feat: 知识库支持目录来源——多个目录批量导入、后缀过滤（默认全部）、可开关递归子目录，只收录文本/图片视频/音频三类其余忽略

apps/desktop/src/commands.rs

@@ -1632,6 +1632,8 @@ pub fn kb_import_documents(
             file_size: bytes.len() as i64,
             char_count: text.chars().count() as i64,
             chunk_count: chunks.len() as i64,
+            file_path: String::new(),
+            source_id: None,
             created_at: String::new(),
             updated_at: String::new(),
         };
@@ -1684,6 +1686,193 @@ pub fn kb_search(
         .map_err(|e| e.to_string())
 }
 
+#[derive(Deserialize)]
+pub struct KbSourceInput {
+    pub kb_id: String,
+    pub path: String,
+    #[serde(default)]
+    pub extensions: String,
+    #[serde(default = "default_true")]
+    pub recursive: bool,
+}
+
+#[derive(Serialize)]
+pub struct KbSourceScanResult {
+    pub imported: usize,
+    pub ignored: usize,
+    pub errors: Vec<String>,
+}
+
+#[tauri::command]
+pub fn kb_list_sources(
+    state: State<'_, App>,
+    kb_id: String,
+) -> Result<Vec<xianren_core::KbSource>, String> {
+    let db = state.core.db.lock().unwrap();
+    kb_db::list_sources(&db, &kb_id).map_err(|e| e.to_string())
+}
+
+/// 添加目录来源并立即扫描导入：只收录文本 / 图片视频 / 音频三类文件，其余忽略。
+#[tauri::command]
+pub fn kb_add_source(
+    state: State<'_, App>,
+    input: KbSourceInput,
+) -> Result<KbSourceScanResult, String> {
+    let dir = std::path::PathBuf::from(input.path.trim());
+    if !dir.is_dir() {
+        return Err(format!("目录不存在或不是文件夹：{}", dir.display()));
+    }
+    let exts = crate::knowledge::normalize_extensions(&input.extensions);
+    let db = state.core.db.lock().unwrap();
+    let kb = kb_db::get_knowledge_base(&db, &input.kb_id)
+        .map_err(|e| e.to_string())?
+        .ok_or_else(|| "知识库不存在".to_string())?;
+    let source_id = uuid::Uuid::new_v4().to_string();
+    let source = xianren_core::KbSource {
+        id: source_id.clone(),
+        kb_id: input.kb_id.clone(),
+        path: dir.to_string_lossy().to_string(),
+        extensions: input.extensions.trim().to_string(),
+        recursive: input.recursive,
+        created_at: String::new(),
+    };
+    kb_db::upsert_source(&db, &source).map_err(|e| e.to_string())?;
+    // 同目录重复添加时沿用已有来源 id，避免文档归属错乱
+    let source_id = kb_db::list_sources(&db, &input.kb_id)
+        .map_err(|e| e.to_string())?
+        .into_iter()
+        .find(|s| s.path == source.path && s.kb_id == source.kb_id)
+        .map(|s| s.id)
+        .unwrap_or(source_id);
+    Ok(scan_and_import(&db, &kb, &source_id, &dir, &exts, input.recursive))
+}
+
+/// 重新扫描该知识库的全部目录来源，导入新增文件（按文件路径去重）。
+#[tauri::command]
+pub fn kb_scan_sources(
+    state: State<'_, App>,
+    kb_id: String,
+) -> Result<KbSourceScanResult, String> {
+    let db = state.core.db.lock().unwrap();
+    let kb = kb_db::get_knowledge_base(&db, &kb_id)
+        .map_err(|e| e.to_string())?
+        .ok_or_else(|| "知识库不存在".to_string())?;
+    let sources = kb_db::list_sources(&db, &kb_id).map_err(|e| e.to_string())?;
+    let mut total = KbSourceScanResult {
+        imported: 0,
+        ignored: 0,
+        errors: Vec::new(),
+    };
+    for src in sources {
+        let dir = std::path::PathBuf::from(&src.path);
+        if !dir.is_dir() {
+            total.errors.push(format!("目录不存在：{}", src.path));
+            continue;
+        }
+        let exts = crate::knowledge::normalize_extensions(&src.extensions);
+        let r = scan_and_import(&db, &kb, &src.id, &dir, &exts, src.recursive);
+        total.imported += r.imported;
+        total.ignored += r.ignored;
+        total.errors.extend(r.errors);
+    }
+    Ok(total)
+}
+
+/// 删除目录来源及其导入的全部文档。
+#[tauri::command]
+pub fn kb_remove_source(state: State<'_, App>, source_id: String) -> Result<usize, String> {
+    let db = state.core.db.lock().unwrap();
+    kb_db::get_source(&db, &source_id)
+        .map_err(|e| e.to_string())?
+        .ok_or_else(|| "目录来源不存在".to_string())?;
+    let removed = kb_db::delete_docs_by_source(&db, &source_id).map_err(|e| e.to_string())?;
+    kb_db::delete_source(&db, &source_id).map_err(|e| e.to_string())?;
+    Ok(removed)
+}
+
+/// 扫描目录并导入文档（文本提取切块；图片/视频/音频以文件名为内容登记；其他忽略）。
+fn scan_and_import(
+    db: &rusqlite::Connection,
+    kb: &xianren_core::KnowledgeBase,
+    source_id: &str,
+    dir: &std::path::Path,
+    extensions: &[String],
+    recursive: bool,
+) -> KbSourceScanResult {
+    let mut result = KbSourceScanResult {
+        imported: 0,
+        ignored: 0,
+        errors: Vec::new(),
+    };
+    let files = crate::knowledge::collect_files(dir, extensions, recursive);
+    for file in files {
+        let name = file
+            .file_name()
+            .and_then(|n| n.to_str())
+            .unwrap_or("")
+            .to_string();
+        let file_path = file.to_string_lossy().to_string();
+        let category = crate::knowledge::classify(&name);
+        if category == crate::knowledge::FileCategory::Other {
+            result.ignored += 1;
+            continue;
+        }
+        if kb_db::document_exists(db, &kb.id, &file_path).unwrap_or(false) {
+            continue;
+        }
+        let bytes = match std::fs::read(&file) {
+            Ok(b) => b,
+            Err(e) => {
+                result.errors.push(format!("{name}: 读取失败 {e}"));
+                continue;
+            }
+        };
+        let content = if category == crate::knowledge::FileCategory::Text {
+            match crate::knowledge::extract_text(&name, &bytes) {
+                Ok(t) => t,
+                Err(e) => {
+                    result.errors.push(format!("{name}: {e}"));
+                    continue;
+                }
+            }
+        } else {
+            // 图片 / 视频 / 音频：以文件名为内容登记，便于按名检索
+            name.clone()
+        };
+        let chunks = if category == crate::knowledge::FileCategory::Text {
+            kb_db::chunk_text(&content, kb.chunk_size as usize, kb.chunk_overlap as usize)
+        } else {
+            vec![content.clone()]
+        };
+        if chunks.is_empty() {
+            result.errors.push(format!("{name}: 切分后没有有效分块"));
+            continue;
+        }
+        let doc = xianren_core::KbDocument {
+            id: uuid::Uuid::new_v4().to_string(),
+            kb_id: kb.id.clone(),
+            name,
+            file_type: file
+                .extension()
+                .map(|e| e.to_string_lossy().to_lowercase())
+                .unwrap_or_default(),
+            file_size: bytes.len() as i64,
+            char_count: content.chars().count() as i64,
+            chunk_count: chunks.len() as i64,
+            file_path,
+            source_id: Some(source_id.to_string()),
+            created_at: String::new(),
+            updated_at: String::new(),
+        };
+        if let Err(e) = kb_db::insert_document_with_chunks(db, &doc, &content, &chunks) {
+            result.errors.push(format!("{}: 写入失败 {e}", doc.name));
+            continue;
+        }
+        result.imported += 1;
+    }
+    result
+}
+
 #[tauri::command]
 pub fn list_conversations(
     state: State<'_, App>,

