {
     public String getHelpCharacterMessage() {
         return languageManager.getMessage("help_character");
     }
+    public String getHelpStatsMessage() {
+        return languageManager.getMessage("help_stats");
+    }
     public String getModelSwitchMessage() {
         return languageManager.getMessage("model_switch");
     }
     public String getChatGPTErrorMessage() {
         return languageManager.getMessage("chatgpt_error");
     }
+    public String getNoApiKeyMessage() {
+        return languageManager.getMessage("no_api_key", "&cNo API key configured. Please set api.keys in config.yml.");
+    }
     public String getChatGPTResponseMessage() {
         return languageManager.getMessage("chatgpt_response", "&b%s: %s");
     }
     public String getQuestionMessage() {
         return languageManager.getMessage("question");
     }
+    public String getQueuedMessage() {
+        return languageManager.getMessage("queued", "&eYour request has been queued. Position: %s");
+    }
+    public String getQueueFullMessage() {
+        return languageManager.getMessage("queue_full", "&cQueue is full. Please try again later.");
+    }
+    public String getQueueFullUserMessage() {
+        return languageManager.getMessage("queue_full_user", "&cYou have too many pending requests. Please wait.");
+    }
+    public String getCooldownMessage() {
+        return languageManager.getMessage("cooldown", "&cYou're sending requests too fast. Please wait %s ms.");
+    }
+    public String getRateLimitedMessage() {
+        return languageManager.getMessage("rate_limited", "&cRate limited. Please try again later.");
+    }
+    public String getStatsHeaderMessage() {
+        return languageManager.getMessage("stats_header", "&e===== MineChatGPT Stats =====");
+    }
+    public String getStatsGlobalMessage() {
+        return languageManager.getMessage("stats_global", "&eGlobal tokens: %s (prompt=%s, completion=%s), requests=%s");
+    }
+    public String getStatsUserMessage() {
+        return languageManager.getMessage("stats_user", "&eYour tokens: %s (prompt=%s, completion=%s), requests=%s");
+    }
+    public String getStatsResetMessage() {
+        return languageManager.getMessage("stats_reset", "&aStats reset.");
+    }
     public String getInvalidModelMessage() {
         return languageManager.getMessage("invalid_model");
     }
@@ -133,6 +201,24 @@ public class ConfigManager {
     public int getMaxHistorySize() {
         return config.getInt("conversation.max_history_size", 10);
     }
+    public int getMaxContextTokens() {
+        return config.getInt("conversation.max_context_tokens", 2000);
+    }
+    public int getReserveCompletionTokens() {
+        return config.getInt("conversation.reserve_completion_tokens", 400);
+    }
+    public boolean isSummarizationEnabled() {
+        return config.getBoolean("conversation.summarization.enabled", true);
+    }
+    public String getSummarizationModel() {
+        return config.getString("conversation.summarization.model", getCurrentModel());
+    }
+    public int getSummarizationTriggerTokens() {
+        return config.getInt("conversation.summarization.trigger_tokens", 1800);
+    }
+    public int getSummarizationKeepLastMessages() {
+        return Math.max(0, config.getInt("conversation.summarization.keep_last_messages", 6));
+    }
     public boolean isContextEnabled() {
         return config.getBoolean("conversation.context_enabled", false);
     }
@@ -158,11 +244,27 @@ public class ConfigManager {
         return languageManager.getMessage("invalid_character", "&cInvalid character. Use /chatgpt character to list available characters.");
     }
     public Map<String, String> getCharacters() {
-        Map<String, String> characters = new HashMap<>();
-        config.getConfigurationSection("characters").getKeys(false).forEach(key -> {
-            characters.put(key, config.getString("characters." + key));
-        });
-        return characters;
+        Map<String, String> cached = charactersCache;
+        if (cached != null) {
+            return cached;
+        }
+
+        ConfigurationSection section = config.getConfigurationSection("characters");
+        if (section == null) {
+            charactersCache = Collections.emptyMap();
+            return charactersCache;
+        }
+
+        Map<String, String> characters = new LinkedHashMap<>();
+        for (String key : section.getKeys(false)) {
+            String prompt = section.getString(key);
+            if (prompt != null) {
+                characters.put(key, prompt);
+            }
+        }
+
+        charactersCache = Collections.unmodifiableMap(characters);
+        return charactersCache;
     }
     public String getCurrentCharacter(String userId) {
         return config.getString("users." + userId + ".character", "ChatGPT");
@@ -171,4 +273,62 @@ public class ConfigManager {
         config.set("users." + userId + ".character", character);
         plugin.saveConfig();
     }
-}
\ No newline at end of file
+
+    public boolean isRateLimitEnabled() {
+        return config.getBoolean("rate_limit.enabled", true);
+    }
+    public String getRateLimitMode() {
+        return config.getString("rate_limit.mode", "both");
+    }
+    public long getRateLimitCooldownMs() {
+        return Math.max(0L, config.getLong("rate_limit.cooldown_ms", 1000L));
+    }
+    public String getRateLimitTokenEstimator() {
+        return config.getString("rate_limit.token_estimator", "approx_chars");
+    }
+    public int getRateLimitAssumedCompletionTokens() {
+        return Math.max(0, config.getInt("rate_limit.assumed_completion_tokens", 300));
+    }
+
+    public int getPerUserRequestsPerMinute() {
+        return Math.max(0, config.getInt("rate_limit.per_user.requests_per_minute", 6));
+    }
+    public int getPerUserBurstRequests() {
+        return Math.max(0, config.getInt("rate_limit.per_user.burst_requests", 3));
+    }
+    public int getPerUserTokensPerMinute() {
+        return Math.max(0, config.getInt("rate_limit.per_user.tokens_per_minute", 4000));
+    }
+    public int getPerUserBurstTokens() {
+        return Math.max(0, config.getInt("rate_limit.per_user.burst_tokens", 2000));
+    }
+
+    public int getGlobalRequestsPerMinute() {
+        return Math.max(0, config.getInt("rate_limit.global.requests_per_minute", 60));
+    }
+    public int getGlobalBurstRequests() {
+        return Math.max(0, config.getInt("rate_limit.global.burst_requests", 20));
+    }
+    public int getGlobalTokensPerMinute() {
+        return Math.max(0, config.getInt("rate_limit.global.tokens_per_minute", 100000));
+    }
+    public int getGlobalBurstTokens() {
+        return Math.max(0, config.getInt("rate_limit.global.burst_tokens", 20000));
+    }
+
+    public boolean isQueueEnabled() {
+        return config.getBoolean("queue.enabled", true);
+    }
+    public int getQueueMaxSize() {
+        return Math.max(0, config.getInt("queue.max_size", 100));
+    }
+    public int getQueueMaxPerUser() {
+        return Math.max(0, config.getInt("queue.max_per_user", 3));
+    }
+    public int getQueueMaxInFlight() {
+        return Math.max(1, config.getInt("queue.max_in_flight", getApiThreadPoolSize()));
+    }
+    public int getQueueDispatchPerTick() {
+        return Math.max(1, config.getInt("queue.dispatch_per_tick", 2));
+    }
+}
diff --git a/src/main/java/com/ddaodan/MineChatGPT/ConversationContext.java b/src/main/java/com/ddaodan/MineChatGPT/ConversationContext.java
index 1cdb283..c1972f9 100644
--- a/src/main/java/com/ddaodan/MineChatGPT/ConversationContext.java
+++ b/src/main/java/com/ddaodan/MineChatGPT/ConversationContext.java
@@ -1,29 +1,84 @@
 package com.ddaodan.MineChatGPT;
 
-import java.util.LinkedList;
-import java.util.Queue;
+import java.util.ArrayDeque;
+import java.util.ArrayList;
+import java.util.Deque;
+import java.util.List;
 
 public class ConversationContext {
-    private Queue<String> conversationHistory;
-    private int maxHistorySize;
+    public static final class Message {
+        private final String role;
+        private final String content;
+
+        public Message(String role, String content) {
+            this.role = role;
+            this.content = content;
+        }
+
+        public String getRole() {
+            return role;
+        }
+
+        public String getContent() {
+            return content;
+        }
+    }
+
+    private final Deque<Message> history;
+    private final int maxHistorySize;
 
     public ConversationContext(int maxHistorySize) {
         this.maxHistorySize = maxHistorySize;
-        this.conversationHistory = new LinkedList<>();
+        this.history = new ArrayDeque<>();
     }
 
-    public void addMessage(String message) {
-        if (conversationHistory.size() >= maxHistorySize) {
-            conversationHistory.poll();
+    public void addUserMessage(String message) {
+        addMessage("user", message);
+    }
+
+    public void addAssistantMessage(String message) {
+        addMessage("assistant", message);
+    }
+
+    private void addMessage(String role, String message) {
+        if (history.size() >= maxHistorySize) {
+            history.pollFirst();
         }
-        conversationHistory.offer(message);
+        history.addLast(new Message(role, message));
     }
 
-    public String getConversationHistory() {
-        return String.join("\n", conversationHistory);
+    public List<Message> getMessages() {
+        return new ArrayList<>(history);
+    }
+
+    public void setMessages(List<Message> messages) {
+        history.clear();
+        if (messages == null || messages.isEmpty()) {
+            return;
+        }
+        int startIndex = Math.max(0, messages.size() - maxHistorySize);
+        for (int i = startIndex; i < messages.size(); i++) {
+            Message msg = messages.get(i);
+            if (msg != null) {
+                history.addLast(new Message(msg.getRole(), msg.getContent()));
+            }
+        }
+    }
+
+    public void trimToLast(int keepLast) {
+        if (keepLast < 0) {
+            keepLast = 0;
+        }
+        while (history.size() > keepLast) {
+            history.pollFirst();
+        }
+    }
+
+    public void removeOldestMessage() {
+        history.pollFirst();
     }
 
     public void clearHistory() {
-        conversationHistory.clear();
+        history.clear();
     }
-}
\ No newline at end of file
+}
diff --git a/src/main/java/com/ddaodan/MineChatGPT/LanguageManager.java b/src/main/java/com/ddaodan/MineChatGPT/LanguageManager.java
index 6bae836..4d1e6dd 100644
--- a/src/main/java/com/ddaodan/MineChatGPT/LanguageManager.java
+++ b/src/main/java/com/ddaodan/MineChatGPT/LanguageManager.java
@@ -1,5 +1,6 @@
 package com.ddaodan.MineChatGPT;
 
+import com.ddaodan.MineChatGPT.util.ConfigFileUpdater;
 import org.bukkit.ChatColor;
 import org.bukkit.configuration.file.FileConfiguration;
 import org.bukkit.configuration.file.YamlConfiguration;
@@ -23,22 +24,29 @@ public class LanguageManager {
     }
 
     public void loadLanguage() {
+        String resourcePath = "lang/" + currentLanguage + ".yml";
         langFile = new File(plugin.getDataFolder(), "lang" + File.separator + currentLanguage + ".yml");
-        
-        // 如果语言文件不存在，创建默认语言文件
-        if (!langFile.exists()) {
-            langFile.getParentFile().mkdirs();
-            plugin.saveResource("lang/" + currentLanguage + ".yml", false);
+
+        if (plugin.getResource(resourcePath) != null) {
+            ConfigFileUpdater.UpdateResult updateResult = ConfigFileUpdater.updateIfMissingKeys(plugin, resourcePath);
+            if (updateResult.updated) {
+                plugin.getLogger().info("Language file updated (" + currentLanguage + "): inserted "
+                        + updateResult.insertedPaths + " missing path(s). Backup: " + updateResult.backupFileName);
+            }
+        } else if (!langFile.exists()) {
+            plugin.getLogger().warning("Language resource not found: " + resourcePath + ", file does not exist.");
         }
-        
+
         langConfig = YamlConfiguration.loadConfiguration(langFile);
-        
-        // 设置默认值，以防语言文件中缺少某些键
-        InputStream defaultLangStream = plugin.getResource("lang/" + currentLanguage + ".yml");
+
+        InputStream defaultLangStream = plugin.getResource(resourcePath);
         if (defaultLangStream != null) {
-            YamlConfiguration defaultLang = YamlConfiguration.loadConfiguration(
-                    new InputStreamReader(defaultLangStream, StandardCharsets.UTF_8));
-            langConfig.setDefaults(defaultLang);
+            try (InputStreamReader reader = new InputStreamReader(defaultLangStream, StandardCharsets.UTF_8)) {
+                YamlConfiguration defaultLang = YamlConfiguration.loadConfiguration(reader);
+                langConfig.setDefaults(defaultLang);
+            } catch (IOException ex) {
+                plugin.getLogger().warning("Failed to load default language resource: " + ex.getMessage());
+            }
         }
     }
 
