search with multiquery redesigned

Cargo.lock

@@ -493,7 +493,7 @@ checksum = "d71b6127be86fdcfddb610f7182ac57211d4b18a3e9c82eb2d17662f2227ad6a"
 
 [[package]]
 name = "canvas"
-version = "0.5.10"
+version = "0.6.3"
 dependencies = [
  "anyhow",
  "async-trait",
@@ -585,7 +585,7 @@ dependencies = [
 
 [[package]]
 name = "client"
-version = "0.5.10"
+version = "0.6.3"
 dependencies = [
  "anyhow",
  "async-trait",
@@ -596,6 +596,7 @@ dependencies = [
  "dotenvy",
  "futures",
  "lazy_static",
+ "nucleo",
  "prost 0.13.5",
  "prost-types 0.13.5",
  "ratatui",
@@ -614,7 +615,6 @@ dependencies = [
  "tonic",
  "tracing",
  "tracing-subscriber",
- "tui-textarea",
  "unicode-segmentation",
  "unicode-width 0.2.0",
  "uuid",
@@ -641,7 +641,7 @@ dependencies = [
 
 [[package]]
 name = "common"
-version = "0.5.10"
+version = "0.6.3"
 dependencies = [
  "prost 0.13.5",
  "prost-build 0.14.1",
@@ -2104,6 +2104,27 @@ dependencies = [
  "windows-sys 0.60.2",
 ]
 
+[[package]]
+name = "nucleo"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5262af4c94921c2646c5ac6ff7900c2af9cbb08dc26a797e18130a7019c039d4"
+dependencies = [
+ "nucleo-matcher",
+ "parking_lot",
+ "rayon",
+]
+
+[[package]]
+name = "nucleo-matcher"
+version = "0.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bf33f538733d1a5a3494b836ba913207f14d9d4a1d3cd67030c5061bdd2cac85"
+dependencies = [
+ "memchr",
+ "unicode-segmentation",
+]
+
 [[package]]
 name = "num-bigint"
 version = "0.4.6"
@@ -3095,7 +3116,7 @@ checksum = "1c107b6f4780854c8b126e228ea8869f4d7b71260f962fefb57b996b8959ba6b"
 
 [[package]]
 name = "search"
-version = "0.5.10"
+version = "0.6.3"
 dependencies = [
  "anyhow",
  "common",
@@ -3194,7 +3215,7 @@ dependencies = [
 
 [[package]]
 name = "server"
-version = "0.5.10"
+version = "0.6.3"
 dependencies = [
  "anyhow",
  "bcrypt",
@@ -4414,18 +4435,6 @@ version = "0.2.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e78122066b0cb818b8afd08f7ed22f7fdbc3e90815035726f0840d0d26c0747a"
 
-[[package]]
-name = "tui-textarea"
-version = "0.7.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0a5318dd619ed73c52a9417ad19046724effc1287fb75cdcc4eca1d6ac1acbae"
-dependencies = [
- "crossterm",
- "ratatui",
- "regex",
- "unicode-width 0.2.0",
-]
-
 [[package]]
 name = "typed-arena"
 version = "2.0.2"

Cargo.toml

@@ -5,7 +5,7 @@ resolver = "2"
 [workspace.package]
 # TODO: idk how to do the name, fix later
 # name = "komp_ac"
-version = "0.5.10"
+version = "0.6.3"
 edition = "2021"
 license = "GPL-3.0-or-later"
 authors = ["Filip Priečinský <filippriec@gmail.com>"]

common/proto/search.proto

@@ -3,18 +3,34 @@ syntax = "proto3";
 package komp_ac.search;
 
 service Searcher {
-  rpc SearchTable(SearchRequest) returns (SearchResponse);
+  rpc Search(SearchRequest) returns (SearchResponse);
+}
+
+enum MatchMode {
+  MATCH_MODE_UNSPECIFIED = 0;
+  MATCH_MODE_FUZZY = 1;
+  MATCH_MODE_EXACT = 2;
+}
+
+message ColumnConstraint {
+  string column = 1;
+  string query = 2;
+  MatchMode mode = 3;
 }
 
 message SearchRequest {
-  string table_name = 1;
-  string query = 2;
+  string profile_name = 1;
+  optional string table_name = 2;
+  string free_query = 3;
+  repeated ColumnConstraint must = 4;
+  optional uint32 limit = 5;
 }
 message SearchResponse {
   message Hit {
     int64 id = 1; // PostgreSQL row ID
     float score = 2;
     string content_json = 3;
+    string table_name = 4;
   }
   repeated Hit hits = 1;
 }

common/proto/table_definition.proto

@@ -18,6 +18,10 @@ service TableDefinition {
   // This provides a tree-like overview of table relationships.
   rpc GetProfileTree(komp_ac.common.Empty) returns (ProfileTreeResponse);
 
+  // Fetches all tables with their columns and scripts for a specific profile.
+  // Pure data retrieval - no business logic.
+  rpc GetProfileDetails(GetProfileDetailsRequest) returns (GetProfileDetailsResponse);
+
   // Drops a table and its metadata, then deletes the profile if it becomes empty.
   rpc DeleteTable(DeleteTableRequest) returns (DeleteTableResponse);
 }
@@ -119,6 +123,35 @@ message ProfileTreeResponse {
   repeated Profile profiles = 1;
 }
 