apps/desktop/src/knowledge.rs

@@ -1,10 +1,95 @@
-/// 按扩展名从文件字节提取纯文本：文本类文件直接按 UTF-8 读取，PDF 走 pdf-extract。
-pub fn extract_text(name: &str, bytes: &[u8]) -> Result<String, String> {
-    let ext = name
-        .rsplit('.')
+use std::path::Path;
+
+/// 文件分类：可读文本 / 图片视频 / 音频 / 其他（其他一律忽略）。
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum FileCategory {
+    Text,
+    Media,
+    Other,
+}
+
+const TEXT_EXTS: &[&str] = &[
+    "txt", "md", "markdown", "json", "csv", "tsv", "log", "yml", "yaml", "toml", "ini",
+    "conf", "cfg", "xml", "html", "htm", "py", "rs", "ts", "tsx", "js", "jsx", "java", "c",
+    "cpp", "h", "hpp", "go", "php", "rb", "sh", "bat", "ps1", "sql", "css", "scss", "sass",
+    "pdf",
+];
+
+const IMAGE_EXTS: &[&str] = &[
+    "png", "jpg", "jpeg", "gif", "webp", "bmp", "svg", "tiff", "tif", "ico", "avif", "heic",
+];
+
+const VIDEO_EXTS: &[&str] = &[
+    "mp4", "mkv", "avi", "mov", "webm", "flv", "wmv", "m4v", "mpg", "mpeg", "3gp",
+];
+
+const AUDIO_EXTS: &[&str] = &[
+    "mp3", "wav", "flac", "ogg", "oga", "m4a", "aac", "wma", "opus", "aiff", "aif",
+];
+
+fn ext_of(name: &str) -> String {
+    name.rsplit('.')
         .next()
         .map(|e| e.to_lowercase())
-        .unwrap_or_default();
+        .unwrap_or_default()
+}
+
+/// 分类文件：文本（可提取内容）/ 媒体（图片、视频、音频）/ 其他。
+pub fn classify(name: &str) -> FileCategory {
+    let ext = ext_of(name);
+    if TEXT_EXTS.contains(&ext.as_str()) {
+        FileCategory::Text
+    } else if IMAGE_EXTS.contains(&ext.as_str())
+        || VIDEO_EXTS.contains(&ext.as_str())
+        || AUDIO_EXTS.contains(&ext.as_str())
+    {
+        FileCategory::Media
+    } else {
+        FileCategory::Other
+    }
+}
+
+/// 规范化用户输入的后缀列表：`.txt, .md；PDF` → ["txt", "md", "pdf"]；空串表示全部。
+pub fn normalize_extensions(raw: &str) -> Vec<String> {
+    raw.split([',', ';', '，', '；', ' '])
+        .map(|s| s.trim().trim_start_matches('.').to_lowercase())
+        .filter(|s| !s.is_empty())
+        .collect::<Vec<_>>()
+}
+
+fn has_matching_ext(name: &str, extensions: &[String]) -> bool {
+    let ext = ext_of(name);
+    extensions.iter().any(|e| e == &ext)
+}
+
+/// 递归/非递归收集目录下符合后缀条件的文件，按路径排序。
+pub fn collect_files(path: &Path, extensions: &[String], recursive: bool) -> Vec<std::path::PathBuf> {
+    let mut out = Vec::new();
+    fn walk(dir: &Path, exts: &[String], recursive: bool, out: &mut Vec<std::path::PathBuf>) {
+        let Ok(entries) = std::fs::read_dir(dir) else {
+            return;
+        };
+        for entry in entries.flatten() {
+            let p = entry.path();
+            if p.is_dir() {
+                if recursive {
+                    walk(&p, exts, recursive, out);
+                }
+            } else if let Some(name) = p.file_name().and_then(|n| n.to_str()) {
+                if exts.is_empty() || has_matching_ext(name, exts) {
+                    out.push(p);
+                }
+            }
+        }
+    }
+    walk(path, extensions, recursive, &mut out);
+    out.sort();
+    out
+}
+
+/// 按扩展名从文件字节提取纯文本：文本类文件直接按 UTF-8 读取，PDF 走 pdf-extract。
+pub fn extract_text(name: &str, bytes: &[u8]) -> Result<String, String> {
+    let ext = ext_of(name);
     let text = match ext.as_str() {
         "pdf" => pdf_extract::extract_text_from_mem(bytes)
             .map_err(|e| format!("PDF 解析失败：{e}"))?,
@@ -31,3 +116,48 @@ pub fn extract_text(name: &str, bytes: &[u8]) -> Result<String, String> {
     }
     Ok(cleaned)
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn classify_three_categories() {
+        assert_eq!(classify("readme.md"), FileCategory::Text);
+        assert_eq!(classify("手册.pdf"), FileCategory::Text);
+        assert_eq!(classify("photo.PNG"), FileCategory::Media);
+        assert_eq!(classify("clip.mp4"), FileCategory::Media);
+        assert_eq!(classify("song.mp3"), FileCategory::Media);
+        assert_eq!(classify("app.exe"), FileCategory::Other);
+        assert_eq!(classify("archive.zip"), FileCategory::Other);
+        assert_eq!(classify("noext"), FileCategory::Other);
+    }
+
+    #[test]
+    fn normalize_extensions_handles_dots_and_separators() {
+        assert_eq!(
+            normalize_extensions(".txt, .md；PDF, png"),
+            vec!["txt", "md", "pdf", "png"]
+        );
+        assert!(normalize_extensions("  ").is_empty());
+    }
+
+    #[test]
+    fn collect_files_filters_and_recurses() {
+        let base = std::env::temp_dir().join(format!("kb_scan_test_{}", uuid::Uuid::new_v4()));
+        std::fs::create_dir_all(base.join("sub")).unwrap();
+        std::fs::write(base.join("a.txt"), "hello").unwrap();
+        std::fs::write(base.join("b.exe"), "x").unwrap();
+        std::fs::write(base.join("sub/c.md"), "hi").unwrap();
+        std::fs::write(base.join("sub/d.png"), "img").unwrap();
+
+        let all = collect_files(&base, &[], true);
+        assert_eq!(all.len(), 4);
+        let txt_only = collect_files(&base, &["txt".to_string()], true);
+        assert_eq!(txt_only.len(), 1);
+        let non_recursive = collect_files(&base, &[], false);
+        assert_eq!(non_recursive.len(), 2);
+
+        std::fs::remove_dir_all(&base).unwrap();
+    }
+}

