Compare commits

...

8 Commits

Author SHA1 Message Date
iven
23c5bbdb40 feat(ai): 知识库 V2 Handler + 路由注册 + State 初始化
5 个端点:GET/POST /ai/knowledge-bases, GET/PUT/DELETE /ai/knowledge-bases/{id}
AiState 新增 knowledge_v2 字段,main.rs 初始化。
2026-05-26 23:25:38 +08:00
iven
2ccf0801b7 feat(ai): 新增 KnowledgeV2Service — 知识库 CRUD + 原子计数器
知识库列表/创建/更新/删除/单条查询,含分页和过滤。
increment_document_count / increment_chunk_count 使用 SQL 原子递增。
2026-05-26 23:19:09 +08:00
iven
86dbd74f3f feat(ai): 新增知识库 V2 Entity(bases/documents/chunks)
三张新表对应的 SeaORM Entity,embedding 字段标记 #[sea_orm(ignore)]
通过 raw SQL 操作 pgvector 列。
2026-05-26 23:13:42 +08:00
iven
0edb475638 feat(db): 创建 ai_knowledge_documents + ai_knowledge_chunks 表迁移
文档表支持多种来源(上传/URL/手动)、处理状态追踪、嵌入进度计数。
切片表含 vector(1536) 列 + HNSW 索引用于向量相似搜索。
2026-05-26 23:10:57 +08:00
iven
a7526455b4 feat(db): 创建 ai_knowledge_bases 表迁移
知识库 V2 第一张表:支持多类型知识库(规则/参考资料/临床指南/FAQ),
含意图关键词 JSONB、切片策略 JSONB、文档/切片计数器。
2026-05-26 23:07:28 +08:00
iven
dda8be9079 docs(ai): AI 知识库 V2 实施计划 — 7 Phase / 26 Tasks
Phase 0: 数据库迁移 + Entity (3 Tasks)
Phase 1: 后端 Service + Handler (4 Tasks)
Phase 2: 文档处理管线 (5 Tasks)
Phase 3: 混合意图路由 (3 Tasks)
Phase 4: 前端管理 UI (4 Tasks)
Phase 5: AI 客服集成 + 数据迁移 (4 Tasks)
Phase 6: 测试 + 验收 (3 Tasks)
2026-05-26 23:00:46 +08:00
iven
af2484e63b docs(ai): 知识库 V2 设计规格 — review 修复
修复 3 CRITICAL + 4 HIGH 问题:
- C1: 明确禁止 format! 拼接 SQL,所有 pgvector 查询参数化
- C2: 迁移 SQL 改用临时映射表精确关联,防数据重复/丢失
- C3: SSRF 防护细化(DNS rebinding + 超时 + 重定向校验)
- H1: document_count/chunk_count 改为原子 SQL 增量
- H2: embedding 部分失败:NULL embedding 写入 + embedded_count 统计
- H3: chunks 表补充 updated_at/updated_by 审计字段
- H4: 新增 validator crate 依赖说明
- M1: 空知识库边界返回空 context 不报错
- M2: 向量索引改用 HNSW(无需预热)
2026-05-26 22:18:28 +08:00
iven
10c28df152 docs(ai): AI 知识库 V2 设计规格 — 统一知识管理平台
替换旧三层模型(rules/references/guides)为 Dify 风格统一知识库:
- 3 张新表:knowledge_bases / documents / chunks(切片+向量)
- 文档处理管线:PDF/Word/Excel/URL/Markdown/手动录入 → 智能切片 → embedding
- 混合意图路由:关键词粗筛 → 向量检索 → LLM 兜底
- AI 客服集成:RAG context 注入 + [ref:xxx] 引用溯源
2026-05-26 22:11:29 +08:00
16 changed files with 3412 additions and 0 deletions

View File

@@ -0,0 +1,31 @@
use sea_orm::entity::prelude::*;
use serde::{Deserialize, Serialize};
#[derive(Clone, Debug, PartialEq, DeriveEntityModel, Serialize, Deserialize)]
#[sea_orm(table_name = "ai_knowledge_bases")]
pub struct Model {
#[sea_orm(primary_key, auto_increment = false)]
pub id: Uuid,
pub tenant_id: Uuid,
pub name: String,
pub kb_type: String,
pub description: Option<String>,
pub icon: Option<String>,
pub chunk_strategy: serde_json::Value,
pub intent_keywords: serde_json::Value,
pub embedding_model: Option<String>,
pub is_enabled: bool,
pub document_count: i32,
pub chunk_count: i32,
pub created_at: DateTimeUtc,
pub updated_at: DateTimeUtc,
pub created_by: Option<Uuid>,
pub updated_by: Option<Uuid>,
pub deleted_at: Option<DateTimeUtc>,
pub version_lock: i32,
}
#[derive(Copy, Clone, Debug, EnumIter, DeriveRelation)]
pub enum Relation {}
impl ActiveModelBehavior for ActiveModel {}

