diff --git a/go/internal/api/routes/admin.go b/go/internal/api/routes/admin.go
index ac4b933..a52f6be 100644
--- a/go/internal/api/routes/admin.go
+++ b/go/internal/api/routes/admin.go
@@ -460,4 +460,56 @@ func (aa *AdminAPI) ListMemories(w http.ResponseWriter, r *http.Request) {
 	respond(w, 200, map[string]interface{}{
 		"memories": items, "count": len(items), "limit": limit,
 	})
+}
+
+// ExportMemoriesMD 导出记忆为 Markdown（P5：EverOS 式 md 真相层）
+// GET /api/v1/memories/export?namespace=hermes-main&limit=1000&format=md
+// 输出人可读的 Markdown 文档：记忆可迁移、可备份、可人工审查
+func (aa *AdminAPI) ExportMemoriesMD(w http.ResponseWriter, r *http.Request) {
+	ns := r.URL.Query().Get("namespace")
+	limit := 1000
+	if l := r.URL.Query().Get("limit"); l != "" {
+		if v, err := fmt.Sscanf(l, "%d", &limit); err != nil || v != 1 || limit < 1 {
+			limit = 1000
+		}
+		if limit > 5000 {
+			limit = 5000
+		}
+	}
+	zeroVec := make([]float32, 1024)
+	results, err := aa.LanceDB.Search("memories", zeroVec, limit, ns)
+	if err != nil {
+		respondError(w, 500, "export memories: "+err.Error())
+		return
+	}
+
+	var sb strings.Builder
+	sb.WriteString("# 织忆记忆导出\n\n")
+	sb.WriteString(fmt.Sprintf("> 导出时间: %s\n", time.Now().Format("2006-01-02 15:04:05")))
+	sb.WriteString(fmt.Sprintf("> 命名空间: %s | 条数: %d\n\n---\n\n", orDefault(ns, "all"), len(results)))
+
+	for i, m := range results {
+		sb.WriteString(fmt.Sprintf("## M%d — %s\n\n", i+1, m.ID))
+		sb.WriteString(fmt.Sprintf("- **分类**: %s\n", orDefault(m.Category, "unknown")))
+		if m.Namespace != "" {
+			sb.WriteString(fmt.Sprintf("- **命名空间**: %s\n", m.Namespace))
+		}
+		sb.WriteString(fmt.Sprintf("- **重要性**: %.2f\n", m.Importance))
+		sb.WriteString(fmt.Sprintf("- **时间**: %s\n", m.CreatedAt.Format("2006-01-02 15:04:05")))
+		sb.WriteString("\n### 内容\n\n")
+		sb.WriteString(m.Content)
+		sb.WriteString("\n\n---\n\n")
+	}
+
+	w.Header().Set("Content-Type", "text/markdown; charset=utf-8")
+	w.Header().Set("Content-Disposition", "attachment; filename=zhiyi-memories-"+time.Now().Format("20060102")+".md")
+	w.WriteHeader(200)
+	w.Write([]byte(sb.String()))
+}
+
+func orDefault(s, def string) string {
+	if s == "" {
+		return def
+	}
+	return s
 }
\ No newline at end of file
diff --git a/go/internal/api/server.go b/go/internal/api/server.go
index 76c0def..5797efd 100644
--- a/go/internal/api/server.go
+++ b/go/internal/api/server.go
@@ -913,6 +913,7 @@ func NewServer() http.Handler {
 	})
 	// GET /api/v1/memories — list all memories (zero-vector search, for plugin compat)
 	mux.HandleFunc("/api/v1/memories", adminAPI.ListMemories)
+	mux.HandleFunc("/api/v1/memories/export", adminAPI.ExportMemoriesMD)
 	// /api/v1/memory/{id}/versions — existing version history endpoint
 	mux.HandleFunc("/api/v1/memory/", func(w http.ResponseWriter, r *http.Request) {
 		if r.Method == "GET" {
diff --git a/go/internal/distill/aaak.go b/go/internal/distill/aaak.go
new file mode 100644
index 0000000..795057b
--- /dev/null
+++ b/go/internal/distill/aaak.go
@@ -0,0 +1,157 @@
+package distill
+
+import (
+	"log"
+	"strings"
+	"unicode"
+)
+
+// AAAKEntry AAAK 风格压缩索引条目（借鉴 mempalace/dialect.py 的 AAAK 设计）
+// 目标：为每条事实生成紧凑结构化摘要，LLM 可读、无需解码器，
+// 召回时先扫索引定位相关事实，再读原文（索引层指向内容层）。
+type AAAKEntry struct {
+	FactID   string   `json:"fact_id"`   // 事实编号（F1, F2, ...）
+	Primary  string   `json:"primary"`   // 主实体（最重要实体，如人名/项目名）
+	Entities []string `json:"entities"`  // 全部相关实体
+	Keywords []string `json:"keywords"`  // 主题关键词（2-4 个）
+	Quote    string   `json:"quote"`     // 关键短语（截断 ≤40 字）
+	Weight   float64  `json:"weight"`    // 权重（由 5D 分数综合）
+	Kind     string   `json:"kind"`      // 类型：fact / decision / action / question / conclusion
+}
+
+// buildAAAKIndex 为事实列表生成 AAAK 压缩索引
+func buildAAAKIndex(facts []string, entities []Entity, overall float64) []AAAKEntry {
+	entries := make([]AAAKEntry, 0, len(facts))
+	entityNames := make([]string, 0, len(entities))