@@ -62,4 +70,4 @@ public class LanguageManager {
     public String getMessage(String path, String defaultValue) {
         return translateColorCodes(langConfig.getString("messages." + path, defaultValue));
     }
-}
\ No newline at end of file
+}
diff --git a/src/main/java/com/ddaodan/MineChatGPT/Main.java b/src/main/java/com/ddaodan/MineChatGPT/Main.java
index 185bc86..1306138 100644
--- a/src/main/java/com/ddaodan/MineChatGPT/Main.java
+++ b/src/main/java/com/ddaodan/MineChatGPT/Main.java
@@ -7,18 +7,21 @@ import java.util.Objects;
 
 public final class Main extends JavaPlugin {
     private ConfigManager configManager;
+    private com.ddaodan.MineChatGPT.service.UserSessionManager sessionManager;
+    private com.ddaodan.MineChatGPT.service.ApiService apiService;
+    private com.ddaodan.MineChatGPT.service.RequestCoordinator requestCoordinator;
     private CommandHandler commandHandler;
     private MineChatGPTTabCompleter tabCompleter;
-    private LanguageManager languageManager;
 
     @Override
     public void onEnable() {
         saveDefaultConfig();
         configManager = new ConfigManager(this);
-        // 初始化语言管理器
-        String language = getConfig().getString("language", "en");
-        languageManager = new LanguageManager(this, language);
-        commandHandler = new CommandHandler(this, configManager);
+        sessionManager = new com.ddaodan.MineChatGPT.service.UserSessionManager(configManager);
+        apiService = new com.ddaodan.MineChatGPT.service.ApiService(this, configManager);
+        requestCoordinator = new com.ddaodan.MineChatGPT.service.RequestCoordinator(this, configManager, apiService, sessionManager);
+        requestCoordinator.start();
+        commandHandler = new CommandHandler(configManager, sessionManager, requestCoordinator);
         tabCompleter = new MineChatGPTTabCompleter(configManager);
         Objects.requireNonNull(getCommand("chatgpt")).setExecutor(commandHandler);
         Objects.requireNonNull(getCommand("chatgpt")).setTabCompleter(tabCompleter);
@@ -32,6 +35,12 @@ public final class Main extends JavaPlugin {
 
     @Override
     public void onDisable() {
+        if (requestCoordinator != null) {
+            requestCoordinator.stop();
+        }
+        if (apiService != null) {
+            apiService.shutdown();
+        }
         saveConfig();
     }
-}
\ No newline at end of file
+}
diff --git a/src/main/java/com/ddaodan/MineChatGPT/MineChatGPTTabCompleter.java b/src/main/java/com/ddaodan/MineChatGPT/MineChatGPTTabCompleter.java
index f3c875b..35b6549 100644
--- a/src/main/java/com/ddaodan/MineChatGPT/MineChatGPTTabCompleter.java
+++ b/src/main/java/com/ddaodan/MineChatGPT/MineChatGPTTabCompleter.java
@@ -26,12 +26,15 @@ public class MineChatGPTTabCompleter implements TabCompleter {
             completions.add("context");
             completions.add("clear");
             completions.add("character");
+            completions.add("stats");
         } else if (args.length == 2) {
             String subCommand = args[0];
             if (subCommand.equalsIgnoreCase("model")) {
                 completions.addAll(configManager.getModels());
             } else if (subCommand.equalsIgnoreCase("character")) {
                 completions.addAll(configManager.getCharacters().keySet());
+            } else if (subCommand.equalsIgnoreCase("stats")) {
+                completions.add("reset");
             }
         }
 
diff --git a/src/main/java/com/ddaodan/MineChatGPT/service/ApiService.java b/src/main/java/com/ddaodan/MineChatGPT/service/ApiService.java
index 34612f5..e65b3b4 100644
--- a/src/main/java/com/ddaodan/MineChatGPT/service/ApiService.java
+++ b/src/main/java/com/ddaodan/MineChatGPT/service/ApiService.java
@@ -1,155 +1,201 @@
 package com.ddaodan.MineChatGPT.service;
 
 import com.ddaodan.MineChatGPT.ConfigManager;
-import com.ddaodan.MineChatGPT.ConversationContext;
-import org.bukkit.command.CommandSender;
+import com.ddaodan.MineChatGPT.Main;
+import jodd.http.HttpRequest;
+import jodd.http.HttpResponse;
 import org.json.JSONArray;
 import org.json.JSONObject;
 
-import java.nio.charset.StandardCharsets;
 import java.util.concurrent.CompletableFuture;
+import java.util.concurrent.ExecutorService;
+import java.util.concurrent.Executors;
+import java.util.concurrent.ThreadFactory;
+import java.util.concurrent.atomic.AtomicInteger;
 import java.util.logging.Level;
 import java.util.logging.Logger;
 
-import jodd.http.HttpRequest;
-import jodd.http.HttpResponse;
-
 /**
- * API服务类，负责处理与OpenAI API的通信
+ * API服务类：负责与 OpenAI 兼容接口通信（/chat/completions）
  */
 public class ApiService {
-    private final ConfigManager configManager;
     private static final Logger logger = Logger.getLogger(ApiService.class.getName());
 
-    public ApiService(ConfigManager configManager) {
+    private final Main plugin;
+    private final ConfigManager configManager;
+    private final ExecutorService executor;
+
+    public ApiService(Main plugin, ConfigManager configManager) {
+        this.plugin = plugin;
         this.configManager = configManager;
+        this.executor = Executors.newFixedThreadPool(
+                configManager.getApiThreadPoolSize(),
+                newNamedThreadFactory("minechatgpt-api-")
+        );
     }
 
-    /**
-     * 向ChatGPT发送请求
-     *
-     * @param sender 命令发送者
-     * @param question 问题内容
-     * @param conversationContext 对话上下文
-     * @param contextEnabled 是否启用上下文
-     * @param userId 用户ID
-     */
-    public void askChatGPT(CommandSender sender, String question, ConversationContext conversationContext, boolean contextEnabled, String userId) {
-        String utf8Question = convertToUTF8(question);
-        JSONObject json = new JSONObject();
-        json.put("model", configManager.getCurrentModel());
-        JSONArray messages = new JSONArray();
-        
-        // 添加自定义 prompt
-        String currentCharacter = configManager.getCurrentCharacter(userId);
-        String customPrompt = configManager.getCharacters().get(currentCharacter);
-        if (customPrompt != null && !customPrompt.isEmpty()) {
-            JSONObject promptMessage = new JSONObject();
-            promptMessage.put("role", "system");
-            promptMessage.put("content", customPrompt);
-            messages.put(promptMessage);
+    public void shutdown() {
+        executor.shutdownNow();
+    }
+
+    public CompletableFuture<ChatCompletionResult> createChatCompletion(String model, JSONArray messages) {
+        if (model == null || model.trim().isEmpty()) {
+            return CompletableFuture.completedFuture(ChatCompletionResult.error("Missing model", null));
         }
-        
-        JSONObject message = new JSONObject();
-        message.put("role", "user");
-        message.put("content", utf8Question);
-        messages.put(message);
-        
-        if (contextEnabled) {
-            String history = conversationContext.getConversationHistory();
-            if (!history.isEmpty()) {
-                JSONObject historyMessage = new JSONObject();
-                historyMessage.put("role", "system");
-                historyMessage.put("content", history);
-                messages.put(historyMessage);
-            }
-        }
-        
-        json.put("messages", messages);
-        
-        if (configManager.isDebugMode()) {
-            logger.info("Built request: " + json.toString());
+        if (messages == null) {
+            messages = new JSONArray();
         }
 
-        HttpRequest request = HttpRequest.post(configManager.getBaseUrl() + "/chat/completions")
+        String apiKey = configManager.getApiKey();
+        if (apiKey == null || apiKey.trim().isEmpty()) {
+            return CompletableFuture.completedFuture(ChatCompletionResult.error(configManager.getNoApiKeyMessage(), null));
+        }
+
+        JSONObject payload = new JSONObject();
+        payload.put("model", model);
+        payload.put("messages", messages);
+
+        if (configManager.isDebugMode()) {
+            logger.info("Built request: " + payload);
+        }
+
+        String baseUrl = normalizeBaseUrl(configManager.getBaseUrl());
+        HttpRequest request = HttpRequest
+                .post(baseUrl + "/chat/completions")
                 .header("Content-Type", "application/json; charset=UTF-8")
-                .header("Authorization", "Bearer " + configManager.getApiKey())
-                .bodyText(json.toString());
+                .header("Accept", "application/json")
+                .header("Authorization", "Bearer " + apiKey)
+                .header("User-Agent", "MineChatGPT/" + plugin.getDescription().getVersion())
+                .connectionTimeout(configManager.getApiConnectTimeoutMs())
+                .timeout(configManager.getApiTimeoutMs())
+                .bodyText(payload.toString());
 
         if (configManager.isDebugMode()) {
-            logger.info("Sending request to ChatGPT: " + request.toString());
+            logger.info("Sending request: " + request);
         }
 
-        CompletableFuture.supplyAsync(() -> request.send())
-                .thenAccept(response -> {
-                    if (configManager.isDebugMode()) {
-                        logger.info("Received response from ChatGPT: " + response.toString());
-                    }
-                    if (response.statusCode() == 200) {
-                        processSuccessResponse(response, sender, conversationContext, contextEnabled, currentCharacter);
-                    } else {
-                        processErrorResponse(response, sender);
-                    }
-                })
+        return CompletableFuture
+                .supplyAsync(request::send, executor)
+                .thenApply(response -> parseResponse(response))
                 .exceptionally(e -> {
                     logger.log(Level.SEVERE, "Exception occurred while processing request: " + e.getMessage(), e);
-                    sender.sendMessage(configManager.getChatGPTErrorMessage());
-                    return null;
+                    return ChatCompletionResult.error(configManager.getChatGPTErrorMessage(), null);
                 });
     }
 