apps/desktop/src/lib.rs

@@ -195,6 +195,10 @@ pub fn run() {
             commands::remove_kb_document,
             commands::kb_rechunk_document,
             commands::kb_search,
+            commands::kb_list_sources,
+            commands::kb_add_source,
+            commands::kb_scan_sources,
+            commands::kb_remove_source,
             commands::list_models,
             commands::import_model,
             commands::remove_model,

crates/core/src/app.rs

@@ -123,6 +123,8 @@ fn open_db(path: &Path) -> Result<Connection> {
     ensure_column(&conn, "scheduled_tasks", "email_mode", "TEXT NOT NULL DEFAULT 'fixed'")?;
     ensure_column(&conn, "scheduled_tasks", "email_subject", "TEXT NOT NULL DEFAULT ''")?;
     ensure_column(&conn, "scheduled_tasks", "email_body", "TEXT NOT NULL DEFAULT ''")?;
+    ensure_column(&conn, "kb_documents", "file_path", "TEXT NOT NULL DEFAULT ''")?;
+    ensure_column(&conn, "kb_documents", "source_id", "TEXT")?;
     Ok(conn)
 }
 

crates/core/src/knowledge_base.rs

@@ -24,10 +24,22 @@ pub struct KbDocument {
     pub file_size: i64,
     pub char_count: i64,
     pub chunk_count: i64,
+    pub file_path: String,
+    pub source_id: Option<String>,
     pub created_at: String,
     pub updated_at: String,
 }
 
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct KbSource {
+    pub id: String,
+    pub kb_id: String,
+    pub path: String,
+    pub extensions: String,
+    pub recursive: bool,
+    pub created_at: String,
+}
+
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct KbDocumentDetail {
     #[serde(flatten)]
@@ -45,7 +57,8 @@ pub struct KbSearchHit {
 }
 
 const KB_COLUMNS: &str = "id, name, description, chunk_size, chunk_overlap, created_at, updated_at";
-const DOC_COLUMNS: &str = "id, kb_id, name, file_type, file_size, char_count, chunk_count, created_at, updated_at";
+const DOC_COLUMNS: &str = "id, kb_id, name, file_type, file_size, char_count, chunk_count, file_path, source_id, created_at, updated_at";
+const SOURCE_COLUMNS: &str = "id, kb_id, path, extensions, recursive, created_at";
 
 // ---------- 知识库 ----------
 
@@ -149,7 +162,7 @@ pub fn get_document_detail(db: &Connection, id: &str) -> Result<Option<KbDocumen
     ))?;
     let mut rows = stmt.query_map(params![id], |row| {
         let doc = row_to_doc(row)?;
-        let content: String = row.get(9)?;
+        let content: String = row.get(11)?;
         Ok(KbDocumentDetail { doc, content })
     })?;
     match rows.next() {
@@ -168,8 +181,8 @@ pub fn insert_document_with_chunks(
     let tx = db.unchecked_transaction()?;
     tx.execute(
         "INSERT INTO kb_documents
-           (id, kb_id, name, file_type, file_size, char_count, chunk_count, content)
-         VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8)",
+           (id, kb_id, name, file_type, file_size, char_count, chunk_count, content, file_path, source_id)
+         VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10)",
         params![
             doc.id,
             doc.kb_id,
@@ -179,6 +192,8 @@ pub fn insert_document_with_chunks(
             doc.char_count,
             chunks.len() as i64,
             content,
+            doc.file_path,
+            doc.source_id,
         ],
     )?;
     for (i, chunk) in chunks.iter().enumerate() {
@@ -203,6 +218,75 @@ pub fn delete_document(db: &Connection, id: &str) -> Result<()> {
     Ok(())
 }
 