View File

@@ -0,0 +1,34 @@
use sea_orm::entity::prelude::*;
use serde::{Deserialize, Serialize};
#[derive(Clone, Debug, PartialEq, DeriveEntityModel, Serialize, Deserialize)]
#[sea_orm(table_name = "ai_knowledge_chunks")]
pub struct Model {
#[sea_orm(primary_key, auto_increment = false)]
pub id: Uuid,
pub tenant_id: Uuid,
pub knowledge_base_id: Uuid,
pub document_id: Uuid,
pub chunk_index: i32,
pub content: String,
pub token_count: Option<i32>,
// pgvector 字段 — SeaORM 不原生支持 vector 类型,查询时用 raw SQL
#[sea_orm(ignore)]
pub embedding: Option<Vec<f32>>,
pub start_offset: Option<i32>,
pub end_offset: Option<i32>,
pub page_number: Option<i32>,
pub metadata: serde_json::Value,
pub hit_count: i32,
pub last_hit_at: Option<DateTimeUtc>,
pub created_at: DateTimeUtc,
pub updated_at: DateTimeUtc,
pub created_by: Option<Uuid>,
pub updated_by: Option<Uuid>,
pub deleted_at: Option<DateTimeUtc>,
}
#[derive(Copy, Clone, Debug, EnumIter, DeriveRelation)]
pub enum Relation {}
impl ActiveModelBehavior for ActiveModel {}

View File

@@ -0,0 +1,36 @@
use sea_orm::entity::prelude::*;
use serde::{Deserialize, Serialize};
#[derive(Clone, Debug, PartialEq, DeriveEntityModel, Serialize, Deserialize)]
#[sea_orm(table_name = "ai_knowledge_documents")]
pub struct Model {
#[sea_orm(primary_key, auto_increment = false)]
pub id: Uuid,
pub tenant_id: Uuid,
pub knowledge_base_id: Uuid,
pub title: String,
pub doc_type: String,
pub source_type: String,
pub source_url: Option<String>,
pub file_name: Option<String>,
pub file_size: Option<i64>,
pub file_mime_type: Option<String>,
pub content: Option<String>,
pub status: String,
pub chunk_count: i32,
pub embedded_count: i32,
pub error_message: Option<String>,
pub processing_started_at: Option<DateTimeUtc>,
pub processing_completed_at: Option<DateTimeUtc>,
pub created_at: DateTimeUtc,
pub updated_at: DateTimeUtc,
pub created_by: Option<Uuid>,
pub updated_by: Option<Uuid>,
pub deleted_at: Option<DateTimeUtc>,
pub version_lock: i32,
}
#[derive(Copy, Clone, Debug, EnumIter, DeriveRelation)]
pub enum Relation {}
impl ActiveModelBehavior for ActiveModel {}

View File

@@ -3,6 +3,9 @@ pub mod ai_analysis_queue;
pub mod ai_chat_message;
pub mod ai_chat_session;
pub mod ai_feature_flags;
pub mod ai_knowledge_bases;
pub mod ai_knowledge_chunks;
pub mod ai_knowledge_documents;
pub mod ai_knowledge_guides;
pub mod ai_knowledge_references;
pub mod ai_knowledge_rules;