+// Request to fetch all tables, columns and scripts for a profile.
+message GetProfileDetailsRequest {
+  // Profile (schema) name to fetch details for.
+  string profile_name = 1;
+}
+
+// Response with all tables, columns and scripts for a profile.
+message GetProfileDetailsResponse {
+  string profile_name = 1;
+  repeated TableDetail tables = 2;
+}
+
+// Describes a table with its columns and associated scripts.
+message TableDetail {
+  string name = 1;
+  int64 id = 2;
+  repeated ColumnDefinition columns = 3;
+  repeated ScriptInfo scripts = 4;
+}
+
+// A script that targets a specific column in a table.
+message ScriptInfo {
+  int64 script_id = 1;
+  string target_column = 2;
+  string target_column_type = 3;
+  string script = 4;
+  string description = 5;
+}
+
 // Request to delete one table definition entirely.
 message DeleteTableRequest {
   // Profile (schema) name owning the table (must exist).

common/src/proto/komp_ac.search.rs

@@ -1,10 +1,25 @@
 // This file is @generated by prost-build.
 #[derive(Clone, PartialEq, ::prost::Message)]
+pub struct ColumnConstraint {
+    #[prost(string, tag = "1")]
+    pub column: ::prost::alloc::string::String,
+    #[prost(string, tag = "2")]
+    pub query: ::prost::alloc::string::String,
+    #[prost(enumeration = "MatchMode", tag = "3")]
+    pub mode: i32,
+}
+#[derive(Clone, PartialEq, ::prost::Message)]
 pub struct SearchRequest {
     #[prost(string, tag = "1")]
-    pub table_name: ::prost::alloc::string::String,
-    #[prost(string, tag = "2")]
-    pub query: ::prost::alloc::string::String,
+    pub profile_name: ::prost::alloc::string::String,
+    #[prost(string, optional, tag = "2")]
+    pub table_name: ::core::option::Option<::prost::alloc::string::String>,
+    #[prost(string, tag = "3")]
+    pub free_query: ::prost::alloc::string::String,
+    #[prost(message, repeated, tag = "4")]
+    pub must: ::prost::alloc::vec::Vec<ColumnConstraint>,
+    #[prost(uint32, optional, tag = "5")]
+    pub limit: ::core::option::Option<u32>,
 }
 #[derive(Clone, PartialEq, ::prost::Message)]
 pub struct SearchResponse {
@@ -22,6 +37,37 @@ pub mod search_response {
         pub score: f32,
         #[prost(string, tag = "3")]
         pub content_json: ::prost::alloc::string::String,
+        #[prost(string, tag = "4")]
+        pub table_name: ::prost::alloc::string::String,
+    }
+}
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash, PartialOrd, Ord, ::prost::Enumeration)]
+#[repr(i32)]
+pub enum MatchMode {
+    Unspecified = 0,
+    Fuzzy = 1,
+    Exact = 2,
+}
+impl MatchMode {
+    /// String value of the enum field names used in the ProtoBuf definition.
+    ///
+    /// The values are not transformed in any way and thus are considered stable
+    /// (if the ProtoBuf definition does not change) and safe for programmatic use.
+    pub fn as_str_name(&self) -> &'static str {
+        match self {
+            Self::Unspecified => "MATCH_MODE_UNSPECIFIED",
+            Self::Fuzzy => "MATCH_MODE_FUZZY",
+            Self::Exact => "MATCH_MODE_EXACT",
+        }
+    }
+    /// Creates an enum from field names used in the ProtoBuf definition.
+    pub fn from_str_name(value: &str) -> ::core::option::Option<Self> {
+        match value {
+            "MATCH_MODE_UNSPECIFIED" => Some(Self::Unspecified),
+            "MATCH_MODE_FUZZY" => Some(Self::Fuzzy),
+            "MATCH_MODE_EXACT" => Some(Self::Exact),
+            _ => None,
+        }
     }
 }
 /// Generated client implementations.
@@ -115,7 +161,7 @@ pub mod searcher_client {
             self.inner = self.inner.max_encoding_message_size(limit);
             self
         }
-        pub async fn search_table(
+        pub async fn search(
             &mut self,
             request: impl tonic::IntoRequest<super::SearchRequest>,
         ) -> std::result::Result<tonic::Response<super::SearchResponse>, tonic::Status> {
@@ -129,11 +175,11 @@ pub mod searcher_client {
                 })?;
             let codec = tonic::codec::ProstCodec::default();
             let path = http::uri::PathAndQuery::from_static(
-                "/komp_ac.search.Searcher/SearchTable",
+                "/komp_ac.search.Searcher/Search",
             );
             let mut req = request.into_request();
             req.extensions_mut()
-                .insert(GrpcMethod::new("komp_ac.search.Searcher", "SearchTable"));
+                .insert(GrpcMethod::new("komp_ac.search.Searcher", "Search"));
             self.inner.unary(req, path, codec).await
         }
     }
@@ -151,7 +197,7 @@ pub mod searcher_server {
     /// Generated trait containing gRPC methods that should be implemented for use with SearcherServer.
     #[async_trait]
     pub trait Searcher: std::marker::Send + std::marker::Sync + 'static {
-        async fn search_table(
+        async fn search(
             &self,
             request: tonic::Request<super::SearchRequest>,
         ) -> std::result::Result<tonic::Response<super::SearchResponse>, tonic::Status>;
@@ -232,11 +278,11 @@ pub mod searcher_server {
         }
         fn call(&mut self, req: http::Request<B>) -> Self::Future {
             match req.uri().path() {
-                "/komp_ac.search.Searcher/SearchTable" => {
+                "/komp_ac.search.Searcher/Search" => {
                     #[allow(non_camel_case_types)]
-                    struct SearchTableSvc<T: Searcher>(pub Arc<T>);
+                    struct SearchSvc<T: Searcher>(pub Arc<T>);
                     impl<T: Searcher> tonic::server::UnaryService<super::SearchRequest>
-                    for SearchTableSvc<T> {
+                    for SearchSvc<T> {
                         type Response = super::SearchResponse;
                         type Future = BoxFuture<
                             tonic::Response<Self::Response>,
@@ -248,7 +294,7 @@ pub mod searcher_server {
                         ) -> Self::Future {
                             let inner = Arc::clone(&self.0);
                             let fut = async move {
-                                <T as Searcher>::search_table(&inner, request).await
+                                <T as Searcher>::search(&inner, request).await
                             };
                             Box::pin(fut)
                         }
@@ -259,7 +305,7 @@ pub mod searcher_server {
                     let max_encoding_message_size = self.max_encoding_message_size;
                     let inner = self.inner.clone();
                     let fut = async move {
-                        let method = SearchTableSvc(inner);
+                        let method = SearchSvc(inner);
                         let codec = tonic::codec::ProstCodec::default();
                         let mut grpc = tonic::server::Grpc::new(codec)
                             .apply_compression_config(