-    /**
-     * 处理成功的API响应
-     *
-     * @param response HTTP响应
-     * @param sender 命令发送者
-     * @param conversationContext 对话上下文
-     * @param contextEnabled 是否启用上下文
-     * @param currentCharacter 当前角色
-     */
-    private void processSuccessResponse(HttpResponse response, CommandSender sender, 
-                                       ConversationContext conversationContext, 
-                                       boolean contextEnabled, String currentCharacter) {
-        String responseBody = response.bodyText();
-        String utf8ResponseBody = new String(responseBody.getBytes(StandardCharsets.ISO_8859_1), StandardCharsets.UTF_8);
-        try {
-            JSONObject jsonResponse = new JSONObject(utf8ResponseBody);
-            String answer = jsonResponse.getJSONArray("choices").getJSONObject(0).getJSONObject("message").getString("content");
-            sender.sendMessage(configManager.getChatGPTResponseMessage().replaceFirst("%s", currentCharacter).replaceFirst("%s", answer));
-            if (contextEnabled) {
-                conversationContext.addMessage(answer); // 仅在启用上下文时添加AI响应到历史记录
+    private ChatCompletionResult parseResponse(HttpResponse response) {
+        if (configManager.isDebugMode()) {
+            logger.info("Received response: " + response);
+        }
+
+        int status = response.statusCode();
+        String body = response.bodyText();
+
+        if (status == 200) {
+            try {
+                JSONObject json = new JSONObject(body);
+                String answer = json
+                        .getJSONArray("choices")
+                        .getJSONObject(0)
+                        .getJSONObject("message")
+                        .getString("content");
+
+                Usage usage = null;
+                if (json.has("usage")) {
+                    JSONObject usageJson = json.getJSONObject("usage");
+                    long prompt = usageJson.optLong("prompt_tokens", -1);
+                    long completion = usageJson.optLong("completion_tokens", -1);
+                    long total = usageJson.optLong("total_tokens", -1);
+                    if (prompt >= 0 && completion >= 0 && total >= 0) {
+                        usage = new Usage(prompt, completion, total);
+                    }
+                }
+
+                return ChatCompletionResult.success(answer, usage);
+            } catch (Exception e) {
+                logger.log(Level.SEVERE, "Failed to parse response: " + e.getMessage(), e);
+                return ChatCompletionResult.error(configManager.getChatGPTErrorMessage(), body);
             }
-        } catch (Exception e) {
-            logger.log(Level.SEVERE, "Failed to parse ChatGPT response: " + e.getMessage(), e);
-            sender.sendMessage(configManager.getChatGPTErrorMessage());
         }
+
+        String debugDetails = tryExtractOpenAiErrorDetails(status, body);
+        logger.log(Level.SEVERE, "Failed to get response (HTTP " + status + "): " + body);
+        return ChatCompletionResult.error(configManager.getChatGPTErrorMessage(), debugDetails);
     }
 
-    /**
-     * 处理错误的API响应
-     *
-     * @param response HTTP响应
-     * @param sender 命令发送者
-     */
-    private void processErrorResponse(HttpResponse response, CommandSender sender) {
-        String errorBody = response.bodyText();
-        logger.log(Level.SEVERE, "Failed to get a response from ChatGPT: " + errorBody);
-        sender.sendMessage(configManager.getChatGPTErrorMessage());
-    }
-
-    /**
-     * 将字符串转换为UTF-8编码
-     *
-     * @param input 输入字符串
-     * @return UTF-8编码的字符串
-     */
-    private String convertToUTF8(String input) {
+    private static String tryExtractOpenAiErrorDetails(int status, String responseBody) {
         try {
-            byte[] bytes = input.getBytes(StandardCharsets.UTF_8);
-            return new String(bytes, StandardCharsets.UTF_8);
-        } catch (Exception e) {
-            logger.severe("Failed to convert input to UTF-8: " + e.getMessage());
-            return input; // 如果转换失败，返回原始输入
+            JSONObject errorJson = new JSONObject(responseBody);
+            if (!errorJson.has("error")) {
+                return null;
+            }
+            JSONObject err = errorJson.getJSONObject("error");
+            String message = err.optString("message", "");
+            String code = err.optString("code", "");
+            if (message.isEmpty()) {
+                return null;
+            }
+            return "[MineChatGPT] HTTP " + status + (code.isEmpty() ? "" : " (" + code + ")") + ": " + message;
+        } catch (Exception ignored) {
+            return null;
         }
     }
-}
\ No newline at end of file
+
+    private static String normalizeBaseUrl(String baseUrl) {
+        if (baseUrl == null) {
+            return "https://api.openai.com/v1";
+        }
+        String trimmed = baseUrl.trim();
+        if (trimmed.endsWith("/")) {
+            return trimmed.substring(0, trimmed.length() - 1);
+        }
+        return trimmed;
+    }
+
+    private static ThreadFactory newNamedThreadFactory(String prefix) {
+        AtomicInteger counter = new AtomicInteger(1);
+        return runnable -> {
+            Thread thread = new Thread(runnable, prefix + counter.getAndIncrement());
+            thread.setDaemon(true);
+            return thread;
+        };
+    }
+
+    public static final class Usage {
+        public final long promptTokens;
+        public final long completionTokens;
+        public final long totalTokens;
+
+        public Usage(long promptTokens, long completionTokens, long totalTokens) {
+            this.promptTokens = promptTokens;
+            this.completionTokens = completionTokens;
+            this.totalTokens = totalTokens;
+        }
+    }
+
+    public static final class ChatCompletionResult {
+        public final String answer;
+        public final Usage usage;
+        public final String errorMessage;
+        public final String debugDetails;
+
+        private ChatCompletionResult(String answer, Usage usage, String errorMessage, String debugDetails) {
+            this.answer = answer;
+            this.usage = usage;
+            this.errorMessage = errorMessage;
+            this.debugDetails = debugDetails;
+        }
+
+        public static ChatCompletionResult success(String answer, Usage usage) {
+            return new ChatCompletionResult(answer, usage, null, null);
+        }
+
+        public static ChatCompletionResult error(String errorMessage, String debugDetails) {
+            return new ChatCompletionResult(null, null, errorMessage, debugDetails);
+        }
+
+        public boolean isSuccess() {
+            return errorMessage == null;
+        }
+    }
+}
diff --git a/src/main/java/com/ddaodan/MineChatGPT/service/ApproxCharTokenEstimator.java b/src/main/java/com/ddaodan/MineChatGPT/service/ApproxCharTokenEstimator.java
new file mode 100644
index 0000000..782d509
--- /dev/null
+++ b/src/main/java/com/ddaodan/MineChatGPT/service/ApproxCharTokenEstimator.java
@@ -0,0 +1,13 @@
+package com.ddaodan.MineChatGPT.service;
+
+public class ApproxCharTokenEstimator implements TokenEstimator {
+    @Override
+    public int estimateTextTokens(String text) {
+        if (text == null || text.isEmpty()) {
+            return 0;
+        }
+        int chars = text.length();
+        return (int) Math.ceil(chars / 4.0);
+    }
+}
+
diff --git a/src/main/java/com/ddaodan/MineChatGPT/service/CommandService.java b/src/main/java/com/ddaodan/MineChatGPT/service/CommandService.java
index 08f7cda..ba344c7 100644
--- a/src/main/java/com/ddaodan/MineChatGPT/service/CommandService.java
+++ b/src/main/java/com/ddaodan/MineChatGPT/service/CommandService.java
@@ -12,12 +12,12 @@ import java.util.Map;
  */
 public class CommandService {
     private final ConfigManager configManager;
-    private final ApiService apiService;
+    private final RequestCoordinator requestCoordinator;
     private final UserSessionManager sessionManager;
 
-    public CommandService(ConfigManager configManager, ApiService apiService, UserSessionManager sessionManager) {
+    public CommandService(ConfigManager configManager, RequestCoordinator requestCoordinator, UserSessionManager sessionManager) {
         this.configManager = configManager;
-        this.apiService = apiService;
+        this.requestCoordinator = requestCoordinator;
         this.sessionManager = sessionManager;
     }
 
@@ -33,6 +33,7 @@ public class CommandService {
             return true;
         }
         configManager.reloadConfig();
+        requestCoordinator.reloadFromConfig();
         sender.sendMessage(configManager.getReloadMessage());
         return true;
     }
@@ -161,15 +162,40 @@ public class CommandService {
             return true;
         }
         String question = String.join(" ", args);
-        ConversationContext conversationContext = sessionManager.getConversationContext(userId);
-        boolean contextEnabled = sessionManager.isContextEnabled(userId);
-        
-        if (contextEnabled) {
-            conversationContext.addMessage(question);
-        }
         
         sender.sendMessage(configManager.getQuestionMessage().replace("%s", question));
-        apiService.askChatGPT(sender, question, conversationContext, contextEnabled, userId);
+        requestCoordinator.submitAsk(sender, question, userId);
+        return true;
+    }
+
+    public boolean handleStatsCommand(CommandSender sender, String[] args, String userId) {
+        if (!sender.hasPermission("minechatgpt.stats")) {
+            sender.sendMessage(configManager.getNoPermissionMessage().replace("%s", "minechatgpt.stats"));
+            return true;
+        }
+
+        if (args.length >= 2 && "reset".equalsIgnoreCase(args[1])) {
+            requestCoordinator.resetUsageTracker();
+            sender.sendMessage(configManager.getStatsResetMessage());
+            return true;
+        }
+
+        UsageTracker.Snapshot global = requestCoordinator.getUsageTracker().getGlobal();
+        UsageTracker.Snapshot user = requestCoordinator.getUsageTracker().getUser(userId);
+
+        sender.sendMessage(configManager.getStatsHeaderMessage());
+        sender.sendMessage(configManager.getStatsGlobalMessage()
+                .replaceFirst("%s", String.valueOf(global.totalTokens))
+                .replaceFirst("%s", String.valueOf(global.promptTokens))
+                .replaceFirst("%s", String.valueOf(global.completionTokens))
+                .replaceFirst("%s", String.valueOf(global.requests)));
+
+        sender.sendMessage(configManager.getStatsUserMessage()
+                .replaceFirst("%s", String.valueOf(user.totalTokens))
+                .replaceFirst("%s", String.valueOf(user.promptTokens))
+                .replaceFirst("%s", String.valueOf(user.completionTokens))
+                .replaceFirst("%s", String.valueOf(user.requests)));
+
         return true;
     }
 