View File

@@ -0,0 +1,172 @@
use axum::Json;
use axum::extract::{Extension, FromRef, Path, State};
use erp_core::rbac::require_permission;
use erp_core::types::{ApiResponse, TenantContext};
use serde::Deserialize;
use crate::service::knowledge_v2::{
CreateKnowledgeBaseReq, ListKnowledgeBasesQuery, UpdateKnowledgeBaseReq,
};
use crate::state::AiState;
#[derive(Debug, Deserialize)]
pub struct ListKnowledgeBasesParams {
pub kb_type: Option<String>,
pub is_enabled: Option<bool>,
pub page: Option<u64>,
pub page_size: Option<u64>,
}
#[utoipa::path(
get,
path = "/ai/knowledge-bases",
responses((status = 200, description = "知识库列表")),
tag = "知识库V2",
security(("bearer_auth" = [])),
)]
pub async fn list_knowledge_bases<S>(
State(state): State<AiState>,
Extension(ctx): Extension<TenantContext>,
axum::extract::Query(params): axum::extract::Query<ListKnowledgeBasesParams>,
) -> Result<Json<ApiResponse<serde_json::Value>>, erp_core::error::AppError>
where
AiState: FromRef<S>,
S: Clone + Send + Sync + 'static,
{
require_permission(&ctx, "ai.knowledge.list")?;
let query = ListKnowledgeBasesQuery {
kb_type: params.kb_type,
is_enabled: params.is_enabled,
page: params.page,
page_size: params.page_size,
};
let (items, total) = state.knowledge_v2.list(ctx.tenant_id, &query).await?;
Ok(Json(ApiResponse::ok(serde_json::json!({
"data": items,
"total": total,
"page": query.page.unwrap_or(1),
"page_size": query.page_size.unwrap_or(20),
}))))
}
#[utoipa::path(
get,
path = "/ai/knowledge-bases/{id}",
responses((status = 200, description = "知识库详情")),
tag = "知识库V2",
security(("bearer_auth" = [])),
)]
pub async fn get_knowledge_base<S>(
State(state): State<AiState>,
Extension(ctx): Extension<TenantContext>,
Path(id): Path<uuid::Uuid>,
) -> Result<Json<ApiResponse<serde_json::Value>>, erp_core::error::AppError>
where
AiState: FromRef<S>,
S: Clone + Send + Sync + 'static,
{
require_permission(&ctx, "ai.knowledge.list")?;
let kb = state.knowledge_v2.get_by_id(ctx.tenant_id, id).await?;
Ok(Json(ApiResponse::ok(
serde_json::to_value(&kb).unwrap_or_default(),
)))
}
#[utoipa::path(
post,
path = "/ai/knowledge-bases",
request_body = CreateKnowledgeBaseReq,
responses((status = 200, description = "创建知识库")),
tag = "知识库V2",
security(("bearer_auth" = [])),
)]
pub async fn create_knowledge_base<S>(
State(state): State<AiState>,
Extension(ctx): Extension<TenantContext>,
Json(body): Json<CreateKnowledgeBaseReq>,
) -> Result<Json<ApiResponse<serde_json::Value>>, erp_core::error::AppError>
where
AiState: FromRef<S>,
S: Clone + Send + Sync + 'static,
{
require_permission(&ctx, "ai.knowledge.manage")?;
if body.name.trim().is_empty() {
return Err(erp_core::error::AppError::Validation(
"知识库名称不能为空".into(),
));
}
if body.kb_type.trim().is_empty() {
return Err(erp_core::error::AppError::Validation(
"知识库类型不能为空".into(),
));
}
let id = state
.knowledge_v2
.create(ctx.tenant_id, ctx.user_id, body)
.await?;
Ok(Json(ApiResponse::ok(serde_json::json!({ "id": id }))))
}
#[utoipa::path(
put,
path = "/ai/knowledge-bases/{id}",
request_body = UpdateKnowledgeBaseReq,
responses((status = 200, description = "更新知识库")),
tag = "知识库V2",
security(("bearer_auth" = [])),
)]
pub async fn update_knowledge_base<S>(
State(state): State<AiState>,
Extension(ctx): Extension<TenantContext>,
Path(id): Path<uuid::Uuid>,
Json(body): Json<UpdateKnowledgeBaseReq>,
) -> Result<Json<ApiResponse<serde_json::Value>>, erp_core::error::AppError>
where
AiState: FromRef<S>,
S: Clone + Send + Sync + 'static,
{
require_permission(&ctx, "ai.knowledge.manage")?;
if let Some(ref name) = body.name
&& name.trim().is_empty()
{
return Err(erp_core::error::AppError::Validation(
"知识库名称不能为空".into(),
));
}
state
.knowledge_v2
.update(ctx.tenant_id, ctx.user_id, id, body)
.await?;
Ok(Json(ApiResponse::ok(serde_json::json!({ "id": id }))))
}
#[utoipa::path(
delete,
path = "/ai/knowledge-bases/{id}",
responses((status = 200, description = "删除知识库")),
tag = "知识库V2",
security(("bearer_auth" = [])),
)]
pub async fn delete_knowledge_base<S>(
State(state): State<AiState>,
Extension(ctx): Extension<TenantContext>,
Path(id): Path<uuid::Uuid>,
) -> Result<Json<ApiResponse<serde_json::Value>>, erp_core::error::AppError>
where
AiState: FromRef<S>,
S: Clone + Send + Sync + 'static,
{
require_permission(&ctx, "ai.knowledge.manage")?;
state.knowledge_v2.delete(ctx.tenant_id, id).await?;
Ok(Json(ApiResponse::ok(serde_json::json!({ "id": id }))))
}

View File

@@ -16,6 +16,7 @@ pub mod chat_handler;
pub mod config_handler;
pub mod insight_handler;
pub mod knowledge_handler;
pub mod knowledge_v2_handler;
pub mod risk_handler;
pub mod rule_handler;
pub mod suggestion_handler;