common/src/proto/komp_ac.table_definition.rs

@@ -110,6 +110,47 @@ pub mod profile_tree_response {
         pub tables: ::prost::alloc::vec::Vec<Table>,
     }
 }
+/// Request to fetch all tables, columns and scripts for a profile.
+#[derive(Clone, PartialEq, ::prost::Message)]
+pub struct GetProfileDetailsRequest {
+    /// Profile (schema) name to fetch details for.
+    #[prost(string, tag = "1")]
+    pub profile_name: ::prost::alloc::string::String,
+}
+/// Response with all tables, columns and scripts for a profile.
+#[derive(Clone, PartialEq, ::prost::Message)]
+pub struct GetProfileDetailsResponse {
+    #[prost(string, tag = "1")]
+    pub profile_name: ::prost::alloc::string::String,
+    #[prost(message, repeated, tag = "2")]
+    pub tables: ::prost::alloc::vec::Vec<TableDetail>,
+}
+/// Describes a table with its columns and associated scripts.
+#[derive(Clone, PartialEq, ::prost::Message)]
+pub struct TableDetail {
+    #[prost(string, tag = "1")]
+    pub name: ::prost::alloc::string::String,
+    #[prost(int64, tag = "2")]
+    pub id: i64,
+    #[prost(message, repeated, tag = "3")]
+    pub columns: ::prost::alloc::vec::Vec<ColumnDefinition>,
+    #[prost(message, repeated, tag = "4")]
+    pub scripts: ::prost::alloc::vec::Vec<ScriptInfo>,
+}
+/// A script that targets a specific column in a table.
+#[derive(Clone, PartialEq, ::prost::Message)]
+pub struct ScriptInfo {
+    #[prost(int64, tag = "1")]
+    pub script_id: i64,
+    #[prost(string, tag = "2")]
+    pub target_column: ::prost::alloc::string::String,
+    #[prost(string, tag = "3")]
+    pub target_column_type: ::prost::alloc::string::String,
+    #[prost(string, tag = "4")]
+    pub script: ::prost::alloc::string::String,
+    #[prost(string, tag = "5")]
+    pub description: ::prost::alloc::string::String,
+}
 /// Request to delete one table definition entirely.
 #[derive(Clone, PartialEq, ::prost::Message)]
 pub struct DeleteTableRequest {
@@ -289,6 +330,37 @@ pub mod table_definition_client {
                 );
             self.inner.unary(req, path, codec).await
         }
+        /// Fetches all tables with their columns and scripts for a specific profile.
+        /// Pure data retrieval - no business logic.
+        pub async fn get_profile_details(
+            &mut self,
+            request: impl tonic::IntoRequest<super::GetProfileDetailsRequest>,
+        ) -> std::result::Result<
+            tonic::Response<super::GetProfileDetailsResponse>,
+            tonic::Status,
+        > {
+            self.inner
+                .ready()
+                .await
+                .map_err(|e| {
+                    tonic::Status::unknown(
+                        format!("Service was not ready: {}", e.into()),
+                    )
+                })?;
+            let codec = tonic::codec::ProstCodec::default();
+            let path = http::uri::PathAndQuery::from_static(
+                "/komp_ac.table_definition.TableDefinition/GetProfileDetails",
+            );
+            let mut req = request.into_request();
+            req.extensions_mut()
+                .insert(
+                    GrpcMethod::new(
+                        "komp_ac.table_definition.TableDefinition",
+                        "GetProfileDetails",
+                    ),
+                );
+            self.inner.unary(req, path, codec).await
+        }
         /// Drops a table and its metadata, then deletes the profile if it becomes empty.
         pub async fn delete_table(
             &mut self,
@@ -353,6 +425,15 @@ pub mod table_definition_server {
             tonic::Response<super::ProfileTreeResponse>,
             tonic::Status,
         >;
+        /// Fetches all tables with their columns and scripts for a specific profile.
+        /// Pure data retrieval - no business logic.
+        async fn get_profile_details(
+            &self,
+            request: tonic::Request<super::GetProfileDetailsRequest>,
+        ) -> std::result::Result<
+            tonic::Response<super::GetProfileDetailsResponse>,
+            tonic::Status,
+        >;
         /// Drops a table and its metadata, then deletes the profile if it becomes empty.
         async fn delete_table(
             &self,
@@ -537,6 +618,52 @@ pub mod table_definition_server {
                     };
                     Box::pin(fut)
                 }
+                "/komp_ac.table_definition.TableDefinition/GetProfileDetails" => {
+                    #[allow(non_camel_case_types)]
+                    struct GetProfileDetailsSvc<T: TableDefinition>(pub Arc<T>);
+                    impl<
+                        T: TableDefinition,
+                    > tonic::server::UnaryService<super::GetProfileDetailsRequest>
+                    for GetProfileDetailsSvc<T> {
+                        type Response = super::GetProfileDetailsResponse;
+                        type Future = BoxFuture<
+                            tonic::Response<Self::Response>,
+                            tonic::Status,
+                        >;
+                        fn call(
+                            &mut self,
+                            request: tonic::Request<super::GetProfileDetailsRequest>,
+                        ) -> Self::Future {
+                            let inner = Arc::clone(&self.0);
+                            let fut = async move {
+                                <T as TableDefinition>::get_profile_details(&inner, request)
+                                    .await
+                            };
+                            Box::pin(fut)
+                        }
+                    }
+                    let accept_compression_encodings = self.accept_compression_encodings;
+                    let send_compression_encodings = self.send_compression_encodings;
+                    let max_decoding_message_size = self.max_decoding_message_size;
+                    let max_encoding_message_size = self.max_encoding_message_size;
+                    let inner = self.inner.clone();
+                    let fut = async move {
+                        let method = GetProfileDetailsSvc(inner);
+                        let codec = tonic::codec::ProstCodec::default();
+                        let mut grpc = tonic::server::Grpc::new(codec)
+                            .apply_compression_config(
+                                accept_compression_encodings,
+                                send_compression_encodings,
+                            )
+                            .apply_max_message_size_config(
+                                max_decoding_message_size,
+                                max_encoding_message_size,
+                            );
+                        let res = grpc.unary(method, req).await;
+                        Ok(res)
+                    };
+                    Box::pin(fut)
+                }
                 "/komp_ac.table_definition.TableDefinition/DeleteTable" => {
                     #[allow(non_camel_case_types)]
                     struct DeleteTableSvc<T: TableDefinition>(pub Arc<T>);