@@ -187,5 +213,6 @@ public class CommandService {
         sender.sendMessage(configManager.getHelpContextMessage());
         sender.sendMessage(configManager.getHelpClearMessage());
         sender.sendMessage(configManager.getHelpCharacterMessage());
+        sender.sendMessage(configManager.getHelpStatsMessage());
     }
-}
\ No newline at end of file
+}
diff --git a/src/main/java/com/ddaodan/MineChatGPT/service/RateLimiter.java b/src/main/java/com/ddaodan/MineChatGPT/service/RateLimiter.java
new file mode 100644
index 0000000..91e5720
--- /dev/null
+++ b/src/main/java/com/ddaodan/MineChatGPT/service/RateLimiter.java
@@ -0,0 +1,110 @@
+package com.ddaodan.MineChatGPT.service;
+
+import com.ddaodan.MineChatGPT.ConfigManager;
+
+import java.util.Map;
+import java.util.concurrent.ConcurrentHashMap;
+
+public class RateLimiter {
+    public static final class Decision {
+        public final boolean allowed;
+        public final String reason;
+        public final long retryAfterMs;
+
+        private Decision(boolean allowed, String reason, long retryAfterMs) {
+            this.allowed = allowed;
+            this.reason = reason;
+            this.retryAfterMs = retryAfterMs;
+        }
+
+        public static Decision allow() {
+            return new Decision(true, null, 0);
+        }
+
+        public static Decision deny(String reason, long retryAfterMs) {
+            return new Decision(false, reason, retryAfterMs);
+        }
+    }
+
+    private static final class UserBuckets {
+        private final TokenBucket requests;
+        private final TokenBucket tokens;
+
+        private UserBuckets(TokenBucket requests, TokenBucket tokens) {
+            this.requests = requests;
+            this.tokens = tokens;
+        }
+    }
+
+    private final ConfigManager configManager;
+    private final Map<String, UserBuckets> perUser = new ConcurrentHashMap<>();
+    private final Map<String, Long> nextAllowedAt = new ConcurrentHashMap<>();
+
+    private volatile TokenBucket globalRequests;
+    private volatile TokenBucket globalTokens;
+
+    public RateLimiter(ConfigManager configManager) {
+        this.configManager = configManager;
+        long now = System.currentTimeMillis();
+        this.globalRequests = new TokenBucket(
+                configManager.getGlobalBurstRequests(),
+                configManager.getGlobalRequestsPerMinute() / 60.0,
+                now
+        );
+        this.globalTokens = new TokenBucket(
+                configManager.getGlobalBurstTokens(),
+                configManager.getGlobalTokensPerMinute() / 60.0,
+                now
+        );
+    }
+
+    public Decision tryAcquire(String userId, int estimatedTotalTokens, long nowMillis) {
+        if (!configManager.isRateLimitEnabled()) {
+            return Decision.allow();
+        }
+
+        long cooldownMs = configManager.getRateLimitCooldownMs();
+        Long next = nextAllowedAt.get(userId);
+        if (next != null && nowMillis < next) {
+            return Decision.deny("cooldown", next - nowMillis);
+        }
+
+        String mode = configManager.getRateLimitMode();
+        boolean limitRequests = "requests".equalsIgnoreCase(mode) || "both".equalsIgnoreCase(mode);
+        boolean limitTokens = "tokens".equalsIgnoreCase(mode) || "both".equalsIgnoreCase(mode);
+
+        UserBuckets buckets = perUser.computeIfAbsent(userId, id -> {
+            TokenBucket req = new TokenBucket(
+                    configManager.getPerUserBurstRequests(),
+                    configManager.getPerUserRequestsPerMinute() / 60.0,
+                    nowMillis
+            );
+            TokenBucket tok = new TokenBucket(
+                    configManager.getPerUserBurstTokens(),
+                    configManager.getPerUserTokensPerMinute() / 60.0,
+                    nowMillis
+            );
+            return new UserBuckets(req, tok);
+        });
+
+        if (limitRequests) {
+            if (!globalRequests.tryConsume(1.0, nowMillis) || !buckets.requests.tryConsume(1.0, nowMillis)) {
+                return Decision.deny("rate_limited", 0);
+            }
+        }
+
+        if (limitTokens) {
+            double cost = Math.max(0, estimatedTotalTokens);
+            if (!globalTokens.tryConsume(cost, nowMillis) || !buckets.tokens.tryConsume(cost, nowMillis)) {
+                return Decision.deny("rate_limited", 0);
+            }
+        }
+
+        if (cooldownMs > 0) {
+            nextAllowedAt.put(userId, nowMillis + cooldownMs);
+        }
+
+        return Decision.allow();
+    }
+}
+
diff --git a/src/main/java/com/ddaodan/MineChatGPT/service/RequestCoordinator.java b/src/main/java/com/ddaodan/MineChatGPT/service/RequestCoordinator.java
new file mode 100644
index 0000000..2f7e610
--- /dev/null
+++ b/src/main/java/com/ddaodan/MineChatGPT/service/RequestCoordinator.java
@@ -0,0 +1,404 @@
+package com.ddaodan.MineChatGPT.service;
+
+import com.ddaodan.MineChatGPT.ConfigManager;
+import com.ddaodan.MineChatGPT.ConversationContext;
+import com.ddaodan.MineChatGPT.Main;
+import org.bukkit.Bukkit;
+import org.bukkit.command.CommandSender;
+import org.bukkit.entity.Player;
+import org.json.JSONArray;
+import org.json.JSONObject;
+
+import java.util.ArrayList;
+import java.util.List;
+import java.util.concurrent.CompletableFuture;
+import java.util.concurrent.atomic.AtomicInteger;
+import java.util.logging.Level;
+
+public class RequestCoordinator {
+    private static final String SUMMARY_SYSTEM_PROMPT =
+            "You are a summarization assistant. Summarize the conversation for future context. " +
+            "Keep it concise, preserve important facts, user preferences, goals, and constraints. " +
+            "Return ONLY the updated summary text.";
+
+    private final Main plugin;
+    private final ConfigManager configManager;
+    private final ApiService apiService;
+    private final UserSessionManager sessionManager;
+    private volatile RateLimiter rateLimiter;
+    private final AtomicInteger inFlight = new AtomicInteger(0);
+
+    private volatile UsageTracker usageTracker;
+    private volatile TokenEstimator tokenEstimator;
+    private volatile RequestQueue<RequestJob> queue;
+    private volatile int dispatchTaskId = -1;
+
+    public RequestCoordinator(Main plugin, ConfigManager configManager, ApiService apiService, UserSessionManager sessionManager) {
+        this.plugin = plugin;
+        this.configManager = configManager;
+        this.apiService = apiService;
+        this.sessionManager = sessionManager;
+        this.usageTracker = new UsageTracker();
+        reloadFromConfig();
+    }
+
+    public void reloadFromConfig() {
+        this.tokenEstimator = createTokenEstimator(configManager.getRateLimitTokenEstimator());
+        this.queue = new RequestQueue<>(configManager.getQueueMaxSize(), configManager.getQueueMaxPerUser());
+        this.rateLimiter = new RateLimiter(configManager);
+    }
+
+    public UsageTracker getUsageTracker() {
+        return usageTracker;
+    }
+
+    public void resetUsageTracker() {
+        this.usageTracker = new UsageTracker();
+    }
+
+    public void start() {
+        if (dispatchTaskId != -1) {
+            return;
+        }
+        dispatchTaskId = Bukkit.getScheduler().runTaskTimer(plugin, () -> {
+            if (!configManager.isQueueEnabled()) {
+                return;
+            }
+            int maxInFlight = configManager.getQueueMaxInFlight();
+            int perTick = configManager.getQueueDispatchPerTick();
+            for (int i = 0; i < perTick; i++) {
+                if (inFlight.get() >= maxInFlight) {
+                    return;
+                }
+                RequestJob job = queue.poll();
+                if (job == null) {
+                    return;
+                }
+                dispatch(job);
+            }
+        }, 1L, 1L).getTaskId();
+    }
+
+    public void stop() {
+        if (dispatchTaskId != -1) {
+            Bukkit.getScheduler().cancelTask(dispatchTaskId);
+            dispatchTaskId = -1;
+        }
+    }
+
+    public void submitAsk(CommandSender sender, String question, String userId) {
+        boolean contextEnabled = sessionManager.isContextEnabled(userId);
+        ConversationContext context = sessionManager.getConversationContext(userId);
+
+        int estimatedTokensForLimit = estimateTokensForLimit(userId, question, contextEnabled, context);
+        long now = System.currentTimeMillis();
+        RateLimiter.Decision decision = rateLimiter.tryAcquire(userId, estimatedTokensForLimit, now);
+        if (!decision.allowed) {
+            if ("cooldown".equals(decision.reason)) {
+                sender.sendMessage(configManager.getCooldownMessage().replace("%s", String.valueOf(decision.retryAfterMs)));
+            } else {
+                sender.sendMessage(configManager.getRateLimitedMessage());
+            }
+            return;
+        }
+
+        RequestJob job = new RequestJob(sender, question, userId, contextEnabled, estimatedTokensForLimit);
+
+        if (!configManager.isQueueEnabled()) {
+            int maxInFlight = configManager.getQueueMaxInFlight();
+            if (inFlight.get() >= maxInFlight) {
+                sender.sendMessage(configManager.getQueueFullMessage());
+                return;
+            }
+            dispatch(job);
+            return;
+        }
+
+        RequestQueue.EnqueueResult result = queue.tryEnqueue(userId, job);
+        if (!result.accepted) {
+            if ("queue_full_user".equals(result.reason)) {
+                sender.sendMessage(configManager.getQueueFullUserMessage());
+            } else {
+                sender.sendMessage(configManager.getQueueFullMessage());
+            }
+            return;
+        }
+
+        sender.sendMessage(configManager.getQueuedMessage().replace("%s", String.valueOf(result.position)));
+    }
+
+    private void dispatch(RequestJob job) {
+        inFlight.incrementAndGet();
+        process(job)
+                .whenComplete((ignored, e) -> {
+                    if (e != null) {
+                        plugin.getLogger().log(Level.SEVERE, "Job failed: " + e.getMessage(), e);
+                    }
+                    inFlight.decrementAndGet();
+                });
+    }
+
+    private CompletableFuture<Void> process(RequestJob job) {
+        if (job.sender instanceof Player) {
+            Player player = (Player) job.sender;
+            if (!player.isOnline()) {
+                return CompletableFuture.completedFuture(null);
+            }
+        }
+
+        ConversationContext context = sessionManager.getConversationContext(job.userId);
+        String characterName = sessionManager.getCurrentCharacter(job.userId);
+        String characterPrompt = configManager.getCharacters().get(characterName);
+        String summary = sessionManager.getSummary(job.userId);
+
+        CompletableFuture<Void> maybeSummarize = ensureContextWithinBudget(job.userId, job.contextEnabled, characterPrompt, summary, context);
+
+        return maybeSummarize.thenCompose(v -> {
+            String updatedSummary = sessionManager.getSummary(job.userId);
+            JSONArray messages = buildMessages(job.contextEnabled, characterPrompt, updatedSummary, context, job.question);
+            String model = configManager.getCurrentModel();
+            return apiService.createChatCompletion(model, messages)
+                    .thenAccept(result -> handleCompletion(job, characterName, result));
+        });
+    }
+
+    private void handleCompletion(RequestJob job, String characterName, ApiService.ChatCompletionResult result) {
+        Bukkit.getScheduler().runTask(plugin, () -> {
+            if (!plugin.isEnabled()) {
+                return;
+            }
+            if (job.sender instanceof Player) {
+                Player player = (Player) job.sender;
+                if (!player.isOnline()) {
+                    return;
+                }
+            }
+
+            if (!result.isSuccess()) {
+                job.sender.sendMessage(result.errorMessage);
+                if (configManager.isDebugMode() && result.debugDetails != null && !result.debugDetails.isEmpty()) {
+                    job.sender.sendMessage(result.debugDetails);
+                }
+                return;
+            }
+
+            String formatted = formatTwoPlaceholders(configManager.getChatGPTResponseMessage(), characterName, result.answer);
+            sendMultiline(job.sender, formatted);
+
+            if (job.contextEnabled) {
+                ConversationContext ctx = sessionManager.getConversationContext(job.userId);
+                ctx.addUserMessage(job.question);
+                ctx.addAssistantMessage(result.answer);
+            }
+
+            long promptTokens = 0;
+            long completionTokens = 0;
+            long totalTokens = 0;
+            if (result.usage != null) {
+                promptTokens = result.usage.promptTokens;
+                completionTokens = result.usage.completionTokens;
+                totalTokens = result.usage.totalTokens;
+            } else {
+                // fallback estimation
+                int estimatedPrompt = estimatePromptTokens(job.userId, job.question, job.contextEnabled);
+                int estimatedCompletion = tokenEstimator.estimateTextTokens(result.answer);
+                promptTokens = Math.max(0, estimatedPrompt);
+                completionTokens = Math.max(0, estimatedCompletion);
+                totalTokens = promptTokens + completionTokens;
+            }
+            usageTracker.record(job.userId, promptTokens, completionTokens, totalTokens);
+        });
+    }
+
+    private CompletableFuture<Void> ensureContextWithinBudget(String userId, boolean contextEnabled, String characterPrompt, String summary, ConversationContext context) {
+        if (!contextEnabled) {
+            return CompletableFuture.completedFuture(null);
+        }
+
+        if (!configManager.isSummarizationEnabled()) {
+            trimContextToBudget(characterPrompt, summary, context);
+            return CompletableFuture.completedFuture(null);
+        }
+
+        int trigger = configManager.getSummarizationTriggerTokens();
+        int estimated = estimateContextTokens(characterPrompt, summary, context);
+        if (estimated <= trigger) {
+            trimContextToBudget(characterPrompt, summary, context);
+            return CompletableFuture.completedFuture(null);
+        }
+
+        int keepLast = configManager.getSummarizationKeepLastMessages();
+        List<ConversationContext.Message> history = context.getMessages();
+        if (history.size() <= keepLast) {
+            trimContextToBudget(characterPrompt, summary, context);
+            return CompletableFuture.completedFuture(null);
+        }
+
+        List<ConversationContext.Message> toSummarize = history.subList(0, history.size() - keepLast);
+        List<ConversationContext.Message> toKeep = history.subList(history.size() - keepLast, history.size());
+
+        JSONArray summarizeMessages = new JSONArray();
+        summarizeMessages.put(new JSONObject().put("role", "system").put("content", SUMMARY_SYSTEM_PROMPT));
+
+        StringBuilder sb = new StringBuilder();
+        String existing = sessionManager.getSummary(userId);
+        if (existing != null && !existing.trim().isEmpty()) {
+            sb.append("Current summary:\n").append(existing.trim()).append("\n\n");
+        }
+        sb.append("Conversation to summarize:\n");
+        for (ConversationContext.Message msg : toSummarize) {