View File

@@ -588,6 +588,27 @@ impl AiModule {
"/ai/knowledge/guides/{id}/re-embed",
axum::routing::post(crate::handler::knowledge_handler::re_embed_guide),
)
// 知识库 V2 路由
.route(
"/ai/knowledge-bases",
axum::routing::get(crate::handler::knowledge_v2_handler::list_knowledge_bases),
)
.route(
"/ai/knowledge-bases",
axum::routing::post(crate::handler::knowledge_v2_handler::create_knowledge_base),
)
.route(
"/ai/knowledge-bases/{id}",
axum::routing::get(crate::handler::knowledge_v2_handler::get_knowledge_base),
)
.route(
"/ai/knowledge-bases/{id}",
axum::routing::put(crate::handler::knowledge_v2_handler::update_knowledge_base),
)
.route(
"/ai/knowledge-bases/{id}",
axum::routing::delete(crate::handler::knowledge_v2_handler::delete_knowledge_base),
)
.route(
"/ai/dialysis/risk-assessment",
axum::routing::post(crate::handler::assess_dialysis_risk),

View File

@@ -0,0 +1,258 @@
use sea_orm::{
ColumnTrait, ConnectionTrait, EntityTrait, PaginatorTrait, QueryFilter, QueryOrder, Set,
};
use serde::{Deserialize, Serialize};
use uuid::Uuid;
use crate::entity::ai_knowledge_bases;
use crate::error::{AiError, AiResult};
// ─── DTO ───
#[derive(Debug, Deserialize, Serialize, utoipa::ToSchema)]
pub struct CreateKnowledgeBaseReq {
pub name: String,
pub kb_type: String,
pub description: Option<String>,
pub icon: Option<String>,
pub chunk_strategy: Option<serde_json::Value>,
pub intent_keywords: Option<serde_json::Value>,
pub embedding_model: Option<String>,
pub is_enabled: Option<bool>,
}
#[derive(Debug, Deserialize, Serialize, utoipa::ToSchema)]
pub struct UpdateKnowledgeBaseReq {
pub name: Option<String>,
pub kb_type: Option<String>,
pub description: Option<String>,
pub icon: Option<String>,
pub chunk_strategy: Option<serde_json::Value>,
pub intent_keywords: Option<serde_json::Value>,
pub embedding_model: Option<String>,
pub is_enabled: Option<bool>,
}
#[derive(Debug, Deserialize, utoipa::IntoParams)]
pub struct ListKnowledgeBasesQuery {
pub kb_type: Option<String>,
pub is_enabled: Option<bool>,
pub page: Option<u64>,
pub page_size: Option<u64>,
}
// ─── Service ───
pub struct KnowledgeV2Service {
db: sea_orm::DatabaseConnection,
}
impl KnowledgeV2Service {
pub fn new(db: sea_orm::DatabaseConnection) -> Self {
Self { db }
}
pub async fn list(
&self,
tenant_id: Uuid,
query: &ListKnowledgeBasesQuery,
) -> AiResult<(Vec<ai_knowledge_bases::Model>, u64)> {
let page = query.page.unwrap_or(1);
let page_size = query.page_size.unwrap_or(20);
let mut find = ai_knowledge_bases::Entity::find()
.filter(ai_knowledge_bases::Column::TenantId.eq(tenant_id))
.filter(ai_knowledge_bases::Column::DeletedAt.is_null());
if let Some(ref kb_type) = query.kb_type {
find = find.filter(ai_knowledge_bases::Column::KbType.eq(kb_type.as_str()));
}
if let Some(is_enabled) = query.is_enabled {
find = find.filter(ai_knowledge_bases::Column::IsEnabled.eq(is_enabled));
}
let paginator = find
.order_by_desc(ai_knowledge_bases::Column::CreatedAt)
.paginate(&self.db, page_size);
let total = paginator.num_items().await?;
let items = paginator.fetch_page(page - 1).await?;
Ok((items, total))
}
pub async fn get_by_id(
&self,
tenant_id: Uuid,
id: Uuid,
) -> AiResult<ai_knowledge_bases::Model> {
ai_knowledge_bases::Entity::find_by_id(id)
.one(&self.db)
.await
.map_err(|e| AiError::DbError(e.to_string()))?
.filter(|m| m.tenant_id == tenant_id && m.deleted_at.is_none())
.ok_or_else(|| AiError::KnowledgeError("知识库不存在".into()))
}
pub async fn create(
&self,
tenant_id: Uuid,
user_id: Uuid,
req: CreateKnowledgeBaseReq,
) -> AiResult<Uuid> {
let id = Uuid::now_v7();
let now = chrono::Utc::now();
let active = ai_knowledge_bases::ActiveModel {
id: Set(id),
tenant_id: Set(tenant_id),
name: Set(req.name),
kb_type: Set(req.kb_type),
description: Set(req.description),
icon: Set(req.icon),
chunk_strategy: Set(req.chunk_strategy.unwrap_or(
serde_json::json!({"strategy": "auto", "chunk_size": 500, "overlap": 50}),