+/// 同一知识库下是否已存在来自该文件路径的文档（目录导入去重）。
+pub fn document_exists(db: &Connection, kb_id: &str, file_path: &str) -> Result<bool> {
+    if file_path.is_empty() {
+        return Ok(false);
+    }
+    let exists: bool = db.query_row(
+        "SELECT EXISTS(SELECT 1 FROM kb_documents WHERE kb_id = ?1 AND file_path = ?2)",
+        params![kb_id, file_path],
+        |row| row.get(0),
+    )?;
+    Ok(exists)
+}
+
+// ---------- 目录来源 ----------
+
+pub fn list_sources(db: &Connection, kb_id: &str) -> Result<Vec<KbSource>> {
+    let mut stmt = db.prepare(&format!(
+        "SELECT {SOURCE_COLUMNS} FROM kb_sources WHERE kb_id = ?1 ORDER BY created_at ASC"
+    ))?;
+    let rows = stmt.query_map(params![kb_id], row_to_source)?;
+    let mut out = Vec::new();
+    for row in rows {
+        out.push(row?);
+    }
+    Ok(out)
+}
+
+pub fn get_source(db: &Connection, id: &str) -> Result<Option<KbSource>> {
+    let mut stmt = db.prepare(&format!(
+        "SELECT {SOURCE_COLUMNS} FROM kb_sources WHERE id = ?1"
+    ))?;
+    let mut rows = stmt.query_map(params![id], row_to_source)?;
+    match rows.next() {
+        Some(row) => Ok(Some(row?)),
+        None => Ok(None),
+    }
+}
+
+pub fn upsert_source(db: &Connection, source: &KbSource) -> Result<()> {
+    db.execute(
+        "INSERT INTO kb_sources (id, kb_id, path, extensions, recursive)
+         VALUES (?1, ?2, ?3, ?4, ?5)
+         ON CONFLICT(kb_id, path) DO UPDATE SET
+           extensions = excluded.extensions, recursive = excluded.recursive",
+        params![
+            source.id,
+            source.kb_id,
+            source.path,
+            source.extensions,
+            source.recursive as i32,
+        ],
+    )?;
+    Ok(())
+}
+
+pub fn delete_source(db: &Connection, id: &str) -> Result<()> {
+    db.execute("DELETE FROM kb_sources WHERE id = ?1", params![id])?;
+    Ok(())
+}
+
+/// 删除某个目录来源导入的全部文档（分块随外键级联）。
+pub fn delete_docs_by_source(db: &Connection, source_id: &str) -> Result<usize> {
+    let n = db.execute(
+        "DELETE FROM kb_documents WHERE source_id = ?1",
+        params![source_id],
+    )?;
+    Ok(n)
+}
+
 /// 重新切分文档：删除旧分块后按新参数重分并重建索引。
 pub fn rechunk_document(
     db: &Connection,
@@ -328,8 +412,22 @@ fn row_to_doc(row: &rusqlite::Row<'_>) -> rusqlite::Result<KbDocument> {
         file_size: row.get(4)?,
         char_count: row.get(5)?,
         chunk_count: row.get(6)?,
-        created_at: row.get(7)?,
-        updated_at: row.get(8)?,
+        file_path: row.get(7)?,
+        source_id: row.get(8)?,
+        created_at: row.get(9)?,
+        updated_at: row.get(10)?,
+    })
+}
+
+fn row_to_source(row: &rusqlite::Row<'_>) -> rusqlite::Result<KbSource> {
+    let recursive: i32 = row.get(4)?;
+    Ok(KbSource {
+        id: row.get(0)?,
+        kb_id: row.get(1)?,
+        path: row.get(2)?,
+        extensions: row.get(3)?,
+        recursive: recursive != 0,
+        created_at: row.get(5)?,
     })
 }
 
@@ -363,6 +461,8 @@ mod tests {
             file_size: content.len() as i64,
             char_count: content.chars().count() as i64,
             chunk_count: 0,
+            file_path: String::new(),
+            source_id: None,
             created_at: String::new(),
             updated_at: String::new(),
         };

crates/core/src/lib.rs

@@ -12,7 +12,7 @@ pub mod workflows;
 
 pub use app::CoreApp;
 pub use error::{CoreError, Result};
-pub use knowledge_base::{KbDocument, KbDocumentDetail, KbSearchHit, KnowledgeBase};
+pub use knowledge_base::{KbDocument, KbDocumentDetail, KbSearchHit, KbSource, KnowledgeBase};
 pub use models::ModelInfo;
 pub use agents::Agent;
 pub use scheduled_tasks::ScheduledTask;

crates/core/src/schema.sql

@@ -147,6 +147,8 @@ CREATE TABLE IF NOT EXISTS kb_documents (
     char_count  INTEGER NOT NULL DEFAULT 0,
     chunk_count INTEGER NOT NULL DEFAULT 0,
     content     TEXT NOT NULL DEFAULT '',
+    file_path   TEXT NOT NULL DEFAULT '',
+    source_id   TEXT,
     created_at  TEXT NOT NULL DEFAULT (datetime('now')),
     updated_at  TEXT NOT NULL DEFAULT (datetime('now'))
 );
@@ -160,6 +162,17 @@ CREATE TABLE IF NOT EXISTS kb_chunks (
     created_at  TEXT NOT NULL DEFAULT (datetime('now'))
 );
 