+            sb.append(msg.getRole()).append(": ").append(msg.getContent()).append("\n");
+        }
+        summarizeMessages.put(new JSONObject().put("role", "user").put("content", sb.toString()));
+
+        String model = configManager.getSummarizationModel();
+        return apiService.createChatCompletion(model, summarizeMessages)
+                .thenCompose(result -> runSync(() -> {
+                    if (!result.isSuccess()) {
+                        return;
+                    }
+                    String newSummary = result.answer == null ? "" : result.answer.trim();
+                    sessionManager.setSummary(userId, newSummary);
+                    context.setMessages(new ArrayList<>(toKeep));
+                    trimContextToBudget(characterPrompt, newSummary, context);
+
+                    if (result.usage != null) {
+                        usageTracker.record(userId, result.usage.promptTokens, result.usage.completionTokens, result.usage.totalTokens);
+                    }
+                }));
+    }
+
+    private CompletableFuture<Void> runSync(Runnable runnable) {
+        CompletableFuture<Void> future = new CompletableFuture<>();
+        Bukkit.getScheduler().runTask(plugin, () -> {
+            try {
+                runnable.run();
+                future.complete(null);
+            } catch (Exception e) {
+                future.completeExceptionally(e);
+            }
+        });
+        return future;
+    }
+
+    private void trimContextToBudget(String characterPrompt, String summary, ConversationContext context) {
+        int budget = configManager.getMaxContextTokens();
+        if (budget <= 0) {
+            return;
+        }
+        while (estimateContextTokens(characterPrompt, summary, context) > budget) {
+            List<ConversationContext.Message> history = context.getMessages();
+            if (history.isEmpty()) {
+                return;
+            }
+            context.removeOldestMessage();
+        }
+    }
+
+    private int estimateContextTokens(String characterPrompt, String summary, ConversationContext context) {
+        int total = 0;
+        if (characterPrompt != null && !characterPrompt.isEmpty()) {
+            total += tokenEstimator.estimateMessageTokens("system", characterPrompt);
+        }
+        if (summary != null && !summary.trim().isEmpty()) {
+            total += tokenEstimator.estimateMessageTokens("system", "Conversation summary:\n" + summary.trim());
+        }
+        for (ConversationContext.Message msg : context.getMessages()) {
+            total += tokenEstimator.estimateMessageTokens(msg.getRole(), msg.getContent());
+        }
+        return total;
+    }
+
+    private int estimatePromptTokens(String userId, String question, boolean contextEnabled) {
+        ConversationContext context = sessionManager.getConversationContext(userId);
+        String characterName = sessionManager.getCurrentCharacter(userId);
+        String characterPrompt = configManager.getCharacters().get(characterName);
+        String summary = sessionManager.getSummary(userId);
+
+        int total = 0;
+        if (characterPrompt != null && !characterPrompt.isEmpty()) {
+            total += tokenEstimator.estimateMessageTokens("system", characterPrompt);
+        }
+        if (contextEnabled) {
+            if (summary != null && !summary.trim().isEmpty()) {
+                total += tokenEstimator.estimateMessageTokens("system", "Conversation summary:\n" + summary.trim());
+            }
+            for (ConversationContext.Message msg : context.getMessages()) {
+                total += tokenEstimator.estimateMessageTokens(msg.getRole(), msg.getContent());
+            }
+        }
+        total += tokenEstimator.estimateMessageTokens("user", question);
+        return total;
+    }
+
+    private int estimateTokensForLimit(String userId, String question, boolean contextEnabled, ConversationContext context) {
+        int assumedCompletion = configManager.getRateLimitAssumedCompletionTokens();
+        int reserve = configManager.getReserveCompletionTokens();
+        int completionBudget = Math.max(assumedCompletion, reserve);
+        int prompt = estimatePromptTokens(userId, question, contextEnabled);
+        return prompt + completionBudget;
+    }
+
+    private JSONArray buildMessages(boolean contextEnabled, String characterPrompt, String summary, ConversationContext context, String question) {
+        JSONArray messages = new JSONArray();
+        if (characterPrompt != null && !characterPrompt.isEmpty()) {
+            messages.put(new JSONObject().put("role", "system").put("content", characterPrompt));
+        }
+        if (contextEnabled) {
+            if (summary != null && !summary.trim().isEmpty()) {
+                messages.put(new JSONObject().put("role", "system").put("content", "Conversation summary:\n" + summary.trim()));
+            }
+            for (ConversationContext.Message msg : context.getMessages()) {
+                messages.put(new JSONObject().put("role", msg.getRole()).put("content", msg.getContent()));
+            }
+        }
+        messages.put(new JSONObject().put("role", "user").put("content", question));
+        return messages;
+    }
+
+    private static void sendMultiline(CommandSender sender, String message) {
+        String[] lines = message.split("\\R", -1);
+        for (String line : lines) {
+            if (!line.isEmpty()) {
+                sender.sendMessage(line);
+            }
+        }
+    }
+
+    private static String formatTwoPlaceholders(String template, String first, String second) {
+        String withFirst = replaceFirstLiteral(template, "%s", first);
+        return replaceFirstLiteral(withFirst, "%s", second);
+    }
+
+    private static String replaceFirstLiteral(String template, String token, String replacement) {
+        int index = template.indexOf(token);
+        if (index < 0) {
+            return template;
+        }
+        return template.substring(0, index) + replacement + template.substring(index + token.length());
+    }
+
+    private static TokenEstimator createTokenEstimator(String type) {
+        if ("approx_chars".equalsIgnoreCase(type)) {
+            return new ApproxCharTokenEstimator();
+        }
+        return new ApproxCharTokenEstimator();
+    }
+
+    private static final class RequestJob {
+        private final CommandSender sender;
+        private final String question;
+        private final String userId;
+        private final boolean contextEnabled;
+        @SuppressWarnings("unused")
+        private final int estimatedTokensForLimit;
+
+        private RequestJob(CommandSender sender, String question, String userId, boolean contextEnabled, int estimatedTokensForLimit) {
+            this.sender = sender;
+            this.question = question;
+            this.userId = userId;
+            this.contextEnabled = contextEnabled;
+            this.estimatedTokensForLimit = estimatedTokensForLimit;
+        }
+    }
+}
diff --git a/src/main/java/com/ddaodan/MineChatGPT/service/RequestQueue.java b/src/main/java/com/ddaodan/MineChatGPT/service/RequestQueue.java
new file mode 100644
index 0000000..57eca68
--- /dev/null
+++ b/src/main/java/com/ddaodan/MineChatGPT/service/RequestQueue.java
@@ -0,0 +1,83 @@
+package com.ddaodan.MineChatGPT.service;
+
+import java.util.ArrayDeque;
+import java.util.Deque;
+import java.util.HashMap;
+import java.util.Map;
+
+public class RequestQueue<T> {
+    public static final class EnqueueResult {
+        public final boolean accepted;
+        public final int position;
+        public final String reason;
+
+        private EnqueueResult(boolean accepted, int position, String reason) {
+            this.accepted = accepted;
+            this.position = position;
+            this.reason = reason;
+        }
+
+        public static EnqueueResult accepted(int position) {
+            return new EnqueueResult(true, position, null);
+        }
+
+        public static EnqueueResult rejected(String reason) {
+            return new EnqueueResult(false, -1, reason);
+        }
+    }
+
+    private final int maxSize;
+    private final int maxPerUser;
+    private final Deque<Entry<T>> queue = new ArrayDeque<>();
+    private final Map<String, Integer> perUserCounts = new HashMap<>();
+
+    private static final class Entry<T> {
+        private final String userId;
+        private final T item;
+
+        private Entry(String userId, T item) {
+            this.userId = userId;
+            this.item = item;
+        }
+    }
+
+    public RequestQueue(int maxSize, int maxPerUser) {
+        this.maxSize = Math.max(0, maxSize);
+        this.maxPerUser = Math.max(0, maxPerUser);
+    }
+
+    public synchronized EnqueueResult tryEnqueue(String userId, T item) {
+        if (maxSize > 0 && queue.size() >= maxSize) {
+            return EnqueueResult.rejected("queue_full");
+        }
+        int count = perUserCounts.getOrDefault(userId, 0);
+        if (maxPerUser > 0 && count >= maxPerUser) {
+            return EnqueueResult.rejected("queue_full_user");
+        }
+        queue.addLast(new Entry<>(userId, item));
+        perUserCounts.put(userId, count + 1);
+        return EnqueueResult.accepted(queue.size());
+    }
+
+    public synchronized T poll() {
+        Entry<T> entry = queue.pollFirst();
+        if (entry == null) {
+            return null;
+        }
+        Integer count = perUserCounts.get(entry.userId);
+        if (count != null) {
+            int next = count - 1;
+            if (next <= 0) {
+                perUserCounts.remove(entry.userId);
+            } else {
+                perUserCounts.put(entry.userId, next);
+            }
+        }
+        return entry.item;
+    }
+
+    public synchronized int size() {
+        return queue.size();
+    }
+}
+
diff --git a/src/main/java/com/ddaodan/MineChatGPT/service/TokenBucket.java b/src/main/java/com/ddaodan/MineChatGPT/service/TokenBucket.java
new file mode 100644
index 0000000..0771e9a
--- /dev/null
+++ b/src/main/java/com/ddaodan/MineChatGPT/service/TokenBucket.java
@@ -0,0 +1,48 @@
+package com.ddaodan.MineChatGPT.service;
+
+public class TokenBucket {
+    private final double capacity;
+    private final double refillTokensPerMillis;
+    private double tokens;
+    private long lastRefillMillis;
+
+    public TokenBucket(double capacity, double refillTokensPerSecond, long nowMillis) {
+        this.capacity = Math.max(0.0, capacity);
+        this.refillTokensPerMillis = Math.max(0.0, refillTokensPerSecond) / 1000.0;
+        this.tokens = this.capacity;
+        this.lastRefillMillis = nowMillis;
+    }
+
+    public synchronized boolean tryConsume(double amount, long nowMillis) {
+        refill(nowMillis);
+        if (amount <= 0) {
+            return true;
+        }
+        if (tokens >= amount) {
+            tokens -= amount;
+            return true;
+        }
+        return false;
+    }
+
+    public synchronized void refund(double amount, long nowMillis) {
+        refill(nowMillis);
+        if (amount <= 0) {
+            return;
+        }
+        tokens = Math.min(capacity, tokens + amount);
+    }
+
+    private void refill(long nowMillis) {
+        if (nowMillis <= lastRefillMillis) {
+            return;
+        }
+        double elapsed = nowMillis - lastRefillMillis;
+        double add = elapsed * refillTokensPerMillis;
+        if (add > 0) {
+            tokens = Math.min(capacity, tokens + add);
+        }
+        lastRefillMillis = nowMillis;
+    }
+}
+
diff --git a/src/main/java/com/ddaodan/MineChatGPT/service/TokenEstimator.java b/src/main/java/com/ddaodan/MineChatGPT/service/TokenEstimator.java
new file mode 100644
index 0000000..9b3920a
--- /dev/null
+++ b/src/main/java/com/ddaodan/MineChatGPT/service/TokenEstimator.java
@@ -0,0 +1,11 @@
+package com.ddaodan.MineChatGPT.service;
+
+public interface TokenEstimator {
+    int estimateTextTokens(String text);
+
+    default int estimateMessageTokens(String role, String content) {
+        int base = 4; // message overhead (approx)
+        return base + estimateTextTokens(role) + estimateTextTokens(content);
+    }
+}
+
diff --git a/src/main/java/com/ddaodan/MineChatGPT/service/UsageTracker.java b/src/main/java/com/ddaodan/MineChatGPT/service/UsageTracker.java
new file mode 100644
index 0000000..e4b852b
--- /dev/null
+++ b/src/main/java/com/ddaodan/MineChatGPT/service/UsageTracker.java
@@ -0,0 +1,72 @@
+package com.ddaodan.MineChatGPT.service;
+
+import java.util.Map;
+import java.util.concurrent.ConcurrentHashMap;
+import java.util.concurrent.atomic.LongAdder;
+
+public class UsageTracker {
+    public static final class Snapshot {
+        public final long requests;
+        public final long promptTokens;
+        public final long completionTokens;
+        public final long totalTokens;
+
+        private Snapshot(long requests, long promptTokens, long completionTokens, long totalTokens) {
+            this.requests = requests;
+            this.promptTokens = promptTokens;
+            this.completionTokens = completionTokens;
+            this.totalTokens = totalTokens;
+        }
+    }
+
+    private static final class Counters {
+        private final LongAdder requests = new LongAdder();
+        private final LongAdder promptTokens = new LongAdder();
+        private final LongAdder completionTokens = new LongAdder();
+        private final LongAdder totalTokens = new LongAdder();
+
+        private Snapshot snapshot() {
+            return new Snapshot(
+                    requests.sum(),
+                    promptTokens.sum(),
+                    completionTokens.sum(),
+                    totalTokens.sum()
+            );
+        }
+    }
+
+    private final Counters global = new Counters();
+    private final Map<String, Counters> perUser = new ConcurrentHashMap<>();
+
+    public void record(String userId, long promptTokens, long completionTokens, long totalTokens) {
+        global.requests.increment();
+        global.promptTokens.add(promptTokens);
+        global.completionTokens.add(completionTokens);
+        global.totalTokens.add(totalTokens);
+
+        Counters user = perUser.computeIfAbsent(userId, k -> new Counters());
+        user.requests.increment();
+        user.promptTokens.add(promptTokens);
+        user.completionTokens.add(completionTokens);
+        user.totalTokens.add(totalTokens);
+    }
+
+    public Snapshot getGlobal() {
+        return global.snapshot();
+    }
+
+    public Snapshot getUser(String userId) {
+        Counters user = perUser.get(userId);
+        if (user == null) {
+            return new Snapshot(0, 0, 0, 0);
+        }
+        return user.snapshot();
+    }
+
+    public void reset() {
+        perUser.clear();
+        // global LongAdder has no reset without recreating
+        // Create a new UsageTracker if you need a hard reset
+    }
+}
+
diff --git a/src/main/java/com/ddaodan/MineChatGPT/service/UserSessionManager.java b/src/main/java/com/ddaodan/MineChatGPT/service/UserSessionManager.java
index b7653e6..d0e5dc5 100644
--- a/src/main/java/com/ddaodan/MineChatGPT/service/UserSessionManager.java
+++ b/src/main/java/com/ddaodan/MineChatGPT/service/UserSessionManager.java
@@ -14,12 +14,14 @@ public class UserSessionManager {
     private final Map<String, ConversationContext> userContexts;
     private final Map<String, Boolean> userContextEnabled;
     private final Map<String, String> userCurrentCharacter;
+    private final Map<String, String> userSummaries;
 