)),
intent_keywords: Set(req.intent_keywords.unwrap_or(serde_json::json!([]))),
embedding_model: Set(req.embedding_model),
is_enabled: Set(req.is_enabled.unwrap_or(true)),
document_count: Set(0),
chunk_count: Set(0),
created_at: Set(now),
updated_at: Set(now),
created_by: Set(Some(user_id)),
updated_by: Set(Some(user_id)),
deleted_at: Set(None),
version_lock: Set(1),
};
ai_knowledge_bases::Entity::insert(active)
.exec(&self.db)
.await
.map_err(|e| AiError::DbError(e.to_string()))?;
Ok(id)
}
pub async fn update(
&self,
tenant_id: Uuid,
user_id: Uuid,
id: Uuid,
req: UpdateKnowledgeBaseReq,
) -> AiResult<()> {
let existing = self.get_by_id(tenant_id, id).await?;
let now = chrono::Utc::now();
let active = ai_knowledge_bases::ActiveModel {
id: Set(existing.id),
tenant_id: Set(existing.tenant_id),
name: Set(req.name.unwrap_or(existing.name)),
kb_type: Set(req.kb_type.unwrap_or(existing.kb_type)),
description: Set(req.description.or(existing.description)),
icon: Set(req.icon.or(existing.icon)),
chunk_strategy: Set(req.chunk_strategy.unwrap_or(existing.chunk_strategy)),
intent_keywords: Set(req.intent_keywords.unwrap_or(existing.intent_keywords)),
embedding_model: Set(req.embedding_model.or(existing.embedding_model)),
is_enabled: Set(req.is_enabled.unwrap_or(existing.is_enabled)),
document_count: Set(existing.document_count),
chunk_count: Set(existing.chunk_count),
created_at: Set(existing.created_at),
updated_at: Set(now),
created_by: Set(existing.created_by),
updated_by: Set(Some(user_id)),
deleted_at: Set(existing.deleted_at),
version_lock: Set(existing.version_lock + 1),
};
ai_knowledge_bases::Entity::update(active)
.exec(&self.db)
.await
.map_err(|e| AiError::DbError(e.to_string()))?;
Ok(())
}
pub async fn delete(&self, tenant_id: Uuid, id: Uuid) -> AiResult<()> {
let existing = self.get_by_id(tenant_id, id).await?;
let now = chrono::Utc::now();
let active = ai_knowledge_bases::ActiveModel {
id: Set(existing.id),
tenant_id: Set(existing.tenant_id),
name: Set(existing.name),
kb_type: Set(existing.kb_type),
description: Set(existing.description),
icon: Set(existing.icon),
chunk_strategy: Set(existing.chunk_strategy),
intent_keywords: Set(existing.intent_keywords),
embedding_model: Set(existing.embedding_model),
is_enabled: Set(existing.is_enabled),
document_count: Set(existing.document_count),
chunk_count: Set(existing.chunk_count),
created_at: Set(existing.created_at),
updated_at: Set(now),
created_by: Set(existing.created_by),
updated_by: Set(existing.updated_by),
deleted_at: Set(Some(now)),
version_lock: Set(existing.version_lock + 1),
};
ai_knowledge_bases::Entity::update(active)
.exec(&self.db)
.await
.map_err(|e| AiError::DbError(e.to_string()))?;
Ok(())
}
/// 原子递增文档计数(用于文档上传成功后)
pub async fn increment_document_count(&self, kb_id: Uuid, delta: i32) -> AiResult<()> {
let sql = r#"
UPDATE ai_knowledge_bases
SET document_count = document_count + $2,
updated_at = $3,
version_lock = version_lock + 1
WHERE id = $1 AND deleted_at IS NULL
"#;
let stmt = sea_orm::Statement::from_sql_and_values(
sea_orm::DatabaseBackend::Postgres,
sql,
[
sea_orm::Value::from(kb_id),
sea_orm::Value::from(delta),
sea_orm::Value::from(chrono::Utc::now()),
],
);
self.db
.execute(stmt)
.await
.map_err(|e| AiError::DbError(e.to_string()))?;
Ok(())
}
/// 原子递增切片计数(用于切片生成后)
pub async fn increment_chunk_count(&self, kb_id: Uuid, delta: i32) -> AiResult<()> {
let sql = r#"
UPDATE ai_knowledge_bases
SET chunk_count = chunk_count + $2,
updated_at = $3,
version_lock = version_lock + 1
WHERE id = $1 AND deleted_at IS NULL
"#;
let stmt = sea_orm::Statement::from_sql_and_values(
sea_orm::DatabaseBackend::Postgres,
sql,
[
sea_orm::Value::from(kb_id),
sea_orm::Value::from(delta),
sea_orm::Value::from(chrono::Utc::now()),
],
);
self.db
.execute(stmt)
.await
.map_err(|e| AiError::DbError(e.to_string()))?;
Ok(())
}
}