+CREATE TABLE IF NOT EXISTS kb_sources (
+    id         TEXT PRIMARY KEY,
+    kb_id      TEXT NOT NULL REFERENCES knowledge_bases(id) ON DELETE CASCADE,
+    path       TEXT NOT NULL,
+    extensions TEXT NOT NULL DEFAULT '',
+    recursive  INTEGER NOT NULL DEFAULT 1,
+    created_at TEXT NOT NULL DEFAULT (datetime('now'))
+);
+
+CREATE UNIQUE INDEX IF NOT EXISTS idx_kb_sources_kb_path ON kb_sources(kb_id, path);
+
 -- 全文检索索引（trigram 分词，对中文/短文本更友好；content='' 为 contentless-delete 模式）
 CREATE VIRTUAL TABLE IF NOT EXISTS kb_chunks_fts USING fts5(content, content = '', tokenize = 'trigram');
 

docs/ARCHITECTURE.md

@@ -12,7 +12,7 @@
 
 ```
 xianren_studio/
-├── apps/desktop/          Tauri 桌面壳（Rust 命令层、tauri.conf.json、推荐模型/预制智能体/预制工作流默认 JSON、工作流执行引擎 workflow.rs、邮件发送 mail.rs、文档提取 knowledge.rs）
+├── apps/desktop/          Tauri 桌面壳（Rust 命令层、tauri.conf.json、推荐模型/预制智能体/预制工作流默认 JSON、工作流执行引擎 workflow.rs、邮件发送 mail.rs、文档提取与目录扫描 knowledge.rs）
 ├── crates/core/           领域核心：SQLite（模型注册表、智能体、工作流、知识库、定时计划、会话、消息、设置）
 ├── crates/engine/         llama-server 生命周期 + 流式/非流式聊天（本地与远程 OpenAI 兼容）
 ├── crates/download/       分片断点续传下载器
@@ -111,6 +111,7 @@ ui/src/
 | 定时计划 | `list_scheduled_tasks`、`add_scheduled_task`、`update_scheduled_task`、`remove_scheduled_task`、`set_scheduled_task_enabled`、`run_scheduled_task_now` |
 | 邮件 | `mail_test`（用当前 SMTP 配置发送测试邮件） |
 | 知识库 | `list_knowledge_bases`、`add_knowledge_base`、`update_knowledge_base`、`remove_knowledge_base`、`list_kb_documents`、`get_kb_document`、`kb_import_documents`、`remove_kb_document`、`kb_rechunk_document`、`kb_search` |
+| 知识库·目录来源 | `kb_list_sources`、`kb_add_source`、`kb_scan_sources`、`kb_remove_source` |
 | 应用/设置 | `app_info`、`autostart_status`、`autostart_set`、`settings_get`、`settings_set` |
 | 模型 | `list_models`、`import_model`、`remove_model`、`set_model_enabled`、`scan_models`、`add_remote_model` |
 | 模型广场 | `search_models`、`list_model_files`、`list_recommended_models`、`import_recommendations`、`fetch_model_page` |
@@ -143,6 +144,7 @@ ui/src/
 - `send_task_email` / `load_smtp_config` / `split_recipients`：读取设置页「邮件」SMTP 配置、拆分多收件人并调用 `mail.rs` 发送。
 - `kb_import_documents`：base64 解码 → `knowledge.rs::extract_text` 提取文本（文本类直接 UTF-8，PDF 走 pdf-extract）→ `kb_db::chunk_text` 切块（按知识库 chunk_size/overlap）→ `insert_document_with_chunks` 事务写入文档与分块，逐文件返回成功/失败结果。
 - `kb_search` / `kb_rechunk_document`：检索走 `kb_db::search`（FTS5 trigram + BM25，短查询回退 LIKE）；重切按知识库最新分块设置重建该文档分块与索引。
+- `kb_add_source` / `kb_scan_sources` / `kb_remove_source`：目录来源注册与扫描——`knowledge.rs::collect_files` 按后缀过滤（空=全部）与递归开关收集文件，`classify` 分成文本 / 图片视频 / 音频三类（其他忽略），文本提取切块入库、媒体以文件名为内容登记，`kb_db::document_exists` 按（kb_id + file_path）去重；删除来源级联删除其导入文档。
 
 **邮件发送模块（`apps/desktop/src/mail.rs`）：**
 
@@ -185,13 +187,14 @@ ui/src/
 | `skills` | `name`、`description`、`content`、`enabled` | 技能工具库 |
 | `mcp_servers` | `name`、`description`、`url`、`auth_token`、`enabled` | MCP 服务配置 |
 | `knowledge_bases` | `name`、`description`、`chunk_size`、`chunk_overlap` | 知识库及分块设置 |
-| `kb_documents` | `kb_id`（级联删除）、`name`、`file_type`、`file_size`、`char_count`、`chunk_count`、`content`（提取后的纯文本） | 知识库文档；删除/重切分块由 FTS 触发器同步 |
+| `kb_documents` | `kb_id`（级联删除）、`name`、`file_type`、`file_size`、`char_count`、`chunk_count`、`content`（提取后的纯文本/媒体文件名）、`file_path`、`source_id` | 知识库文档；目录导入记录文件路径与来源，`file_path` 用于去重；删除/重切分块由 FTS 触发器同步 |
 | `kb_chunks` | `kb_id`、`document_id`（级联删除）、`seq`、`content` | 文档分块 |
 | `kb_chunks_fts` | FTS5 虚拟表（contentless-delete + trigram 分词） | 全文检索索引，由 `kb_chunks` 触发器维护 |
+| `kb_sources` | `kb_id`（级联删除）、`path`、`extensions`、`recursive`；`(kb_id, path)` 唯一 | 目录来源（多个）；添加时立即扫描，可重新扫描增量导入 |
 
 迁移清单（`ensure_column`）：`models.kind/base_url/api_key/api_model/enabled`、`messages.elapsed_ms/first_token_ms/images_json/model_id`、`conversations.pinned/favorite/tools_json/agent_id`。
 
-新增列迁移：`scheduled_tasks.email_enabled / email_to / email_mode / email_subject / email_body`（`ensure_column` 幂等补齐）。
+新增列迁移：`scheduled_tasks.email_enabled / email_to / email_mode / email_subject / email_body`、`kb_documents.file_path / source_id`（`ensure_column` 幂等补齐）。
 
 ### 5.3 `models.rs`
 
@@ -219,7 +222,7 @@ ui/src/
 
 ### 5.9 `knowledge_base.rs`
 
-知识库 CRUD（`list/get/insert/update/delete`）、文档 CRUD（`list/get/get_detail`）、`insert_document_with_chunks`（文档 + 分块同事务写入，FTS 触发器自动建索引）、`delete_document`（级联删除分块并同步索引）、`rechunk_document`、`chunk_text`（按字符切块 + 重叠）、`search`（FTS5 trigram BM25 排序，查询 <3 字或 MATCH 失败回退 LIKE）。