     public UserSessionManager(ConfigManager configManager) {
         this.configManager = configManager;
         this.userContexts = new HashMap<>();
         this.userContextEnabled = new HashMap<>();
         this.userCurrentCharacter = new HashMap<>();
+        this.userSummaries = new HashMap<>();
     }
 
     /**
@@ -67,6 +69,7 @@ public class UserSessionManager {
     public void clearConversationHistory(String userId) {
         ConversationContext context = getConversationContext(userId);
         context.clearHistory();
+        clearSummary(userId);
     }
 
     /**
@@ -92,4 +95,20 @@ public class UserSessionManager {
         userCurrentCharacter.put(userId, character);
         configManager.setCurrentCharacter(userId, character);
     }
-}
\ No newline at end of file
+
+    public String getSummary(String userId) {
+        return userSummaries.getOrDefault(userId, "");
+    }
+
+    public void setSummary(String userId, String summary) {
+        if (summary == null) {
+            userSummaries.remove(userId);
+            return;
+        }
+        userSummaries.put(userId, summary);
+    }
+
+    public void clearSummary(String userId) {
+        userSummaries.remove(userId);
+    }
+}
diff --git a/src/main/java/com/ddaodan/MineChatGPT/util/ConfigFileUpdater.java b/src/main/java/com/ddaodan/MineChatGPT/util/ConfigFileUpdater.java
new file mode 100644
index 0000000..ace100a
--- /dev/null
+++ b/src/main/java/com/ddaodan/MineChatGPT/util/ConfigFileUpdater.java
@@ -0,0 +1,424 @@
+package com.ddaodan.MineChatGPT.util;
+
+import org.bukkit.configuration.file.YamlConfiguration;
+import org.bukkit.plugin.java.JavaPlugin;
+
+import java.io.BufferedReader;
+import java.io.File;
+import java.io.IOException;
+import java.io.InputStream;
+import java.io.InputStreamReader;
+import java.io.Reader;
+import java.nio.charset.StandardCharsets;
+import java.nio.file.Files;
+import java.nio.file.StandardCopyOption;
+import java.text.SimpleDateFormat;
+import java.util.ArrayDeque;
+import java.util.ArrayList;
+import java.util.Collections;
+import java.util.Date;
+import java.util.LinkedHashMap;
+import java.util.List;
+import java.util.Locale;
+import java.util.Map;
+import java.util.Set;
+import java.util.TreeSet;
+
+public final class ConfigFileUpdater {
+    private ConfigFileUpdater() {
+    }
+
+    public static UpdateResult updateIfMissingKeys(JavaPlugin plugin, String resourcePath) {
+        File configFile = new File(plugin.getDataFolder(), resourcePath);
+        if (!configFile.exists()) {
+            plugin.saveResource(resourcePath, false);
+            return UpdateResult.noChange();
+        }
+
+        try {
+            YamlConfiguration currentYaml = YamlConfiguration.loadConfiguration(configFile);
+            YamlConfiguration defaultYaml = loadYamlFromResource(plugin, resourcePath);
+            if (defaultYaml == null) {
+                return UpdateResult.noChange();
+            }
+
+            List<String> defaultLines = readResourceLines(plugin, resourcePath);
+            if (defaultLines.isEmpty()) {
+                return UpdateResult.noChange();
+            }
+            DocumentModel defaultModel = parseDocument(defaultLines);
+
+            List<String> missingTopLevelPaths = collectMissingTopPaths(currentYaml, defaultYaml, defaultModel);
+            if (missingTopLevelPaths.isEmpty()) {
+                return UpdateResult.noChange();
+            }
+
+            List<String> currentLines = Files.readAllLines(configFile.toPath(), StandardCharsets.UTF_8);
+            int inserted = 0;
+            for (String missingPath : missingTopLevelPaths) {
+                Entry source = defaultModel.entriesByPath.get(missingPath);
+                if (source == null) {
+                    continue;
+                }
+
+                List<String> snippet = new ArrayList<>(
+                        defaultLines.subList(source.startLine, source.endLine + 1)
+                );
+                if (snippet.isEmpty()) {
+                    continue;
+                }
+
+                DocumentModel currentModel = parseDocument(currentLines);
+                int insertAt = findInsertIndex(currentLines, currentModel, defaultModel, missingPath);
+                if (insertAt < 0 || insertAt > currentLines.size()) {
+                    insertAt = currentLines.size();
+                }
+                insertSnippetWithSpacing(currentLines, insertAt, snippet);
+                inserted++;
+            }
+
+            if (inserted == 0) {
+                return UpdateResult.noChange();
+            }
+
+            File backup = createBackup(configFile);
+            Files.write(configFile.toPath(), currentLines, StandardCharsets.UTF_8);
+            return UpdateResult.updated(inserted, backup.getName());
+        } catch (Exception ex) {
+            plugin.getLogger().warning("Failed to auto-update config.yml: " + ex.getMessage());
+            return UpdateResult.noChange();
+        }
+    }
+
+    private static YamlConfiguration loadYamlFromResource(JavaPlugin plugin, String resourcePath) {
+        InputStream inputStream = plugin.getResource(resourcePath);
+        if (inputStream == null) {
+            return null;
+        }
+        try (Reader reader = new InputStreamReader(inputStream, StandardCharsets.UTF_8)) {
+            return YamlConfiguration.loadConfiguration(reader);
+        } catch (IOException ex) {
+            plugin.getLogger().warning("Failed to read default config resource: " + ex.getMessage());
+            return null;
+        }
+    }
+
+    private static List<String> collectMissingTopPaths(
+            YamlConfiguration currentYaml,
+            YamlConfiguration defaultYaml,
+            DocumentModel defaultModel
+    ) {
+        Set<String> missing = new TreeSet<String>();
+        for (String key : defaultYaml.getKeys(true)) {
+            if (!currentYaml.contains(key)) {
+                missing.add(key);
+            }
+        }
+        if (missing.isEmpty()) {
+            return Collections.emptyList();
+        }
+
+        List<String> top = new ArrayList<String>();
+        for (Entry entry : defaultModel.entriesInOrder) {
+            String path = entry.path;
+            if (!missing.contains(path)) {
+                continue;
+            }
+            if (!hasAncestor(top, path)) {
+                top.add(path);
+            }
+        }
+        return top;
+    }
+
+    private static boolean hasAncestor(List<String> selected, String path) {
+        int index = path.lastIndexOf('.');
+        while (index > 0) {
+            String parent = path.substring(0, index);
+            if (selected.contains(parent)) {
+                return true;
+            }
+            index = parent.lastIndexOf('.');
+        }
+        return false;
+    }
+
+    private static List<String> readResourceLines(JavaPlugin plugin, String resourcePath) throws IOException {
+        InputStream inputStream = plugin.getResource(resourcePath);
+        if (inputStream == null) {
+            return Collections.emptyList();
+        }
+        List<String> lines = new ArrayList<String>();
+        try (BufferedReader reader = new BufferedReader(new InputStreamReader(inputStream, StandardCharsets.UTF_8))) {
+            String line;
+            while ((line = reader.readLine()) != null) {
+                lines.add(line);
+            }
+        }
+        return lines;
+    }
+
+    private static int findInsertIndex(
+            List<String> currentLines,
+            DocumentModel currentModel,
+            DocumentModel defaultModel,
+            String path
+    ) {
+        String parentPath = getParentPath(path);
+        int beforeSibling = findNextExistingSiblingStartLine(currentModel, defaultModel, path, parentPath);
+        if (beforeSibling >= 0) {
+            return beforeSibling;
+        }
+
+        if (parentPath == null) {
+            Entry lastRoot = null;
+            for (Entry entry : currentModel.entriesInOrder) {
+                if (getParentPath(entry.path) == null) {
+                    lastRoot = entry;
+                }
+            }
+            return lastRoot == null ? currentLines.size() : lastRoot.endLine + 1;
+        }
+
+        Entry parent = currentModel.entriesByPath.get(parentPath);
+        if (parent == null) {
+            return currentLines.size();
+        }
+        return parent.endLine + 1;
+    }
+
+    private static int findNextExistingSiblingStartLine(
+            DocumentModel currentModel,
+            DocumentModel defaultModel,
+            String path,
+            String parentPath
+    ) {
+        boolean foundSelf = false;
+        for (Entry defaultEntry : defaultModel.entriesInOrder) {
+            if (!isDirectChildOf(defaultEntry.path, parentPath)) {
+                continue;
+            }
+
+            if (!foundSelf) {
+                if (defaultEntry.path.equals(path)) {
+                    foundSelf = true;
+                }
+                continue;
+            }
+
+            Entry existingSibling = currentModel.entriesByPath.get(defaultEntry.path);
+            if (existingSibling != null) {
+                return existingSibling.startLine;
+            }
+        }
+        return -1;
+    }
+
+    private static boolean isDirectChildOf(String path, String parentPath) {
+        String actualParent = getParentPath(path);