View File

@@ -11,6 +11,7 @@ pub mod embedding;
pub mod feature_flag_service;
pub mod insight_service;
pub mod knowledge;
pub mod knowledge_v2;
pub mod local_rules;
pub mod output_parser;
pub mod post_process;

View File

@@ -12,6 +12,7 @@ use crate::service::chat_session::ChatSessionService;
use crate::service::feature_flag_service::FeatureFlagService;
use crate::service::insight_service::InsightService;
use crate::service::knowledge::KnowledgeService;
use crate::service::knowledge_v2::KnowledgeV2Service;
use crate::service::prompt::PromptService;
use crate::service::quota::QuotaService;
use crate::service::risk_service::RiskService;
@@ -34,6 +35,7 @@ pub struct AiState {
pub insight_service: Arc<InsightService>,
pub feature_flags: Arc<FeatureFlagService>,
pub knowledge: Arc<KnowledgeService>,
pub knowledge_v2: Arc<KnowledgeV2Service>,
pub chat_session: Arc<ChatSessionService>,
pub chat_message: Arc<ChatMessageService>,
}

View File

@@ -172,6 +172,8 @@ mod m20260522_000162_seed_patient_miniprogram_permissions;
mod m20260526_000163_points_rule_unique_event_type;
mod m20260526_000164_ai_prompt_add_analysis_type;
mod m20260526_000165_ai_prompt_fix_analysis_type;
mod m20260526_000166_create_ai_knowledge_bases;
mod m20260526_000167_create_ai_knowledge_documents;
pub struct Migrator;
@@ -351,6 +353,8 @@ impl MigratorTrait for Migrator {
Box::new(m20260526_000163_points_rule_unique_event_type::Migration),
Box::new(m20260526_000164_ai_prompt_add_analysis_type::Migration),
Box::new(m20260526_000165_ai_prompt_fix_analysis_type::Migration),
Box::new(m20260526_000166_create_ai_knowledge_bases::Migration),
Box::new(m20260526_000167_create_ai_knowledge_documents::Migration),
]
}
}

View File

@@ -0,0 +1,141 @@
use sea_orm_migration::prelude::*;
#[derive(DeriveMigrationName)]
pub struct Migration;
#[derive(Iden)]
enum AiKnowledgeBases {
Table,
Id,
TenantId,
Name,
KbType,
Description,
Icon,
ChunkStrategy,
IntentKeywords,
EmbeddingModel,
IsEnabled,
DocumentCount,
ChunkCount,
CreatedAt,
UpdatedAt,
CreatedBy,
UpdatedBy,
DeletedAt,
VersionLock,
}
#[async_trait::async_trait]
impl MigrationTrait for Migration {
async fn up(&self, manager: &SchemaManager) -> Result<(), DbErr> {
manager
.create_table(
Table::create()
.table(AiKnowledgeBases::Table)
.if_not_exists()
.col(
ColumnDef::new(AiKnowledgeBases::Id)
.uuid()
.not_null()
.primary_key(),
)
.col(ColumnDef::new(AiKnowledgeBases::TenantId).uuid().not_null())
.col(
ColumnDef::new(AiKnowledgeBases::Name)
.string_len(200)
.not_null(),
)
.col(
ColumnDef::new(AiKnowledgeBases::KbType)
.string_len(50)
.not_null(),
)
.col(ColumnDef::new(AiKnowledgeBases::Description).text())
.col(ColumnDef::new(AiKnowledgeBases::Icon).string_len(50))
.col(
ColumnDef::new(AiKnowledgeBases::ChunkStrategy)
.json_binary()
.not_null()
.default(Expr::cust("'{}'")),
)
.col(
ColumnDef::new(AiKnowledgeBases::IntentKeywords)
.json_binary()
.not_null()
.default(Expr::cust("'[]'")),
)
.col(ColumnDef::new(AiKnowledgeBases::EmbeddingModel).string_len(100))
.col(
ColumnDef::new(AiKnowledgeBases::IsEnabled)
.boolean()
.not_null()
.default(true),
)
.col(
ColumnDef::new(AiKnowledgeBases::DocumentCount)
.integer()
.not_null()
.default(0),
)
.col(
ColumnDef::new(AiKnowledgeBases::ChunkCount)
.integer()
.not_null()
.default(0),
)
.col(
ColumnDef::new(AiKnowledgeBases::CreatedAt)
.timestamp_with_time_zone()
.not_null()
.default(Expr::current_timestamp()),
)
.col(
ColumnDef::new(AiKnowledgeBases::UpdatedAt)
.timestamp_with_time_zone()
.not_null()
.default(Expr::current_timestamp()),
)
.col(ColumnDef::new(AiKnowledgeBases::CreatedBy).uuid())
.col(ColumnDef::new(AiKnowledgeBases::UpdatedBy).uuid())
.col(ColumnDef::new(AiKnowledgeBases::DeletedAt).timestamp_with_time_zone())
.col(
ColumnDef::new(AiKnowledgeBases::VersionLock)
.integer()
.not_null()
.default(1),
)
.to_owned(),
)
.await?;
manager
.create_index(
Index::create()
.if_not_exists()
.name("idx_kb_tenant")
.table(AiKnowledgeBases::Table)
.col(AiKnowledgeBases::TenantId)
.to_owned(),
)
.await?;
manager
.create_index(
Index::create()
.if_not_exists()
.name("idx_kb_type")
.table(AiKnowledgeBases::Table)
.col(AiKnowledgeBases::TenantId)
.col(AiKnowledgeBases::KbType)
.to_owned(),
)
.await
}
async fn down(&self, manager: &SchemaManager) -> Result<(), DbErr> {
manager
.drop_table(Table::drop().table(AiKnowledgeBases::Table).to_owned())
.await
}
}