+知识库 CRUD（`list/get/insert/update/delete`）、文档 CRUD（`list/get/get_detail`）、`insert_document_with_chunks`（文档 + 分块同事务写入，FTS 触发器自动建索引）、`delete_document`（级联删除分块并同步索引）、`document_exists`（目录导入按文件路径去重）、`rechunk_document`、`chunk_text`（按字符切块 + 重叠）、`search`（FTS5 trigram BM25 排序，查询 <3 字或 MATCH 失败回退 LIKE）；目录来源 `list/get/upsert/delete_sources` 与 `delete_docs_by_source`。
 
 ---
 

docs/FEATURES.md

@@ -133,6 +133,11 @@
 - 左侧导航新增「知识库」选项卡，页面分左右两栏：左侧知识库列表，右侧选中库的管理区。
 - **知识库管理**：新建 / 编辑 / 删除知识库，字段含名称、描述、**分块大小**（100–10000 字，默认 500）与**块间重叠**（默认 50）；删除知识库会级联删除其文档与分块。
 - **文档导入**：支持多文件上传，格式覆盖文本类文件（txt / md / json / csv / 代码 / HTML 等）与 **PDF**（后端提取文本）；导入后按知识库分块设置自动切块并写入全文索引。
+- **目录来源（批量导入）**：可为知识库添加多个本地目录，扫描目录批量导入文档：
+  - **后缀过滤**：可指定只导入的后缀列表（如 `.txt,.md,.pdf`），留空默认全部；
+  - **递归子目录**：可开关是否递归扫描所有子目录；
+  - **类型过滤**：只收录**可读文本**（提取内容并分块检索）、**图片 / 视频**、**音频**三类文件（图片视频音频以文件名为内容登记，可按文件名检索），检测到其他类型自动忽略；
+  - 同一目录重复添加自动更新配置不产生重复来源；按文件路径去重，可一键「重新扫描」增量导入新增文件；删除目录来源会同时删除它导入的文档。
 - **文档操作**：文档列表展示类型 / 大小 / 字数 / 分块数；支持全文预览、「重切」（按最新分块设置重新切分，适用于修改设置后）与删除。
 - **检索**：全文检索基于 SQLite FTS5（trigram 分词，对中文友好），按 BM25 相关性排序；支持单库检索（也可全库检索），结果展示来源文档与分块序号，可一键复制分块内容用于对话。
 - 数据表：`knowledge_bases`、`kb_documents`（含提取后的纯文本）、`kb_chunks`（分块）、`kb_chunks_fts`（FTS5 全文索引，contentless-delete 模式 + 触发器同步）。
@@ -224,6 +229,7 @@
 
 ### 2026-08-17
 
+- 知识库新增「目录来源」：支持添加多个本地目录批量导入（后缀过滤默认全部、可开关递归子目录），只收录文本 / 图片视频 / 音频三类文件其余忽略，支持重新扫描增量导入与按来源删除。
 - 新增「知识库」选项卡与完整操作界面：知识库管理（新建/编辑/删除）、多格式文档上传（文本类 + PDF）、自动分块与全文检索（FTS5 trigram + BM25）、文档预览 / 重新切分 / 删除、检索结果一键复制。
 - 计划任务支持发送邮件：任务设置中可开启邮件并填写收件人，内容支持固定内容与大模型生成两种模式；设置页新增「邮件」SMTP 配置与测试发送。
 - 设置页新增「启动」：开机启动开关（写入 Windows 注册表启动项）；「自动加载模型」列表（多个本地模型按顺序在启动时逐个加载，最后一个保持运行）。

ui/src/api.ts

@@ -245,6 +245,28 @@ export interface KbImportResult {
   error: string | null;
 }
 
+export interface KbSource {
+  id: string;
+  kb_id: string;
+  path: string;
+  extensions: string;
+  recursive: boolean;
+  created_at: string;
+}
+
+export interface KbSourceInput {
+  kb_id: string;
+  path: string;
+  extensions: string;
+  recursive: boolean;
+}
+
+export interface KbSourceScanResult {
+  imported: number;
+  ignored: number;
+  errors: string[];
+}
+
 export interface McpServer {
   id: string;
   name: string;
@@ -483,6 +505,13 @@ export const api = {
     invoke<number>("kb_rechunk_document", { documentId }),
   kbSearch: (kbId: string | null, query: string, limit?: number) =>
     invoke<KbSearchHit[]>("kb_search", { kbId, query, limit: limit ?? 10 }),
+  kbListSources: (kbId: string) => invoke<KbSource[]>("kb_list_sources", { kbId }),
+  kbAddSource: (input: KbSourceInput) =>
+    invoke<KbSourceScanResult>("kb_add_source", { input }),
+  kbScanSources: (kbId: string) =>
+    invoke<KbSourceScanResult>("kb_scan_sources", { kbId }),
+  kbRemoveSource: (sourceId: string) =>
+    invoke<number>("kb_remove_source", { sourceId }),
   listModels: () => invoke<ModelInfo[]>("list_models"),
   importModel: (path: string) => invoke<ModelInfo>("import_model", { path }),
   removeModel: (id: string) => invoke<void>("remove_model", { id }),

ui/src/pages/KnowledgeBasePage.tsx

@@ -4,6 +4,8 @@ import {
   KbDocument,
   KbDocumentDetail,
   KbSearchHit,
+  KbSource,
+  KbSourceScanResult,
   KnowledgeBase,
   KnowledgeBaseInput,
 } from "../api";
@@ -56,6 +58,9 @@ export default function KnowledgeBasePage() {
   const [query, setQuery] = useState("");
   const [searching, setSearching] = useState(false);
   const [searchResults, setSearchResults] = useState<KbSearchHit[] | null>(null);
+  const [sources, setSources] = useState<KbSource[]>([]);
+  const [sourceModal, setSourceModal] = useState(false);
+  const [sourceMsg, setSourceMsg] = useState<string | null>(null);
   const uploadInputRef = useRef<HTMLInputElement>(null);
 
   const selected = useMemo(
@@ -88,6 +93,10 @@ export default function KnowledgeBasePage() {
       .listKbDocuments(selectedId)
       .then(setDocs)
       .catch((e) => setMsg(`加载文档失败：${String(e)}`));
+    api
+      .kbListSources(selectedId)
+      .then(setSources)
+      .catch((e) => setMsg(`加载目录来源失败：${String(e)}`));
   }, [selectedId]);
 
   async function handleSaveKb(input: KnowledgeBaseInput) {
@@ -189,6 +198,76 @@ export default function KnowledgeBasePage() {
     }
   }
 
+  async function refreshSourcesAndDocs() {
+    if (!selected) return;
+    const [d, s] = await Promise.all([
+      api.listKbDocuments(selected.id),
+      api.kbListSources(selected.id),
+    ]);
+    setDocs(d);
+    setSources(s);
+    await refreshKbs();
+  }
+
+  function formatScanMsg(label: string, r: KbSourceScanResult) {
+    const errPart =
+      r.errors.length > 0
+        ? `，${r.errors.length} 个错误（${r.errors
+            .slice(0, 3)