+        if (parentPath == null) {
+            return actualParent == null;
+        }
+        return parentPath.equals(actualParent);
+    }
+
+    private static String getParentPath(String path) {
+        int dot = path.lastIndexOf('.');
+        if (dot <= 0) {
+            return null;
+        }
+        return path.substring(0, dot);
+    }
+
+    private static void insertSnippetWithSpacing(List<String> lines, int insertAt, List<String> snippet) {
+        if (snippet.isEmpty()) {
+            return;
+        }
+
+        // Avoid carrying leading/trailing blank lines from the template snippet.
+        int from = 0;
+        int to = snippet.size() - 1;
+        while (from <= to && snippet.get(from).trim().isEmpty()) {
+            from++;
+        }
+        while (to >= from && snippet.get(to).trim().isEmpty()) {
+            to--;
+        }
+        if (from > to) {
+            return;
+        }
+
+        List<String> normalized = new ArrayList<String>(snippet.subList(from, to + 1));
+        int index = insertAt;
+        boolean isTopLevel = isTopLevelSnippet(normalized);
+        if (isTopLevel
+                && index > 0
+                && !lines.get(index - 1).trim().isEmpty()
+                && !normalized.get(0).trim().isEmpty()
+                && (index >= lines.size() || !lines.get(index).trim().isEmpty())) {
+            lines.add(index, "");
+            index++;
+        }
+        lines.addAll(index, normalized);
+    }
+
+    private static boolean isTopLevelSnippet(List<String> snippet) {
+        for (String line : snippet) {
+            KeyLine keyLine = parseKeyLine(line);
+            if (keyLine != null) {
+                return keyLine.indent == 0;
+            }
+        }
+        return false;
+    }
+
+    private static File createBackup(File configFile) throws IOException {
+        String suffix = new SimpleDateFormat("yyyyMMdd-HHmmss", Locale.ROOT).format(new Date());
+        File backup = new File(configFile.getParentFile(), configFile.getName() + ".bak." + suffix);
+        Files.copy(configFile.toPath(), backup.toPath(), StandardCopyOption.REPLACE_EXISTING);
+        return backup;
+    }
+
+    private static DocumentModel parseDocument(List<String> lines) {
+        List<Entry> ordered = new ArrayList<Entry>();
+        Map<String, Entry> byPath = new LinkedHashMap<String, Entry>();
+        ArrayDeque<Entry> stack = new ArrayDeque<Entry>();
+
+        for (int i = 0; i < lines.size(); i++) {
+            KeyLine keyLine = parseKeyLine(lines.get(i));
+            if (keyLine == null) {
+                continue;
+            }
+
+            while (!stack.isEmpty() && keyLine.indent <= stack.peek().indent) {
+                stack.pop();
+            }
+
+            String path = stack.isEmpty() ? keyLine.key : stack.peek().path + "." + keyLine.key;
+            Entry entry = new Entry(path, keyLine.indent, i);
+            ordered.add(entry);
+            byPath.put(path, entry);
+            stack.push(entry);
+        }
+
+        for (Entry entry : ordered) {
+            entry.startLine = findSnippetStartLine(lines, entry.line);
+            entry.endLine = findValueEndLine(lines, entry.line, entry.indent);
+        }
+
+        return new DocumentModel(ordered, byPath);
+    }
+
+    private static int findSnippetStartLine(List<String> lines, int keyLineIndex) {
+        int start = keyLineIndex;
+        for (int i = keyLineIndex - 1; i >= 0; i--) {
+            String trimmed = lines.get(i).trim();
+            if (trimmed.startsWith("#")) {
+                start = i;
+                continue;
+            }
+            break;
+        }
+        return start;
+    }
+
+    private static int findValueEndLine(List<String> lines, int keyLineIndex, int keyIndent) {
+        int end = keyLineIndex;
+        for (int i = keyLineIndex + 1; i < lines.size(); i++) {
+            String line = lines.get(i);
+            KeyLine keyLine = parseKeyLine(line);
+            if (keyLine != null && keyLine.indent <= keyIndent) {
+                break;
+            }
+
+            String trimmed = line.trim();
+            if (!trimmed.isEmpty() && !trimmed.startsWith("#")) {
+                end = i;
+            }
+        }
+        return end;
+    }
+
+    private static KeyLine parseKeyLine(String line) {
+        if (line == null) {
+            return null;
+        }
+
+        String trimmed = line.trim();
+        if (trimmed.isEmpty() || trimmed.startsWith("#") || trimmed.startsWith("-")) {
+            return null;
+        }
+
+        int colon = trimmed.indexOf(':');
+        if (colon <= 0) {
+            return null;
+        }
+
+        String key = trimmed.substring(0, colon).trim();
+        if (key.isEmpty()) {
+            return null;
+        }
+
+        int indent = 0;
+        while (indent < line.length() && line.charAt(indent) == ' ') {
+            indent++;
+        }
+        return new KeyLine(key, indent);
+    }
+
+    public static final class UpdateResult {
+        public final boolean updated;
+        public final int insertedPaths;
+        public final String backupFileName;
+
+        private UpdateResult(boolean updated, int insertedPaths, String backupFileName) {
+            this.updated = updated;
+            this.insertedPaths = insertedPaths;
+            this.backupFileName = backupFileName;
+        }
+
+        private static UpdateResult noChange() {
+            return new UpdateResult(false, 0, "");
+        }
+
+        private static UpdateResult updated(int insertedPaths, String backupFileName) {
+            return new UpdateResult(true, insertedPaths, backupFileName);
+        }
+    }
+
+    private static final class DocumentModel {
+        private final List<Entry> entriesInOrder;
+        private final Map<String, Entry> entriesByPath;
+
+        private DocumentModel(List<Entry> entriesInOrder, Map<String, Entry> entriesByPath) {
+            this.entriesInOrder = entriesInOrder;
+            this.entriesByPath = entriesByPath;
+        }
+    }
+
+    private static final class Entry {
+        private final String path;
+        private final int indent;
+        private final int line;
+        private int startLine;
+        private int endLine;
+
+        private Entry(String path, int indent, int line) {
+            this.path = path;
+            this.indent = indent;
+            this.line = line;
+            this.startLine = line;
+            this.endLine = line;
+        }
+    }
+
+    private static final class KeyLine {
+        private final String key;
+        private final int indent;
+
+        private KeyLine(String key, int indent) {
+            this.key = key;
+            this.indent = indent;
+        }
+    }
+}
diff --git a/src/main/resources/config.yml b/src/main/resources/config.yml
index 53992f2..fe57eea 100644
--- a/src/main/resources/config.yml
+++ b/src/main/resources/config.yml
@@ -4,7 +4,10 @@
 # ======================================================
 # Available languages: en, zh
 # 可用语言：en（英文）, zh（中文）
+# This only affects plugin prompts, not the model's answer language.
+# 该项仅影响插件提示语言，不影响模型回答语言。
 language: "en"
+
 # ======================================================
 # API Configuration
 # API 设置
@@ -14,32 +17,58 @@ api:
   # To obtain an API key, visit https://platform.openai.com/account/api-keys and create a new API key
   # 你的 OpenAI API key，用于身份验证
   # 获取 API key 的方法：访问 https://platform.openai.com/account/api-keys 并创建一个新的 API key
+  # Placeholder keys such as sk-your_openai_api_key_1 are treated as invalid.
+  # 占位 key（如 sk-your_openai_api_key_1）会被视为无效。
   keys:
     - "sk-your_openai_api_key_1"
     # You can add multiple API keys below
     # 可以添加多个API key
     # - "sk-your_openai_api_key_2"
     # - "sk-your_openai_api_key_3"
-  
+
   # API key selection method: "round_robin" or "random"
   # Round Robin: Use each API key in turn
   # Random: Randomly select an API key
   # API key 选择方法："round_robin"（轮询）或 "random"（随机）
   # 轮询：依次使用每个API key
   # 随机：随机选择一个API key
+  # Use round_robin for steadier distribution in production.
+  # 生产环境更推荐 round_robin，分配更均匀。
   selection_method: "round_robin"
-  
+
   # The base URL for the OpenAI API, used to construct requests
   # If you cannot access the official API, you can use a proxy service
   # OpenAI API 的基础 URL，用于构建请求
   # 如果你无法访问官方API，可以使用代理服务
+  # Keep the /v1 path in the URL.
+  # URL 保留 /v1 路径。
   base_url: "https://api.openai.com/v1"
+
+  # HTTP timeouts (milliseconds)
+  # HTTP 超时（毫秒）
+  # connect_timeout_ms controls connection phase, timeout_ms controls total request time.
+  # connect_timeout_ms 控制建连阶段，timeout_ms 控制整次请求耗时。
+  # Increase carefully on unstable networks to avoid slow failure feedback.
+  # 网络不稳定时可适当调大，但过大可能导致失败反馈变慢。
+  connect_timeout_ms: 10000
+  timeout_ms: 30000
+
+  # Thread pool size for API requests
+  # API 请求线程池大小
+  # Higher value increases concurrency but also upstream pressure.
+  # 值越大并发越高，但也会增加上游 API 压力。
+  # Tune together with queue.max_in_flight.
+  # 建议与 queue.max_in_flight 配合调优。
+  thread_pool_size: 4
+
 # ======================================================
 # Model Configuration
 # 模型设置
 # ======================================================
 # List of supported models
 # 支持的模型列表
+# This is a selectable whitelist; actual availability depends on your provider.
+# 这里只是可选白名单，具体是否可用取决于服务商支持情况。
 models:
   # OpenAI ChatGPT
   - "gpt-4"
@@ -58,7 +87,10 @@ models:
 