View File

@@ -0,0 +1,297 @@
use sea_orm_migration::prelude::*;
#[derive(DeriveMigrationName)]
pub struct Migration;
#[derive(Iden)]
enum AiKnowledgeDocuments {
Table,
Id,
TenantId,
KnowledgeBaseId,
Title,
DocType,
SourceType,
SourceUrl,
FileName,
FileSize,
FileMimeType,
Content,
Status,
ChunkCount,
EmbeddedCount,
ErrorMessage,
ProcessingStartedAt,
ProcessingCompletedAt,
CreatedAt,
UpdatedAt,
CreatedBy,
UpdatedBy,
DeletedAt,
VersionLock,
}
#[derive(Iden)]
enum AiKnowledgeChunks {
Table,
Id,
TenantId,
KnowledgeBaseId,
DocumentId,
ChunkIndex,
Content,
TokenCount,
StartOffset,
EndOffset,
PageNumber,
Metadata,
HitCount,
LastHitAt,
CreatedAt,
UpdatedAt,
CreatedBy,
UpdatedBy,
DeletedAt,
}
#[async_trait::async_trait]
impl MigrationTrait for Migration {
async fn up(&self, manager: &SchemaManager) -> Result<(), DbErr> {
// ai_knowledge_documents
manager
.create_table(
Table::create()
.table(AiKnowledgeDocuments::Table)
.if_not_exists()
.col(
ColumnDef::new(AiKnowledgeDocuments::Id)
.uuid()
.not_null()
.primary_key(),
)
.col(
ColumnDef::new(AiKnowledgeDocuments::TenantId)
.uuid()
.not_null(),
)
.col(
ColumnDef::new(AiKnowledgeDocuments::KnowledgeBaseId)
.uuid()
.not_null(),
)
.col(
ColumnDef::new(AiKnowledgeDocuments::Title)
.string_len(500)
.not_null(),
)
.col(
ColumnDef::new(AiKnowledgeDocuments::DocType)
.string_len(30)
.not_null()
.default("manual"),
)
.col(
ColumnDef::new(AiKnowledgeDocuments::SourceType)
.string_len(30)
.not_null()
.default("manual"),
)
.col(ColumnDef::new(AiKnowledgeDocuments::SourceUrl).text())
.col(ColumnDef::new(AiKnowledgeDocuments::FileName).string_len(500))
.col(ColumnDef::new(AiKnowledgeDocuments::FileSize).big_integer())
.col(ColumnDef::new(AiKnowledgeDocuments::FileMimeType).string_len(100))
.col(ColumnDef::new(AiKnowledgeDocuments::Content).text())
.col(
ColumnDef::new(AiKnowledgeDocuments::Status)
.string_len(20)
.not_null()
.default("pending"),
)
.col(
ColumnDef::new(AiKnowledgeDocuments::ChunkCount)
.integer()
.not_null()
.default(0),
)
.col(
ColumnDef::new(AiKnowledgeDocuments::EmbeddedCount)
.integer()
.not_null()
.default(0),
)
.col(ColumnDef::new(AiKnowledgeDocuments::ErrorMessage).text())
.col(
ColumnDef::new(AiKnowledgeDocuments::ProcessingStartedAt)
.timestamp_with_time_zone(),
)
.col(
ColumnDef::new(AiKnowledgeDocuments::ProcessingCompletedAt)
.timestamp_with_time_zone(),
)
.col(
ColumnDef::new(AiKnowledgeDocuments::CreatedAt)
.timestamp_with_time_zone()
.not_null()
.default(Expr::current_timestamp()),
)
.col(
ColumnDef::new(AiKnowledgeDocuments::UpdatedAt)
.timestamp_with_time_zone()
.not_null()
.default(Expr::current_timestamp()),
)
.col(ColumnDef::new(AiKnowledgeDocuments::CreatedBy).uuid())
.col(ColumnDef::new(AiKnowledgeDocuments::UpdatedBy).uuid())
.col(ColumnDef::new(AiKnowledgeDocuments::DeletedAt).timestamp_with_time_zone())
.col(
ColumnDef::new(AiKnowledgeDocuments::VersionLock)
.integer()
.not_null()
.default(1),
)
.to_owned(),
)
.await?;
manager
.create_index(
Index::create()
.if_not_exists()
.name("idx_doc_kb")
.table(AiKnowledgeDocuments::Table)
.col(AiKnowledgeDocuments::KnowledgeBaseId)
.to_owned(),
)
.await?;
manager
.create_index(
Index::create()
.if_not_exists()
.name("idx_doc_status")