+	for _, e := range entities {
+		if e.Name != "" {
+			entityNames = append(entityNames, e.Name)
+		}
+	}
+
+	for i, fact := range facts {
+		if strings.TrimSpace(fact) == "" {
+			continue
+		}
+		entry := AAAKEntry{
+			FactID:   "F" + itoa(i+1),
+			Primary:  pickPrimary(fact, entityNames),
+			Entities: pickRelatedEntities(fact, entityNames, 4),
+			Keywords: pickKeywords(fact, 3),
+			Quote:    truncate(fact, 40),
+			Weight:   overall,
+			Kind:     classifyFact(fact),
+		}
+		entries = append(entries, entry)
+	}
+	return entries
+}
+
+// classifyFact 事实类型分类
+func classifyFact(fact string) string {
+	switch {
+	case containsAny(fact, []string{"决定", "选择", "采用", "确定", "配置", "改", "换"}):
+		return "decision"
+	case containsAny(fact, []string{"完成", "实现", "部署", "安装", "修复", "上线", "验证", "测试"}):
+		return "action"
+	case containsAny(fact, []string{"?", "？", "是否", "吗", "未", "待", "需要"}):
+		return "question"
+	case containsAny(fact, []string{"结论", "因此", "所以", "总之", "意味着"}):
+		return "conclusion"
+	default:
+		return "fact"
+	}
+}
+
+// pickPrimary 选取主实体（事实中第一个出现的已知实体，否则第一个词）
+func pickPrimary(fact string, entityNames []string) string {
+	for _, name := range entityNames {
+		if name != "" && strings.Contains(fact, name) {
+			return name
+		}
+	}
+	// 退而取第一个非停用词 token
+	fields := strings.Fields(fact)
+	for _, f := range fields {
+		clean := strings.Trim(f, ",.;:!?，。；：！？、\"'()（）[]【】")
+		if len([]rune(clean)) >= 2 && !isStopWord(strings.ToLower(clean)) {
+			return clean
+		}
+	}
+	return ""
+}
+
+// pickRelatedEntities 选取事实中出现的相关实体（最多 max 个）
+func pickRelatedEntities(fact string, entityNames []string, max int) []string {
+	var picked []string
+	for _, name := range entityNames {
+		if name != "" && strings.Contains(fact, name) {
+			picked = append(picked, name)
+			if len(picked) >= max {
+				break
+			}
+		}
+	}
+	return picked
+}
+
+// pickKeywords 提取主题关键词（从事实中挑有意义的词，最多 max 个）
+func pickKeywords(fact string, max int) []string {
+	var kws []string
+	seen := make(map[string]bool)
+	fields := strings.Fields(fact)
+	for _, f := range fields {
+		clean := strings.Trim(f, ",.;:!?，。；：！？、\"'()（）[]【】")
+		runes := []rune(clean)
+		if len(runes) < 2 || len(runes) > 10 {
+			continue
+		}
+		// 跳过纯标点/停用词
+		if isStopWord(strings.ToLower(clean)) || isPunctuationOnly(clean) {
+			continue
+		}
+		// 优先中文实体和技术词
+		key := strings.ToLower(clean)
+		if !seen[key] {
+			seen[key] = true
+			kws = append(kws, clean)
+			if len(kws) >= max {
+				break
+			}
+		}
+	}
+	return kws
+}
+
+func isPunctuationOnly(s string) bool {
+	for _, r := range s {
+		if unicode.IsLetter(r) || unicode.IsDigit(r) {
+			return false
+		}
+	}
+	return true
+}
+
+// itoa 简单整数转字符串（避免引入 strconv 依赖之外的复杂度）
+func itoa(n int) string {
+	if n == 0 {
+		return "0"
+	}
+	digits := []byte{}
+	for n > 0 {
+		digits = append([]byte{byte('0' + n%10)}, digits...)
+		n /= 10
+	}
+	return string(digits)
+}
+
+// logAAAKIndex 输出索引（调试用）
+func logAAAKIndex(entries []AAAKEntry) {
+	if len(entries) == 0 {
+		return
+	}
+	for _, e := range entries {
+		log.Printf("[aaak] %s|%s|%s|%.2f|%s",
+			e.FactID, e.Primary, strings.Join(e.Keywords, ","), e.Weight, e.Kind)
+	}
+}
diff --git a/go/internal/distill/engine.go b/go/internal/distill/engine.go
index 2b3a1ab..0847ef4 100644
--- a/go/internal/distill/engine.go
+++ b/go/internal/distill/engine.go
@@ -47,6 +47,7 @@ type DistillResult struct {
 	Entities []Entity
 	Score5D  FiveDScore
 	Overall  float64
+	Index    []AAAKEntry // P4: AAAK 压缩索引（每条事实的紧凑摘要）
 }
 
 // Entity 实体
@@ -272,11 +273,18 @@ func (e *Engine) distillOne(input DistillInput) DistillResult {
 		facts = heuristicFacts
 	}
 
+	// P4: 生成 AAAK 压缩索引（每条事实的紧凑摘要，供召回快速定位）
+	index := buildAAAKIndex(facts, entities, overall)
+	if len(index) > 0 {
+		logAAAKIndex(index)
+	}
+
 	return DistillResult{
 		Facts:    facts,
 		Entities: entities,
 		Score5D:  score,
 		Overall:  overall,
+		Index:    index,
 	}
 }