 # The default model to use
 # 默认使用的模型
+# Use a cost-effective model as default for daily usage.
+# 日常建议选择更稳定且成本更低的模型作为默认模型。
 default_model: "gpt-4o-mini"
+
 # ======================================================
 # Conversation Settings
 # 对话设置
@@ -68,13 +100,141 @@ conversation:
   # When enabled, the plugin will remember the conversation history
   # 连续对话开关
   # 启用时，插件将记住对话历史
+  # Enabling improves continuity but increases token usage.
+  # 开启后连贯性更好，但 token 消耗会增加。
   context_enabled: false
-  
+
   # Maximum number of historical records retained
   # Increasing this value will consume more memory
   # 最大历史记录保留数量
   # 增加此值将消耗更多内存
+  # Usually 8~20 is a practical range.
+  # 通常 8~20 是较实用的范围。
   max_history_size: 10
+
+  # Token budget for context (approximate)
+  # 上下文 token 预算（近似）
+  # Exceeding this budget may trigger trimming or summarization.
+  # 超过预算会触发上下文裁剪或总结。
+  max_context_tokens: 2000
+
+  # Reserve tokens for the model output (approximate)
+  # 为模型输出预留 token（近似）
+  # Too low may truncate replies; too high reduces usable context.
+  # 值太小可能截断回复，值太大则压缩可用上下文。
+  reserve_completion_tokens: 400
+
+  # Automatic summarization to keep context within budget
+  # 自动总结以控制上下文长度
+  summarization:
+    # If disabled, long sessions are more likely to hit context limits.
+    # 关闭后长会话更容易触发上下文超限。
+    enabled: true
+    # Model used for summarization
+    # 用于总结的模型
+    # Prefer a faster/cheaper model for summarization.
+    # 建议用更快/更便宜的模型做总结。
+    model: "gpt-4o-mini"
+    # When estimated context tokens exceed this value, summarization may run
+    # 当估算的上下文 tokens 超过该值时可能触发总结
+    # Usually set slightly lower than max_context_tokens (e.g. 100~300 lower).
+    # 通常设为略低于 max_context_tokens（如低 100~300）。
+    trigger_tokens: 1800
+    # Keep the last N messages (user/assistant) after summarization
+    # 总结后保留最近 N 条消息（user/assistant）
+    # Higher value keeps more detail but uses more tokens.
+    # 值越大保留细节越多，但占用也更高。
+    keep_last_messages: 6
+
+# ======================================================
+# Rate Limit / Queue Settings
+# 限流 / 排队 设置
+# ======================================================
+rate_limit:
+  # Master switch for cooldown/token-bucket limiting.
+  # 冷却与令牌桶限流总开关。
+  enabled: true
+  # Mode: "requests" | "tokens" | "both"
+  # 模式："requests" | "tokens" | "both"
+  # requests = limit by request count
+  # tokens = limit by token budget
+  # both = enforce both limits (recommended)
+  # requests = 仅按请求次数
+  # tokens = 仅按 token 预算
+  # both = 两者同时生效（推荐）
+  mode: "both"
+
+  # Cooldown per user after accepting a request (milliseconds)
+  # 每个用户请求冷却（毫秒）
+  # Helps reduce short burst command spam from one player.
+  # 用于抑制单玩家短时间刷命令。
+  cooldown_ms: 1000
+
+  # Token estimation method: "approx_chars"
+  # token 估算方法："approx_chars"
+  # This estimator is approximate by design.
+  # 该估算器是近似估算。
+  token_estimator: "approx_chars"
+
+  # If token-mode is enabled, assume this many completion tokens per request for limiting
+  # 如果启用 token 限流，按该值假设每次回复的 completion tokens（用于限流）
+  # Increase this if users often ask for long outputs.
+  # 用户经常要求长回复时可适当提高该值。
+  assumed_completion_tokens: 300
+
+  per_user:
+    # Per-user request refill rate (requests/min).
+    # 单用户请求额度恢复速率（每分钟）。
+    requests_per_minute: 6
+    # Per-user request burst capacity.
+    # 单用户请求突发容量。
+    burst_requests: 3
+    # Per-user token refill rate (tokens/min).
+    # 单用户 token 额度恢复速率（每分钟）。
+    tokens_per_minute: 4000
+    # Per-user token burst capacity.
+    # 单用户 token 突发容量。
+    burst_tokens: 2000
+
+  global:
+    # Global request refill rate (requests/min).
+    # 全局请求额度恢复速率（每分钟）。
+    requests_per_minute: 60
+    # Global request burst capacity.
+    # 全局请求突发容量。
+    burst_requests: 20
+    # Global token refill rate (tokens/min).
+    # 全局 token 额度恢复速率（每分钟）。
+    tokens_per_minute: 100000
+    # Global token burst capacity.
+    # 全局 token 突发容量。
+    burst_tokens: 20000
+
+queue:
+  # If enabled, requests can wait in queue instead of immediate rejection.
+  # 开启后，请求可进入队列等待，而不是直接拒绝。
+  enabled: true
+  # Maximum queued jobs globally
+  # 全局队列最大长度
+  # New jobs are rejected after this limit.
+  # 达到上限后新请求会被拒绝。
+  max_size: 100
+  # Maximum queued jobs per user
+  # 每个用户队列最大长度
+  # Prevent one player from occupying all queue slots.
+  # 防止单玩家占满队列。
+  max_per_user: 3
+  # Maximum in-flight (concurrent) API requests
+  # 同时进行中的 API 请求上限
+  # Usually set <= api.thread_pool_size.
+  # 通常建议不高于 api.thread_pool_size。
+  max_in_flight: 4
+  # How many jobs to dispatch per tick (Spigot main thread)
+  # 每 tick 尝试派发多少个任务（主线程）
+  # Higher value drains queue faster but increases main-thread scheduling pressure.
+  # 值越大清队越快，但主线程调度压力更高。
+  dispatch_per_tick: 2
+
 # ======================================================
 # Character Settings
 # 角色设置
@@ -86,10 +246,12 @@ characters:
   # 格式：
   # 角色名称: "角色提示词"
   # 提示词将作为系统消息发送给AI
-  
+  # Keep prompts clear and concise to avoid unnecessary token overhead.
+  # 建议提示词简洁明确，避免无谓 token 开销。
+
   # Default character
   ChatGPT: "You are a helpful assistant."
-  
+
   # You can add more characters below
   # 你可以在下方添加更多角色
   # Example:
diff --git a/src/main/resources/lang/en.yml b/src/main/resources/lang/en.yml
index a390cf7..fd79054 100644
--- a/src/main/resources/lang/en.yml
+++ b/src/main/resources/lang/en.yml
@@ -11,17 +11,28 @@ messages:
   help_context: "&e/chatgpt context - Toggle context mode."
   help_clear: "&e/chatgpt clear - Clear conversation history."
   help_character: "&e/chatgpt character [character_name] - List or switch to a character."
+  help_stats: "&e/chatgpt stats - Show token usage statistics."
   context_toggle: "&eContext is now %s."
   context_toggle_enabled: "&aenabled"
   context_toggle_disabled: "&cdisabled"
   current_model_info: "&eCurrent model: %s. Use /chatgpt model <model_name> to switch models."
   model_switch: "&aModel switched to %s"
   chatgpt_error: "&cFailed to contact ChatGPT."
+  no_api_key: "&cNo API key configured. Please set api.keys in config.yml."
   chatgpt_response: "&b%s: %s"
   question: "&bYou: %s"
+  queued: "&eYour request has been queued. Position: %s"
+  queue_full: "&cQueue is full. Please try again later."
+  queue_full_user: "&cYou have too many pending requests. Please wait."
+  cooldown: "&cYou're sending requests too fast. Please wait %s ms."
+  rate_limited: "&cRate limited. Please try again later."
+  stats_header: "&e===== MineChatGPT Stats ====="
+  stats_global: "&eGlobal tokens: %s (prompt=%s, completion=%s), requests=%s"
+  stats_user: "&eYour tokens: %s (prompt=%s, completion=%s), requests=%s"
+  stats_reset: "&aStats reset."
   character_switched: "&aSwitched to character: %s"
   available_characters: "&eAvailable characters:"
   invalid_character: "&cInvalid character. Use /chatgpt character to list available characters."
   invalid_model: "&cInvalid model. Use /chatgpt modellist to see available models."
   available_models: "&eAvailable models:"
-  no_permission: "&cYou do not have permission to use this command. Required permission: %s"
\ No newline at end of file
+  no_permission: "&cYou do not have permission to use this command. Required permission: %s"
diff --git a/src/main/resources/lang/zh.yml b/src/main/resources/lang/zh.yml
index 0df8c19..24a894b 100644
--- a/src/main/resources/lang/zh.yml
+++ b/src/main/resources/lang/zh.yml
@@ -11,17 +11,28 @@ messages:
   help_context: "&e/chatgpt context - 切换连续对话模式"
   help_clear: "&e/chatgpt clear - 清空对话历史"
   help_character: "&e/chatgpt character [character_name] - 列出或切换角色"
+  help_stats: "&e/chatgpt stats - 查看 token 消耗统计"
   context_toggle: "&e连续对话模式已%s。"
   context_toggle_enabled: "&a开启"
   context_toggle_disabled: "&c关闭"
   current_model_info: "&e当前模型：%s，输入 /chatgpt model <model_name> 来切换模型。"
   model_switch: "&a已切换至模型 %s"
   chatgpt_error: "&c无法联系ChatGPT。"
+  no_api_key: "&c未配置 API Key，请在 config.yml 的 api.keys 中设置。"
   chatgpt_response: "&b%s: %s"
   question: "&b你: %s"
+  queued: "&e你的请求已进入队列，当前位置：%s"
+  queue_full: "&c队列已满，请稍后再试。"
+  queue_full_user: "&c你有太多等待中的请求，请先等待。"
+  cooldown: "&c请求过快，请等待 %s 毫秒。"
+  rate_limited: "&c触发限流，请稍后再试。"
+  stats_header: "&e===== MineChatGPT 统计 ====="
+  stats_global: "&e全局 tokens：%s（prompt=%s，completion=%s），请求数=%s"
+  stats_user: "&e你的 tokens：%s（prompt=%s，completion=%s），请求数=%s"
+  stats_reset: "&a统计已重置。"
   character_switched: "&a已切换至角色： %s"
   available_characters: "&e可用的角色列表："
   invalid_character: "&c无效的角色。使用 /chatgpt character 查看所有可用的角色。"
   invalid_model: "&c模型无效。使用 /chatgpt modellist 查看可用模型。"
   available_models: "&e可用模型列表："
-  no_permission: "&c你没有权限使用这个指令。需要的权限：%s"
\ No newline at end of file
+  no_permission: "&c你没有权限使用这个指令。需要的权限：%s"
diff --git a/src/main/resources/plugin.yml b/src/main/resources/plugin.yml
index b6137c3..13a3d72 100644
--- a/src/main/resources/plugin.yml
+++ b/src/main/resources/plugin.yml
@@ -30,4 +30,7 @@ permissions:
     default: true
   minechatgpt.character:
     description: Allows switching characters
-    default: true
\ No newline at end of file
+    default: true
+  minechatgpt.stats:
+    description: Allows viewing/resetting usage statistics
+    default: op