common/src/search.rs

@@ -1,75 +1,182 @@
-// common/src/search.rs
+use std::path::{Path, PathBuf};
 
-use tantivy::schema::*;
-use tantivy::tokenizer::*;
+use tantivy::schema::{
+    Field, IndexRecordOption, JsonObjectOptions, Schema, TextFieldIndexing, Term, INDEXED,
+    STORED, STRING,
+};
+use tantivy::tokenizer::{
+    AsciiFoldingFilter, LowerCaser, NgramTokenizer, RawTokenizer, RemoveLongFilter,
+    SimpleTokenizer, TextAnalyzer, TokenStream,
+};
 use tantivy::Index;
 
-/// Creates a hybrid Slovak search schema with optimized prefix fields.
+pub const F_PG_ID: &str = "pg_id";
+pub const F_TABLE_NAME: &str = "table_name";
+pub const F_ROW_KEY: &str = "row_key";
+pub const F_DATA_WORD: &str = "data_word";
+pub const F_DATA_NGRAM: &str = "data_ngram";
+pub const F_DATA_EXACT: &str = "data_exact";
+
+pub const TOK_WORD: &str = "kw_word";
+pub const TOK_NGRAM: &str = "kw_ngram";
+pub const TOK_EXACT: &str = "kw_exact";
+
+/// Returns the on-disk path for a profile search index.
+pub fn search_index_path(root: &Path, profile_name: &str) -> PathBuf {
+    root.join(profile_name)
+}
+
+/// Returns the unique index key for one table row inside a profile index.
+pub fn search_row_key(table_name: &str, row_id: i64) -> String {
+    format!("{}:{}", table_name, row_id)
+}
+
+/// Normalizes user-entered values for exact-mode terms.
+pub fn normalize_exact(input: &str) -> String {
+    let trimmed = input.trim();
+    if trimmed.is_empty() {
+        return String::new();
+    }
+
+    let mut analyzer = exact_analyzer();
+    let mut stream = analyzer.token_stream(trimmed);
+    let mut out = String::with_capacity(trimmed.len());
+    while let Some(token) = stream.next() {
+        out.push_str(&token.text);
+    }
+    out
+}
+
+/// Normalizes a column name to the JSON-key form used at index time.
+pub fn normalize_column_name(column: &str) -> String {
+    column.to_ascii_lowercase()
+}
+
+/// Creates the column-aware search schema.
 pub fn create_search_schema() -> Schema {
     let mut schema_builder = Schema::builder();
 
-    schema_builder.add_u64_field("pg_id", INDEXED | STORED);
+    schema_builder.add_u64_field(F_PG_ID, INDEXED | STORED);
+    schema_builder.add_text_field(F_TABLE_NAME, STRING | STORED);
+    schema_builder.add_text_field(F_ROW_KEY, STRING | STORED);
 
-    // FIELD 1: For prefixes (1-4 chars).
-    let short_prefix_indexing = TextFieldIndexing::default()
-        .set_tokenizer("slovak_prefix_edge")
-        .set_index_option(IndexRecordOption::WithFreqsAndPositions);
-    let short_prefix_options = TextOptions::default()
-        .set_indexing_options(short_prefix_indexing)
-        .set_stored();
-    schema_builder.add_text_field("prefix_edge", short_prefix_options);
-
-    // FIELD 2: For the full word.
-    let full_word_indexing = TextFieldIndexing::default()
-        .set_tokenizer("slovak_prefix_full")
-        .set_index_option(IndexRecordOption::WithFreqsAndPositions);
-    let full_word_options = TextOptions::default()
-        .set_indexing_options(full_word_indexing)
-        .set_stored();
-    schema_builder.add_text_field("prefix_full", full_word_options);
-
-    // NGRAM FIELD: For substring matching.
-    let ngram_field_indexing = TextFieldIndexing::default()
-        .set_tokenizer("slovak_ngram")
-        .set_index_option(IndexRecordOption::WithFreqsAndPositions);
-    let ngram_options = TextOptions::default()
-        .set_indexing_options(ngram_field_indexing)
-        .set_stored();
-    schema_builder.add_text_field("text_ngram", ngram_options);
+    schema_builder.add_json_field(F_DATA_WORD, json_options(TOK_WORD, true, false));
+    schema_builder.add_json_field(F_DATA_NGRAM, json_options(TOK_NGRAM, true, false));
+    schema_builder.add_json_field(F_DATA_EXACT, json_options(TOK_EXACT, false, false));
 