+            .join("；")}${r.errors.length > 3 ? "…" : ""}）`
+        : "";
+    return `${label}完成：导入 ${r.imported} 个，忽略 ${r.ignored} 个${errPart}`;
+  }
+
+  async function handleAddSource(path: string, extensions: string, recursive: boolean) {
+    if (!selected) return;
+    setBusy(true);
+    setSourceMsg(null);
+    try {
+      const r = await api.kbAddSource({
+        kb_id: selected.id,
+        path,
+        extensions,
+        recursive,
+      });
+      await refreshSourcesAndDocs();
+      setSourceMsg(formatScanMsg("添加目录", r));
+    } catch (e) {
+      setSourceMsg(`添加目录失败：${String(e)}`);
+    } finally {
+      setBusy(false);
+      setSourceModal(false);
+    }
+  }
+
+  async function handleScanAll() {
+    if (!selected) return;
+    setBusy(true);
+    setSourceMsg(null);
+    try {
+      const r = await api.kbScanSources(selected.id);
+      await refreshSourcesAndDocs();
+      setSourceMsg(formatScanMsg("重新扫描", r));
+    } catch (e) {
+      setSourceMsg(`重新扫描失败：${String(e)}`);
+    } finally {
+      setBusy(false);
+    }
+  }
+
+  async function handleRemoveSource(src: KbSource) {
+    if (!confirm(`确定删除目录来源「${src.path}」吗？将同时删除由它导入的文档。`)) return;
+    try {
+      const n = await api.kbRemoveSource(src.id);
+      setSourceMsg(`已删除目录来源及其 ${n} 个文档`);
+      if (selected) {
+        await refreshSourcesAndDocs();
+      }
+    } catch (e) {
+      setSourceMsg(`删除失败：${String(e)}`);
+    }
+  }
+
   return (
     <div className="flex h-full overflow-hidden">
       {/* 左侧：知识库列表 */}
@@ -343,6 +422,81 @@ export default function KnowledgeBasePage() {
               ) : null}
             </section>
 
+            {/* 目录来源 */}
+            <section className="mb-6 rounded-xl border border-border bg-panel p-4">
+              <div className="mb-2 flex items-center justify-between">
+                <div className="text-xs uppercase text-slate-500">
+                  目录来源（{sources.length}）
+                </div>
+                <div className="flex gap-2">
+                  <button
+                    className="btn-secondary !px-2.5 !py-1 text-xs"
+                    onClick={handleScanAll}
+                    disabled={busy || sources.length === 0}
+                  >
+                    <Icon name="refresh" className="mr-1 inline h-3 w-3" />
+                    重新扫描
+                  </button>
+                  <button
+                    className="btn-primary !px-2.5 !py-1 text-xs"
+                    onClick={() => setSourceModal(true)}
+                  >
+                    + 添加目录
+                  </button>
+                </div>
+              </div>
+              <p className="mb-3 text-[11px] text-slate-500">
+                按目录批量导入文档：只收录可读文本、图片视频、音频三类文件，其余自动忽略；可指定后缀过滤（默认全部），可选是否递归子目录，同一知识库可添加多个目录
+              </p>
+              {sourceMsg ? (
+                <div className="mb-2 rounded-lg border border-border bg-panel-2/60 px-3 py-2 text-xs text-slate-400">
+                  {sourceMsg}
+                </div>
+              ) : null}
+              {sources.length === 0 ? (
+                <div className="rounded-lg border border-dashed border-border bg-panel-2/40 px-3 py-6 text-center text-xs text-slate-500">
+                  还没有目录来源，点击「添加目录」批量导入文件夹中的文档
+                </div>
+              ) : (
+                <div className="space-y-2">
+                  {sources.map((src) => (
+                    <div
+                      key={src.id}
+                      className="flex items-center gap-3 rounded-lg border border-border bg-panel-2/60 p-3"
+                    >
+                      <Icon name="folder" className="h-4 w-4 shrink-0 text-slate-400" />
+                      <div className="min-w-0 flex-1">
+                        <div className="truncate text-xs font-medium text-slate-300">
+                          {src.path}
+                        </div>
+                        <div className="mt-0.5 flex items-center gap-1.5 text-[11px] text-slate-500">
+                          <span className="rounded bg-panel-2 px-1.5 py-0.5">
+                            {src.extensions.trim()
+                              ? src.extensions
+                                  .split(/[,;，；\s]+/)
+                                  .filter(Boolean)
+                                  .map((e) => (e.startsWith(".") ? e : `.${e}`))
+                                  .join(" ")
+                              : "全部后缀"}
+                          </span>