.table(AiKnowledgeDocuments::Table)
.col(AiKnowledgeDocuments::KnowledgeBaseId)
.col(AiKnowledgeDocuments::Status)
.to_owned(),
)
.await?;
// ai_knowledge_chunks
manager
.create_table(
Table::create()
.table(AiKnowledgeChunks::Table)
.if_not_exists()
.col(
ColumnDef::new(AiKnowledgeChunks::Id)
.uuid()
.not_null()
.primary_key(),
)
.col(
ColumnDef::new(AiKnowledgeChunks::TenantId)
.uuid()
.not_null(),
)
.col(
ColumnDef::new(AiKnowledgeChunks::KnowledgeBaseId)
.uuid()
.not_null(),
)
.col(
ColumnDef::new(AiKnowledgeChunks::DocumentId)
.uuid()
.not_null(),
)
.col(
ColumnDef::new(AiKnowledgeChunks::ChunkIndex)
.integer()
.not_null(),
)
.col(ColumnDef::new(AiKnowledgeChunks::Content).text().not_null())
.col(ColumnDef::new(AiKnowledgeChunks::TokenCount).integer())
// embedding is vector(1536) — added via raw SQL below
.col(ColumnDef::new(AiKnowledgeChunks::StartOffset).integer())
.col(ColumnDef::new(AiKnowledgeChunks::EndOffset).integer())
.col(ColumnDef::new(AiKnowledgeChunks::PageNumber).integer())
.col(
ColumnDef::new(AiKnowledgeChunks::Metadata)
.json_binary()
.not_null()
.default(Expr::cust("'{}'")),
)
.col(
ColumnDef::new(AiKnowledgeChunks::HitCount)
.integer()
.not_null()
.default(0),
)
.col(ColumnDef::new(AiKnowledgeChunks::LastHitAt).timestamp_with_time_zone())
.col(
ColumnDef::new(AiKnowledgeChunks::CreatedAt)
.timestamp_with_time_zone()
.not_null()
.default(Expr::current_timestamp()),
)
.col(
ColumnDef::new(AiKnowledgeChunks::UpdatedAt)
.timestamp_with_time_zone()
.not_null()
.default(Expr::current_timestamp()),
)
.col(ColumnDef::new(AiKnowledgeChunks::CreatedBy).uuid())
.col(ColumnDef::new(AiKnowledgeChunks::UpdatedBy).uuid())
.col(ColumnDef::new(AiKnowledgeChunks::DeletedAt).timestamp_with_time_zone())
.to_owned(),
)
.await?;
// Add embedding column as vector type (SeaORM doesn't support this natively)
manager
.get_connection()
.execute_unprepared(
"ALTER TABLE ai_knowledge_chunks ADD COLUMN IF NOT EXISTS embedding vector(1536)",
)
.await?;
// HNSW index for vector similarity search
manager
.get_connection()
.execute_unprepared(
"CREATE INDEX IF NOT EXISTS idx_chunk_embedding ON ai_knowledge_chunks USING hnsw (embedding vector_cosine_ops)",
)
.await?;
manager
.create_index(
Index::create()
.if_not_exists()
.name("idx_chunk_document")
.table(AiKnowledgeChunks::Table)
.col(AiKnowledgeChunks::DocumentId)
.to_owned(),
)
.await?;
manager
.create_index(
Index::create()
.if_not_exists()
.name("idx_chunk_kb")
.table(AiKnowledgeChunks::Table)
.col(AiKnowledgeChunks::KnowledgeBaseId)
.to_owned(),
)
.await
}
async fn down(&self, manager: &SchemaManager) -> Result<(), DbErr> {
manager
.drop_table(Table::drop().table(AiKnowledgeChunks::Table).to_owned())
.await?;
manager
.drop_table(Table::drop().table(AiKnowledgeDocuments::Table).to_owned())
.await
}
}

View File

@@ -616,6 +616,9 @@ async fn main() -> anyhow::Result<()> {
erp_ai::service::embedding::EmbeddingService::from_settings(&db).await,
),
)),
knowledge_v2: std::sync::Arc::new(
erp_ai::service::knowledge_v2::KnowledgeV2Service::new(db.clone()),
),
chat_session: std::sync::Arc::new(
erp_ai::service::chat_session::ChatSessionService::new(db.clone()),
),

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff