use echo_agent::agent::Agent;
use echo_agent::memory::store::{InMemoryStore, Store};
use echo_agent::memory::{Embedder, EmbeddingStore, HttpEmbedder};
use echo_agent::prelude::*;
use echo_core::memory::SearchQuery;
use serde_json::json;
use std::sync::Arc;
use std::time::Duration;
#[tokio::main]
async fn main() -> echo_agent::error::Result<()> {
dotenvy::dotenv().ok();
tracing_subscriber::fmt()
.with_env_filter(
std::env::var("RUST_LOG")
.unwrap_or_else(|_| "echo_agent=warn,demo18_semantic_memory=info".into()),
)
.init();
print_banner();
separator("Part 1: 关键词搜索 vs 语义搜索");
demo_keyword_vs_semantic().await?;
separator("Part 2: remember + recall 工具（语义模式）");
demo_agent_with_semantic_memory().await?;
separator("Part 3: set_memory_store — 运行时替换 Store");
demo_set_memory_store().await?;
println!("\n{}", "═".repeat(64));
println!(" demo18 完成");
println!("{}", "═".repeat(64));
Ok(())
}
async fn demo_keyword_vs_semantic() -> echo_agent::error::Result<()> {
println!(" 对比「关键词检索」和「语义检索」在跨语言查询上的差异\n");
let embedder = load_verified_embedder_from_config().await?;
let memories = [
("用户偏好深色主题", vec!["偏好", "界面"]),
("用户喜欢古典音乐，尤其是肖邦", vec!["偏好", "音乐"]),
("用户是 Rust 程序员，主要做后端开发", vec!["职业", "编程"]),
];
let kw_store = Arc::new(InMemoryStore::new());
let inner = Arc::new(InMemoryStore::new());
let sem_store = Arc::new(EmbeddingStore::new(inner.clone(), embedder));
let ns = &["demo", "memories"];
for (i, (content, tags)) in memories.iter().enumerate() {
let key = format!("mem-{i:03}");
let value = json!({ "content": content, "tags": tags });
kw_store.put(ns, &key, value.clone()).await?;
sem_store.put(ns, &key, value).await?;
}
println!(" 已写入 {} 条记忆（中文）\n", memories.len());
let queries = [
("music preference", "英文查询「音乐偏好」"),
("dark mode", "英文查询「深色模式」"),
];
for (query, desc) in &queries {
let kw_hits = kw_store.search(ns, query, 3).await?;
let sem_hits = sem_store
.search_with(ns, SearchQuery::semantic(query, 3))
.await?;
if sem_hits.is_empty() {
return Err(echo_agent::error::ReactError::Other(format!(
"demo18 验收失败：语义查询 `{query}` 没有命中"
)));
}
println!(" 🔍 查询: \"{query}\" ({desc})");
println!(
" 关键词检索: {} 条命中 {}",
kw_hits.len(),
if kw_hits.is_empty() { "❌" } else { "✅" }
);
println!(
" 语义搜索: {} 条命中 {}",
sem_hits.len(),
if !sem_hits.is_empty() { "✅" } else { "❌" }
);
println!();
}
Ok(())
}
async fn demo_agent_with_semantic_memory() -> echo_agent::error::Result<()> {
println!(" Agent 使用语义记忆，通过 recall 工具语义检索\n");
let embedder = load_verified_embedder_from_config().await?;
let inner = Arc::new(InMemoryStore::new());
let store = Arc::new(EmbeddingStore::new(inner as Arc<dyn Store>, embedder));
let ns = ["memory_agent".to_string(), "memories".to_string()];
let ns_ref: Vec<&str> = ns.iter().map(String::as_str).collect();
store
.put(
&ns_ref,
"m1",
json!({"content": "用户叫 Alice，是一名数据科学家"}),
)
.await?;
store
.put(
&ns_ref,
"m2",
json!({"content": "用户偏好 Python 和 PyTorch"}),
)
.await?;
println!(" 📚 预填充 2 条长期记忆\n");
let mut agent = ReactAgentBuilder::new()
.model("qwen3-max")
.name("memory_agent")
.system_prompt("你是 Alice 的私人助手，结合长期记忆给出个性化建议。")
.enable_tools()
.max_iterations(5)
.build()?;
agent.set_memory_store(store);
println!(" 👤 用户: 帮我推荐适合数据科学研究的 Rust 库");
let answer = agent.execute("帮我推荐适合数据科学研究的 Rust 库").await?;
if answer.trim().is_empty() {
return Err(echo_agent::error::ReactError::Other(
"demo18 验收失败：Agent 语义记忆回答为空".to_string(),
));
}
println!(" 🤖 Agent: {answer}\n");
Ok(())
}
async fn demo_set_memory_store() -> echo_agent::error::Result<()> {
println!(" 展示 set_memory_store() 在 new() 之后热替换 Store 的用法\n");
let mut agent = ReactAgentBuilder::new()
.model("qwen3-max")
.name("hotswap_agent")
.system_prompt("你是一个助手。")
.max_iterations(3)
.build()?;
println!(" ✅ ReactAgent 创建完成（无 Store）");
let embedder = load_verified_embedder_from_config().await?;
let inner = Arc::new(InMemoryStore::new());
let store = Arc::new(EmbeddingStore::new(inner as Arc<dyn Store>, embedder));
agent.set_memory_store(store.clone());
println!(" ✅ EmbeddingStore 已挂载");
let names = agent.tool_names();
if !names.contains(&"remember") || !names.contains(&"recall") || !names.contains(&"forget") {
return Err(echo_agent::error::ReactError::Other(
"demo18 验收失败：热挂载 EmbeddingStore 后记忆工具未完整注册".to_string(),
));
}
println!("\n ℹ️ 接下来 Agent 的 remember/recall 工具将使用向量检索");
Ok(())
}
fn print_banner() {
println!("{}", "═".repeat(64));
println!(" Echo Agent × Store 语义搜索 (demo18)");
println!("{}", "═".repeat(64));
println!();
}
fn separator(title: &str) {
println!("{}", "─".repeat(64));
println!("{title}\n");
}
fn load_embedder_from_config() -> Option<Arc<dyn Embedder>> {
let cfg = echo_agent::llm::config::Config::get_embedding().ok()?;
let embedder = HttpEmbedder::with_endpoint(cfg.url, cfg.api_key, cfg.model)
.with_timeout(Duration::from_secs(cfg.timeout_secs));
Some(Arc::new(embedder))
}
async fn load_verified_embedder_from_config() -> echo_agent::error::Result<Arc<dyn Embedder>> {
let embedder = load_embedder_from_config().ok_or_else(|| {
echo_agent::error::ReactError::Other("demo18 验收失败：缺少 embedding 配置".to_string())
})?;
embedder
.embed("demo18 semantic memory health check")
.await
.map_err(|e| {
echo_agent::error::ReactError::Other(format!(
"demo18 验收失败：embedding 健康检查失败: {e}"
))
})?;
Ok(embedder)
}