Improve repository discoverability messaging

Author: zxxz1000

README.md

@@ -54,6 +54,16 @@ It also treats image, audio, and video as first-class citizens. Text, audio, and
 
 **Core keywords:** multimodal memory, video ingest, audio ingest, knowledge graph, TKG, tenant isolation, agent memory
 
+## 🎯 Use Cases
+
+Use LATRACE when you need one of these product capabilities:
+
+- Build long-term memory for AI agents, copilots, or chatbots without hand-rolling a memory stack
+- Add unified text, audio, and video memory to the same application instead of splitting modalities into separate silos
+- Keep memory tenant-isolated for SaaS products where `tenant_id`, `user_id`, and `memory_domain` must stay clean
+- Explain why a retrieval result exists by walking back through evidence, timeline, entity, and graph edges
+- Turn session data into a typed graph that supports retrieval, timeline views, entity profiles, and agent tools
+
 ## 🌐 One Graph, All Modalities
 
 LATRACE uses the same memory backbone for every modality:
@@ -65,6 +75,24 @@ LATRACE uses the same memory backbone for every modality:
 
 The practical result is simple: a spoken sentence, a frame sequence, and a dialog snippet can be queried, explained, and published together instead of living in separate systems.
 
+## ❓ FAQ
+
+**What is LATRACE?**
+
+LATRACE is a multimodal memory service for AI products. It turns text, audio, image, and video into the same tenant-isolated typed graph and retrieval surface.
+
+**How is LATRACE different from plain RAG?**
+
+Plain RAG retrieves chunks. LATRACE writes memory as entities, events, states, timelines, and evidence chains, then lets you retrieve or query them in structured ways.
+
+**How do audio and video fit into the same graph as text?**
+
+They enter through source-based ingest endpoints, are compiled into the same TKG backbone, and reuse the same isolation, time semantics, and write contracts as dialog memory.
+
+**When should I use Retrieval API vs ADK tools?**
+
+Use the Retrieval API when you want the fastest path to memory-enhanced prompting. Use ADK tools when your agent should ask explicit memory questions at runtime.
+
 ---
 
 ## 🚀 Quick Start

README_zh.md

@@ -54,6 +54,16 @@ LATRACE 是一个**生产级多模态记忆服务**，面向聊天助手、Copil
 
 **核心关键词：** 多模态记忆、视频接入、音频接入、知识图谱、TKG、租户隔离、Agent 记忆
 
+## 🎯 适用场景
+
+当你的产品需要下面这些能力时，LATRACE 就是合适的底座：
+
+- 为聊天助手、Copilot 或 Agent 系统提供长期记忆，而不是自己从零拼一套 memory stack
+- 把文本、音频、图片、视频统一接入同一张图，而不是让不同模态各自形成孤岛
+- 在 SaaS 场景下严格按 `tenant_id`、`user_id`、`memory_domain` 做记忆隔离
+- 不只“召回一段文本”，还能回溯证据链、时间线、实体关系和图谱边
+- 把会话与媒体数据沉淀成 typed graph，支撑检索、时间线、实体画像和 Agent 工具调用
+
 ## 🌐 全模态同图谱
 
 LATRACE 统一使用同一套记忆底座承接所有模态：
@@ -65,6 +75,24 @@ LATRACE 统一使用同一套记忆底座承接所有模态：
 
 结果很直接：对白、语音片段、视频片段不再分属不同系统，而是能一起被查询、解释和发布。
 
+## ❓ 常见问题
+
+**LATRACE 到底是什么？**
+
+LATRACE 是一个面向 AI 产品的多模态记忆服务。它把文本、音频、图片和视频统一编译进同一张带隔离语义的 typed graph 和检索面。
+
+**它和普通 RAG 有什么不同？**
+
+普通 RAG 主要检索文本块。LATRACE 会把记忆写成实体、事件、状态、时间线和证据链，再提供结构化的召回和查询能力。
+
+**音频和视频为什么能和文本落在同一张图里？**
+
+因为它们通过 source-based ingest 入口进入系统，随后被编译到同一套 TKG 主干上，复用和文本一样的隔离键、时间语义与写入契约。
+
+**什么时候该用 Retrieval API，什么时候该用 ADK tools？**
+
+如果你想最快把长期记忆接进 prompt 流程，就用 Retrieval API；如果你希望 Agent 在运行时主动发起记忆查询，就用 ADK tools。
+
 ---
 
 <a id="quick-start"></a>