+                          <span className="rounded bg-panel-2 px-1.5 py-0.5">
+                            {src.recursive ? "递归子目录" : "仅当前目录"}
+                          </span>
+                        </div>
+                      </div>
+                      <button
+                        className="rounded-lg border border-border bg-panel-2 px-2.5 py-1.5 text-xs text-slate-400 hover:text-red-400"
+                        onClick={() => handleRemoveSource(src)}
+                        title="删除该目录来源及其导入的文档"
+                      >
+                        <Icon name="trash" className="h-3.5 w-3.5" />
+                      </button>
+                    </div>
+                  ))}
+                </div>
+              )}
+            </section>
+
             {/* 文档列表 */}
             <section>
               <div className="mb-2 text-xs uppercase text-slate-500">
@@ -381,6 +535,10 @@ export default function KnowledgeBasePage() {
         />
       ) : null}
 
+      {sourceModal ? (
+        <SourceModal busy={busy} onSave={handleAddSource} onClose={() => setSourceModal(false)} />
+      ) : null}
+
       {preview ? (
         <div
           className="fixed inset-0 z-50 flex items-center justify-center bg-black/70 p-6"
@@ -605,3 +763,107 @@ function KbModal({
     </div>
   );
 }
+
+function SourceModal({
+  busy,
+  onSave,
+  onClose,
+}: {
+  busy: boolean;
+  onSave: (path: string, extensions: string, recursive: boolean) => void;
+  onClose: () => void;
+}) {
+  const [path, setPath] = useState("");
+  const [extensions, setExtensions] = useState("");
+  const [recursive, setRecursive] = useState(true);
+  const valid = path.trim().length > 0;
+
+  return (
+    <div
+      className="fixed inset-0 z-50 flex items-center justify-center bg-black/70 p-6"
+      onClick={onClose}
+    >
+      <div
+        className="w-full max-w-md rounded-xl border border-border bg-panel"
+        onClick={(e) => e.stopPropagation()}
+      >
+        <div className="flex items-center justify-between border-b border-border px-5 py-3">
+          <span className="text-sm font-medium">添加目录来源</span>
+          <button className="text-slate-500 hover:text-slate-200" onClick={onClose}>
+            ×
+          </button>
+        </div>
+        <div className="space-y-4 px-5 py-4">
+          <div>
+            <label className="mb-1 block text-xs text-slate-400">目录路径</label>
+            <input
+              className="input"
+              value={path}
+              placeholder="例如：D:\资料\项目文档"
+              onChange={(e) => setPath(e.target.value)}
+            />
+          </div>
+          <div>
+            <label className="mb-1 block text-xs text-slate-400">后缀过滤（可选）</label>
+            <input
+              className="input"
+              value={extensions}
+              placeholder="留空表示全部；多个用逗号分隔，如 .txt,.md,.pdf,.png"
+              onChange={(e) => setExtensions(e.target.value)}
+            />
+          </div>
+          <div className="flex items-center gap-2">
+            <Toggle
+              checked={recursive}
+              onChange={(v) => setRecursive(v)}
+            />
+            <span className="text-xs text-slate-400">
+              {recursive ? "递归扫描所有子目录" : "仅扫描当前目录"}
+            </span>
+          </div>
+          <p className="text-[11px] text-slate-500">
+            导入时只收录可读文本、图片视频、音频三类文件，检测到其他类型会自动忽略；图片 / 视频 / 音频以文件名为内容登记，可按文件名检索。
+          </p>
+        </div>
+        <div className="flex justify-end gap-2 border-t border-border px-5 py-3">
+          <button className="btn-secondary" onClick={onClose} disabled={busy}>
+            取消
+          </button>
+          <button
+            className="btn-primary"
+            disabled={busy || !valid}
+            onClick={() => onSave(path.trim(), extensions.trim(), recursive)}
+          >
+            {busy ? "扫描中…" : "添加并扫描"}
+          </button>
+        </div>
+      </div>
+    </div>
+  );
+}
+
+function Toggle({
+  checked,
+  onChange,
+}: {
+  checked: boolean;
+  onChange: (v: boolean) => void;
+}) {
+  return (
+    <button
+      type="button"
+      role="switch"
+      aria-checked={checked}
+      onClick={() => onChange(!checked)}
+      className={`relative h-6 w-11 shrink-0 rounded-full transition-colors ${
+        checked ? "bg-accent" : "bg-panel-2"
+      }`}
+    >
+      <span
+        className={`absolute top-0.5 h-5 w-5 rounded-full bg-white shadow transition-all ${
+          checked ? "left-[22px]" : "left-0.5"
+        }`}
+      />
+    </button>
+  );
+}