diff --git a/cmd/homed/main.go b/cmd/homed/main.go
index fdb5ee4..8049cb1 100644
--- a/cmd/homed/main.go
+++ b/cmd/homed/main.go
@@ -81,6 +81,7 @@ func main() {
 	}
 
 	memIdx := memory.NewIndexer(memDB)
+	memIdx.Sync() // 启动时立即同步，避免前30分钟空窗
 	socialStore := social.New(memDB)
 
 	distiller := pipeline.NewDistiller(memDB, *dataDir, pipeline.DistillerConfig{
@@ -194,14 +195,14 @@ func main() {
 						if err := textMem.Append(te); err != nil {
 							log.Printf("[homed] text memory append: %v", err)
 						}
-					}
+				}
 
-					if input != "" {
-						distiller.Append("agent", "user", input)
-					}
-					if response != "" {
-						distiller.Append("agent", "assistant", response)
-					}
+				if input != "" && memDB != nil {
+					distiller.Append("agent", "user", input)
+				}
+				if response != "" && memDB != nil {
+					distiller.Append("agent", "assistant", response)
+				}
 				}
 			}
 		}
@@ -323,20 +324,16 @@ func main() {
 如需异步发送消息或通知，使用 output_send 指定通道和内容。
 使用 output_list_channels 查看可用通道及其能力。
 
-你有以下核心工具:
-1. memory_recall — 查询图记忆（历史/个人信息）
-2. memory_commit — 写入图记忆（记住新信息）
-3. memory_introspect — 查看记忆统计
-4. person_query — 查询人物特质与社交关系网
-5. person_set_trait — 记录人物特质（性格/喜好等）
-6. person_relate — 建立人物间社交关系
-7. person_network — 查看社交网络
-8. knowledge_search — 搜索知识库
-9. doc_query — 查询文档记忆
-10. doc_commit — 写入文档记忆
-11. plgreload — 热重载插件
-12. llm_list_sources — 列出所有可用的 LLM 源
-13. llm_set_source — 切换到指定 LLM 源
+可用工具列表会由系统自动传入，按需使用即可。以下是你尤其需要关注的几类工具：
+- memory_* — 图记忆（长期记忆，记录和查询个人信息/事实）
+- knowledge_* — 知识库（查阅预设知识文档）
+- doc_* — 文档记忆（近期对话的存档，查询后自动清除）
+- person_* — 人物特质与社交关系网
+- llm_* — LLM 源管理（列出/切换模型提供商）
+- output_* — 输出通道管理（切换/发送消息）
+- timer_set — 设置定时提醒
+- plgreload — 热重载插件
+- spawn_child — 生成子 Agent 执行独立任务
 
 回复你的真实想法，用自然语言与用户交流。`,
 		Provider:        provider,
diff --git a/internal/agent/core/agent.go b/internal/agent/core/agent.go
index 9e2bd98..0b1f04c 100644
--- a/internal/agent/core/agent.go
+++ b/internal/agent/core/agent.go
@@ -986,12 +986,10 @@ func (a *Agent) executeDocTool(tc agentAPI.ToolCall) string {
 		if query == "" {
 			return "请输入查询内容"
 		}
-		docs := a.docStore.Query(query, topK)
+		docs := a.docStore.Consume(query, topK)
 		if len(docs) == 0 {
 			return "未找到相关文档记忆"
 		}
-		// 返回内容后从冷层移除，避免后续自动注入重复
-		a.docStore.Consume(query, topK)
 		var parts []string
 		for i, d := range docs {
 			parts = append(parts, fmt.Sprintf("[%d] %s (来源: %s)", i+1, d.Summary, d.Source))
diff --git a/internal/memory/pipeline/pipeline.go b/internal/memory/pipeline/pipeline.go
index 432ec9b..26a4815 100644
--- a/internal/memory/pipeline/pipeline.go
+++ b/internal/memory/pipeline/pipeline.go
@@ -6,6 +6,7 @@ import (
 	"log"
 	"os"
 	"path/filepath"
+	"strings"
 	"sync"
 	"time"
 
@@ -213,15 +214,179 @@ func (d *Distiller) cleanupRawFiles() {
 
 func extractKeyTriples(userContent, assistantContent string) []memory.Triple {
 	var triples []memory.Triple
-	if len(userContent) > 0 && len(userContent) < 500 {
-		triples = append(triples, memory.Triple{Subject: "用户", Relation: "提及", Object: truncate(userContent, 200)})
+
+	// 提取对话中的关键信息，而不是直接 dump 原文
+	// 规则1: "我的名字是X" / "我叫X" → (用户, 姓名, X)
+	if name := extractName(userContent); name != "" {
+		triples = append(triples, memory.Triple{Subject: "用户", Relation: "姓名", Object: name})
 	}
-	if len(assistantContent) > 0 && len(assistantContent) < 500 {
-		triples = append(triples, memory.Triple{Subject: "AI", Relation: "回应", Object: truncate(assistantContent, 200)})
+	// 规则2: "我住在X" / "我家在X" → (用户, 居住地, X)
+	if loc := extractLocation(userContent); loc != "" {
+		triples = append(triples, memory.Triple{Subject: "用户", Relation: "居住地", Object: loc})
 	}
+	// 规则3: "我喜欢X" / "我爱X" → (用户, 喜好, X)
+	if like := extractLike(userContent); like != "" {
+		triples = append(triples, memory.Triple{Subject: "用户", Relation: "喜好", Object: like})
+	}
+	// 规则4: "我X岁" / "我的年龄是X" → (用户, 年龄, X)
+	if age := extractAge(userContent); age != "" {
+		triples = append(triples, memory.Triple{Subject: "用户", Relation: "年龄", Object: age})
+	}
+	// 规则5: "我的工作是X" / "我在X工作" → (用户, 职业, X)
+	if job := extractJob(userContent); job != "" {
+		triples = append(triples, memory.Triple{Subject: "用户", Relation: "职业", Object: job})
+	}
+
 	return triples
 }
 
+func extractName(s string) string {
+	patterns := []struct {
+		prefix string
+		suffix string
+	}{
+		{"我叫", ""},
+		{"我的名字是", ""},
+		{"我是", ""},
+		{"名字是", ""},
+	}
+	s = strings.TrimSpace(s)
+	for _, p := range patterns {
+		if strings.HasPrefix(s, p.prefix) {
+			candidate := strings.TrimPrefix(s, p.prefix)
+			if p.suffix != "" && strings.Contains(candidate, p.suffix) {
+				candidate = candidate[:strings.Index(candidate, p.suffix)]
+			}
+			candidate = strings.TrimSpace(candidate)
+			// 取第一个空格/逗号/句号前的内容
+			for _, sep := range []string{"，", "。", " ", ","} {
+				if idx := strings.Index(candidate, sep); idx > 0 {
+					candidate = candidate[:idx]
+				}
+			}
+			if len(candidate) > 0 && len(candidate) < 20 {
+				return candidate
+			}
+		}
+	}
+	return ""
+}
+
+func extractLocation(s string) string {
+	s = strings.TrimSpace(s)
+	after := ""
+	switch {
+	case strings.HasPrefix(s, "我住在"):
+		after = strings.TrimPrefix(s, "我住在")
+	case strings.HasPrefix(s, "我家在"):
+		after = strings.TrimPrefix(s, "我家在")
+	case strings.HasPrefix(s, "我居住在"):
+		after = strings.TrimPrefix(s, "我居住在")
+	case strings.HasPrefix(s, "住在"):
+		after = strings.TrimPrefix(s, "住在")
+	default:
+		return ""
+	}
+	for _, sep := range []string{"。", "，", " ", ","} {
+		if idx := strings.Index(after, sep); idx > 0 {
+			after = after[:idx]
+		}
+	}
+	if len(after) > 0 && len(after) < 50 {
+		return strings.TrimSpace(after)
+	}
+	return ""
+}
+
+func extractLike(s string) string {
+	s = strings.TrimSpace(s)
+	after := ""
+	switch {
+	case strings.HasPrefix(s, "我喜欢"):
+		after = strings.TrimPrefix(s, "我喜欢")
+	case strings.HasPrefix(s, "我爱"):
+		after = strings.TrimPrefix(s, "我爱")
+	case strings.HasPrefix(s, "我最喜欢"):
+		after = strings.TrimPrefix(s, "我最喜欢")
+	default:
+		return ""
+	}
+	for _, sep := range []string{"。", "，", " ", ","} {
+		if idx := strings.Index(after, sep); idx > 0 {
+			after = after[:idx]
+		}
+	}
+	if len(after) > 0 && len(after) < 50 {
+		return strings.TrimSpace(after)
+	}
+	return ""
+}
+
+func extractAge(s string) string {
+	s = strings.TrimSpace(s)
+	after := ""
+	switch {
+	case strings.HasPrefix(s, "我"):
+		rest := strings.TrimPrefix(s, "我")
+		if strings.Contains(rest, "岁") {
+			after = rest[:strings.Index(rest, "岁")]
+		} else if strings.HasPrefix(rest, "的年龄是") {
+			after = strings.TrimPrefix(rest, "的年龄是")
+		} else {
+			return ""
+		}
+	default:
+		return ""
+	}
+	for _, sep := range []string{"。", "，", " ", ","} {
+		if idx := strings.Index(after, sep); idx > 0 {
+			after = after[:idx]
+		}
+	}
+	if len(after) > 0 && len(after) < 5 {
+		return strings.TrimSpace(after)
+	}
+	return ""
+}
+
+func extractJob(s string) string {
+	s = strings.TrimSpace(s)
+	after := ""
+	switch {
+	case strings.HasPrefix(s, "我的工作是"):
+		after = strings.TrimPrefix(s, "我的工作是")
+	case strings.HasPrefix(s, "我在"):
+		rest := strings.TrimPrefix(s, "我在")
+		if strings.Contains(rest, "工作") {
+			after = rest[:strings.Index(rest, "工作")]
+		} else {
+			return ""
+		}
+	case strings.HasPrefix(s, "我是"):
+		rest := strings.TrimPrefix(s, "我是")
+		// "我是一个程序员" / "我是老师"
+		for _, keyword := range []string{"一个", "一名", "一位"} {
+			if strings.HasPrefix(rest, keyword) {
+				rest = strings.TrimPrefix(rest, keyword)
+				break
+			}
+		}
+		// 职业通常较短，先看看
+		after = rest
+	default:
+		return ""
+	}
+	for _, sep := range []string{"。", "，", " ", ",", "。"} {
+		if idx := strings.Index(after, sep); idx > 0 {
+			after = after[:idx]
+		}
+	}
+	if len(after) > 0 && len(after) < 20 {
+		return strings.TrimSpace(after)
+	}
+	return ""
+}
+
 func truncate(s string, max int) string {
 	if len(s) > max {
 		return s[:max] + "..."