     schema_builder.build()
 }
 
-/// Registers all necessary Slovak tokenizers with the index.
-///
-/// This must be called by ANY process that opens the index
-/// to ensure the tokenizers are loaded into memory.
-pub fn register_slovak_tokenizers(index: &Index) -> tantivy::Result<()> {
+fn json_options(
+    tokenizer_name: &str,
+    with_positions: bool,
+    stored: bool,
+) -> JsonObjectOptions {
+    let index_option = if with_positions {
+        IndexRecordOption::WithFreqsAndPositions
+    } else {
+        IndexRecordOption::Basic
+    };
+
+    let indexing = TextFieldIndexing::default()
+        .set_tokenizer(tokenizer_name)
+        .set_index_option(index_option);
+
+    let mut options = JsonObjectOptions::default().set_indexing_options(indexing);
+    if stored {
+        options = options.set_stored();
+    }
+    options
+}
+
+/// Registers all required tokenizers with the index.
+pub fn register_tokenizers(index: &Index) -> tantivy::Result<()> {
     let tokenizer_manager = index.tokenizers();
 
-    // TOKENIZER for `prefix_edge`: Edge N-gram (1-4 chars)
-    let edge_tokenizer = TextAnalyzer::builder(NgramTokenizer::new(1, 4, true)?)
-        .filter(RemoveLongFilter::limit(40))
-        .filter(LowerCaser)
-        .filter(AsciiFoldingFilter)
-        .build();
-    tokenizer_manager.register("slovak_prefix_edge", edge_tokenizer);
-
-    // TOKENIZER for `prefix_full`: Simple word tokenizer
-    let full_tokenizer = TextAnalyzer::builder(SimpleTokenizer::default())
-        .filter(RemoveLongFilter::limit(40))
-        .filter(LowerCaser)
-        .filter(AsciiFoldingFilter)
-        .build();
-    tokenizer_manager.register("slovak_prefix_full", full_tokenizer);
-
-    // NGRAM TOKENIZER: For substring matching.
-    let ngram_tokenizer = TextAnalyzer::builder(NgramTokenizer::new(3, 3, false)?)
-        .filter(RemoveLongFilter::limit(40))
-        .filter(LowerCaser)
-        .filter(AsciiFoldingFilter)
-        .build();
-    tokenizer_manager.register("slovak_ngram", ngram_tokenizer);
+    tokenizer_manager.register(TOK_WORD, word_analyzer());
+    tokenizer_manager.register(TOK_NGRAM, ngram_analyzer()?);
+    tokenizer_manager.register(TOK_EXACT, exact_analyzer());
 
     Ok(())
 }
+
+fn word_analyzer() -> TextAnalyzer {
+    TextAnalyzer::builder(SimpleTokenizer::default())
+        .filter(RemoveLongFilter::limit(80))
+        .filter(LowerCaser)
+        .filter(AsciiFoldingFilter)
+        .build()
+}
+
+fn ngram_analyzer() -> tantivy::Result<TextAnalyzer> {
+    Ok(TextAnalyzer::builder(NgramTokenizer::new(3, 3, false)?)
+        .filter(RemoveLongFilter::limit(80))
+        .filter(LowerCaser)
+        .filter(AsciiFoldingFilter)
+        .build())
+}
+
+fn exact_analyzer() -> TextAnalyzer {
+    TextAnalyzer::builder(RawTokenizer::default())
+        .filter(LowerCaser)
+        .filter(AsciiFoldingFilter)
+        .build()
+}
+
+/// Tokenizes text the same way `data_word` is indexed.
+pub fn tokenize_word(text: &str) -> Vec<String> {
+    tokenize_with(word_analyzer(), text)
+}
+
+/// Tokenizes text the same way `data_ngram` is indexed.
+pub fn tokenize_ngram(text: &str) -> Vec<String> {
+    match ngram_analyzer() {
+        Ok(analyzer) => tokenize_with(analyzer, text),
+        Err(_) => Vec::new(),
+    }
+}
+
+fn tokenize_with(mut analyzer: TextAnalyzer, text: &str) -> Vec<String> {
+    let mut stream = analyzer.token_stream(text);
+    let mut out = Vec::new();
+    while let Some(token) = stream.next() {
+        out.push(token.text.clone());
+    }
+    out
+}
+
+/// Builds a term scoped to a specific JSON path within a JSON field.
+pub fn json_path_term(field: Field, column: &str, text: &str) -> Term {
+    let mut term = Term::from_field_json_path(field, column, false);
+    term.append_type_and_str(text);
+    term
+}
+
+/// Returns all required schema fields or fails loudly on mismatch.
+pub struct SchemaFields {
+    pub pg_id: Field,
+    pub table_name: Field,
+    pub row_key: Field,
+    pub data_word: Field,
+    pub data_ngram: Field,
+    pub data_exact: Field,
+}
+
+impl SchemaFields {
+    pub fn from(schema: &Schema) -> tantivy::Result<Self> {
+        Ok(Self {
+            pg_id: get_field(schema, F_PG_ID)?,
+            table_name: get_field(schema, F_TABLE_NAME)?,
+            row_key: get_field(schema, F_ROW_KEY)?,
+            data_word: get_field(schema, F_DATA_WORD)?,
+            data_ngram: get_field(schema, F_DATA_NGRAM)?,
+            data_exact: get_field(schema, F_DATA_EXACT)?,
+        })
+    }
+}
+
+fn get_field(schema: &Schema, name: &str) -> tantivy::Result<Field> {
+    schema.get_field(name).map_err(|e| {
+        tantivy::TantivyError::SchemaError(format!("schema is missing field '{name}': {e}"))
+    })
+}

search/.gitignore

@@ -0,0 +1 @@
+.codex

search/src/lib.rs

@@ -1,279 +1,426 @@
-// src/lib.rs
+mod query_builder;
 
 use std::collections::HashMap;
 use std::path::Path;
-use tantivy::collector::TopDocs;
-use tantivy::query::{
-    BooleanQuery, BoostQuery, FuzzyTermQuery, Occur, Query, QueryParser, TermQuery,
-};
-use tantivy::schema::{IndexRecordOption, Value};
-use tantivy::{Index, TantivyDocument, Term};
-use tonic::{Request, Response, Status};
+use std::sync::{Arc, Mutex};
 
 use common::proto::komp_ac::search::searcher_server::Searcher;
-pub use common::proto::komp_ac::search::searcher_server::SearcherServer;
 use common::proto::komp_ac::search::{search_response::Hit, SearchRequest, SearchResponse};
-use common::search::register_slovak_tokenizers;
+pub use common::proto::komp_ac::search::searcher_server::SearcherServer;
+use common::search::{register_tokenizers, search_index_path, SchemaFields};
+use query_builder::{build_master_query, ConstraintMode, SearchConstraint};
 use sqlx::{PgPool, Row};
+use tantivy::collector::TopDocs;
+use tantivy::schema::Value;
+use tantivy::{Index, IndexReader, ReloadPolicy, TantivyDocument};
+use tonic::{Request, Response, Status};
 use tracing::info;
 
-// We need to hold the database pool in our service struct.
+const INDEX_ROOT: &str = "./tantivy_indexes";
+const DEFAULT_RESULT_LIMIT: usize = 25;
+const HARD_RESULT_LIMIT: usize = 200;
+const DEFAULT_LIST_LIMIT: usize = 5;
+
 pub struct SearcherService {
     pub pool: PgPool,
+    profiles: Mutex<HashMap<String, Arc<ProfileIndex>>>,
 }
 
-// normalize_slovak_text function remains unchanged...
-fn normalize_slovak_text(text: &str) -> String {
-    // ... function content is unchanged ...
-    text.chars()
-        .map(|c| match c {
-            'á' | 'à' | 'â' | 'ä' | 'ă' | 'ā' => 'a',
-            'Á' | 'À' | 'Â' | 'Ä' | 'Ă' | 'Ā' => 'A',
-            'é' | 'è' | 'ê' | 'ë' | 'ě' | 'ē' => 'e',
-            'É' | 'È' | 'Ê' | 'Ë' | 'Ě' | 'Ē' => 'E',
-            'í' | 'ì' | 'î' | 'ï' | 'ī' => 'i',
-            'Í' | 'Ì' | 'Î' | 'Ï' | 'Ī' => 'I',
-            'ó' | 'ò' | 'ô' | 'ö' | 'ō' | 'ő' => 'o',
-            'Ó' | 'Ò' | 'Ô' | 'Ö' | 'Ō' | 'Ő' => 'O',
-            'ú' | 'ù' | 'û' | 'ü' | 'ū' | 'ű' => 'u',
-            'Ú' | 'Ù' | 'Û' | 'Ü' | 'Ū' | 'Ű' => 'U',
-            'ý' | 'ỳ' | 'ŷ' | 'ÿ' => 'y',
-            'Ý' | 'Ỳ' | 'Ŷ' | 'Ÿ' => 'Y',
-            'č' => 'c',
-            'Č' => 'C',
-            'ď' => 'd',
-            'Ď' => 'D',
-            'ľ' => 'l',
-            'Ľ' => 'L',
-            'ň' => 'n',
-            'Ň' => 'N',
-            'ř' => 'r',
-            'Ř' => 'R',
-            'š' => 's',
-            'Š' => 'S',
-            'ť' => 't',
-            'Ť' => 'T',
-            'ž' => 'z',
-            'Ž' => 'Z',
-            _ => c,
-        })
-        .collect()
-}
+impl SearcherService {
+    pub fn new(pool: PgPool) -> Self {
+        Self {
+            pool,
+            profiles: Mutex::new(HashMap::new()),
+        }
+    }
 
-#[tonic::async_trait]
-impl Searcher for SearcherService {
-    async fn search_table(
-        &self,
-        request: Request<SearchRequest>,
-    ) -> Result<Response<SearchResponse>, Status> {
+    async fn run_rpc(&self, request: Request<SearchRequest>) -> Result<Response<SearchResponse>, Status> {
         let req = request.into_inner();
-        let table_name = req.table_name;
-        let query_str = req.query;
+        let normalized = normalize_request(req)?;
+
+        if !profile_exists(&self.pool, &normalized.profile_name).await? {
+            return Err(Status::not_found(format!(
+                "Profile '{}' was not found",
+                normalized.profile_name
+            )));
+        }
+
+        if let Some(table_name) = normalized.table_name.as_deref() {
+            if !table_exists(&self.pool, &normalized.profile_name, table_name).await? {
+                return Err(Status::not_found(format!(
+                    "Table '{}' was not found in profile '{}'",
+                    table_name, normalized.profile_name
+                )));
+            }
+        }
+
+        if !normalized.has_input() {
+            let Some(table_name) = normalized.table_name.as_deref() else {
+                return Err(Status::invalid_argument(
+                    "table_name is required when query is empty",
+                ));
+            };
+
+            let hits = fetch_latest_rows(
+                &self.pool,
+                &normalized.profile_name,
+                table_name,
+                normalized.limit.unwrap_or(DEFAULT_LIST_LIMIT),
+            )
+            .await?;
+            return Ok(Response::new(SearchResponse { hits }));
+        }
+
+        let index_path = search_index_path(Path::new(INDEX_ROOT), &normalized.profile_name);
+        if !index_path.exists() {
+            return Err(Status::not_found(format!(
+                "No search index found for profile '{}'",
+                normalized.profile_name
+            )));
+        }
+
+        let profile = profile_index(&self.profiles, &normalized.profile_name, &index_path)?;
+        let mut hits = run_search(
+            &self.pool,
+            &profile,
+            &normalized.profile_name,
+            normalized.table_name.as_deref(),
+            &normalized.free_query,
+            &normalized.must,
+            normalized.limit.unwrap_or(DEFAULT_RESULT_LIMIT),
+        )
+        .await?;
+
+        hits.sort_by(|left, right| right.score.total_cmp(&left.score));
+        if let Some(limit) = normalized.limit {
+            if hits.len() > limit {
+                hits.truncate(limit);
+            }
+        }
 
-        // --- MODIFIED LOGIC ---
-        // If the query is empty, fetch the 5 most recent records.
-        if query_str.trim().is_empty() {
         info!(
-                "Empty query for table '{}'. Fetching default results.",
-                table_name
-            );
-            let qualified_table = format!("gen.\"{}\"", table_name);
-            let sql = format!(
-                "SELECT id, to_jsonb(t) AS data FROM {} t ORDER BY id DESC LIMIT 5",
-                qualified_table
+            "search: profile={} table={:?} free='{}' constraints={} hits={}",
+            normalized.profile_name,
+            normalized.table_name,
+            normalized.free_query,
+            normalized.must.len(),
+            hits.len()
         );
 
-            let rows = sqlx::query(&sql).fetch_all(&self.pool).await.map_err(|e| {
-                Status::internal(format!("DB query for default results failed: {}", e))
+        Ok(Response::new(SearchResponse { hits }))
+    }
+}
+
+struct ProfileIndex {
+    index: Index,
+    reader: IndexReader,
+    fields: SchemaFields,
+}
+
+impl ProfileIndex {
+    fn open(path: &Path) -> Result<Self, Status> {
+        let index = Index::open_in_dir(path)
+            .map_err(|e| Status::internal(format!("Failed to open index: {}", e)))?;
+        register_tokenizers(&index)
+            .map_err(|e| Status::internal(format!("Failed to register tokenizers: {}", e)))?;
+        let reader = index
+            .reader_builder()
+            .reload_policy(ReloadPolicy::OnCommitWithDelay)
+            .try_into()
+            .map_err(|e| Status::internal(format!("Failed to build index reader: {}", e)))?;
+        let fields = SchemaFields::from(&index.schema()).map_err(|e| {
+            Status::internal(format!(
+                "Search index schema mismatch. Reindex required: {}",
+                e
+            ))
         })?;
 
-            let hits: Vec<Hit> = rows
+        Ok(Self {
+            index,
+            reader,
+            fields,
+        })
+    }
+}
+
+#[derive(Debug)]
+struct NormalizedSearchRequest {
+    profile_name: String,
+    table_name: Option<String>,
+    free_query: String,
+    must: Vec<SearchConstraint>,
+    limit: Option<usize>,
+}
+
+impl NormalizedSearchRequest {
+    fn has_input(&self) -> bool {
+        !self.free_query.is_empty() || !self.must.is_empty()
+    }
+}
+
+fn profile_index(
+    cache: &Mutex<HashMap<String, Arc<ProfileIndex>>>,
+    profile_name: &str,
+    path: &Path,
+) -> Result<Arc<ProfileIndex>, Status> {
+    {
+        let cache_guard = cache
+            .lock()
+            .map_err(|_| Status::internal("Profile index cache lock poisoned"))?;
+        if let Some(index) = cache_guard.get(profile_name) {
+            return Ok(index.clone());
+        }
+    }
+
+    let opened = Arc::new(ProfileIndex::open(path)?);
+    let mut cache_guard = cache
+        .lock()
+        .map_err(|_| Status::internal("Profile index cache lock poisoned"))?;
+    if let Some(index) = cache_guard.get(profile_name) {
+        return Ok(index.clone());
+    }
+    cache_guard.insert(profile_name.to_string(), opened.clone());
+    Ok(opened)
+}
+
+fn validate_identifier(value: &str, field_name: &str) -> Result<(), Status> {
+    let mut chars = value.chars();
+    let Some(first) = chars.next() else {
+        return Err(Status::invalid_argument(format!(
+            "{field_name} must not be empty"
+        )));
+    };
+
+    if !(first.is_ascii_alphabetic() || first == '_')
+        || !chars.all(|ch| ch.is_ascii_alphanumeric() || ch == '_')
+    {
+        return Err(Status::invalid_argument(format!(
+            "{field_name} contains invalid characters"
+        )));
+    }
+
+    Ok(())
+}
+
+fn qualify_profile_table(profile_name: &str, table_name: &str) -> String {
+    format!("\"{}\".\"{}\"", profile_name, table_name)
+}
+
+async fn profile_exists(pool: &PgPool, profile_name: &str) -> Result<bool, Status> {
+    let exists =
+        sqlx::query_scalar::<_, bool>("SELECT EXISTS(SELECT 1 FROM schemas WHERE name = $1)")
+            .bind(profile_name)
+            .fetch_one(pool)
+            .await
+            .map_err(|e| Status::internal(format!("Profile lookup failed: {}", e)))?;
+    Ok(exists)
+}
+
+async fn table_exists(pool: &PgPool, profile_name: &str, table_name: &str) -> Result<bool, Status> {
+    let exists = sqlx::query_scalar::<_, bool>(
+        r#"
+        SELECT EXISTS(
+            SELECT 1
+            FROM table_definitions td
+            JOIN schemas s ON td.schema_id = s.id
+            WHERE s.name = $1 AND td.table_name = $2
+        )
+        "#,
+    )
+    .bind(profile_name)
+    .bind(table_name)
+    .fetch_one(pool)
+    .await
+    .map_err(|e| Status::internal(format!("Table lookup failed: {}", e)))?;
+    Ok(exists)
+}
+
+fn normalize_request(req: SearchRequest) -> Result<NormalizedSearchRequest, Status> {
+    let profile_name = req.profile_name.trim();
+    if profile_name.is_empty() {
+        return Err(Status::invalid_argument("profile_name is required"));
+    }
+    validate_identifier(profile_name, "profile_name")?;
+
+    let table_name = match req.table_name.as_deref().map(str::trim) {
+        Some(table_name) if !table_name.is_empty() => {
+            validate_identifier(table_name, "table_name")?;
+            Some(table_name.to_string())
+        }
+        _ => None,
+    };
+
+    let free_query = req.free_query.trim().to_string();
+    let mut must = Vec::new();
+
+    for constraint in req.must {
+        let column = constraint.column.trim();
+        if column.is_empty() {
+            return Err(Status::invalid_argument(
+                "constraint.column must not be empty",
+            ));
+        }
+        validate_identifier(column, "constraint.column")?;
+
+        let query = constraint.query.trim();
+        if query.is_empty() {
+            return Err(Status::invalid_argument(
+                "constraint.query must not be empty",
+            ));
+        }
+
+        must.push(SearchConstraint {
+            column: column.to_string(),
+            query: query.to_string(),
+            mode: constraint_mode_from_proto(constraint.mode),
+        });
+    }
+
+    let limit = req.limit.map(|value| (value as usize).min(HARD_RESULT_LIMIT));
+
+    Ok(NormalizedSearchRequest {
+        profile_name: profile_name.to_string(),
+        table_name,
+        free_query,
+        must,
+        limit,
+    })
+}
+
+fn constraint_mode_from_proto(raw_mode: i32) -> ConstraintMode {
+    match raw_mode {
+        2 => ConstraintMode::Exact,
+        _ => ConstraintMode::Fuzzy,
+    }
+}
+
+async fn fetch_latest_rows(
+    pool: &PgPool,
+    profile_name: &str,
+    table_name: &str,
+    limit: usize,
+) -> Result<Vec<Hit>, Status> {
+    let sql = format!(
+        "SELECT id, to_jsonb(t) AS data FROM {} t WHERE deleted = FALSE ORDER BY id DESC LIMIT $1",
+        qualify_profile_table(profile_name, table_name)
+    );
+
+    let rows = sqlx::query(&sql)
+        .bind(limit as i64)
+        .fetch_all(pool)
+        .await
+        .map_err(|e| Status::internal(format!("DB query for default results failed: {}", e)))?;
+
+    Ok(rows
         .into_iter()
         .map(|row| {
             let id: i64 = row.try_get("id").unwrap_or_default();
             let json_data: serde_json::Value = row.try_get("data").unwrap_or_default();
             Hit {
                 id,
-                        // Score is 0.0 as this is not a relevance-ranked search
                 score: 0.0,
                 content_json: json_data.to_string(),
+                table_name: table_name.to_string(),
             }
         })
-                .collect();
+        .collect())
+}
 
-            info!(
-                "--- SERVER: Successfully processed empty query. Returning {} default hits. ---",
-                hits.len()
-            );
-            return Ok(Response::new(SearchResponse { hits }));
-        }
-        // --- END OF MODIFIED LOGIC ---
-
-        let index_path = Path::new("./tantivy_indexes").join(&table_name);
-        if !index_path.exists() {
-            return Err(Status::not_found(format!(
-                "No search index found for table '{}'",
-                table_name
-            )));
-        }
-
-        let index = Index::open_in_dir(&index_path)
-            .map_err(|e| Status::internal(format!("Failed to open index: {}", e)))?;
-
-        register_slovak_tokenizers(&index).map_err(|e| {
-            Status::internal(format!("Failed to register Slovak tokenizers: {}", e))
-        })?;
-
-        let reader = index
-            .reader()
-            .map_err(|e| Status::internal(format!("Failed to create index reader: {}", e)))?;
-        let searcher = reader.searcher();
-        let schema = index.schema();
-
-        let pg_id_field = schema
-            .get_field("pg_id")
-            .map_err(|_| Status::internal("Schema is missing the 'pg_id' field."))?;
-
-        // --- Query Building Logic (no changes here) ---
-        let prefix_edge_field = schema.get_field("prefix_edge").unwrap();
-        let prefix_full_field = schema.get_field("prefix_full").unwrap();
-        let text_ngram_field = schema.get_field("text_ngram").unwrap();
-        let normalized_query = normalize_slovak_text(&query_str);
-        let words: Vec<&str> = normalized_query.split_whitespace().collect();
-        if words.is_empty() {
-            return Ok(Response::new(SearchResponse { hits: vec![] }));
-        }
-        let mut query_layers: Vec<(Occur, Box<dyn Query>)> = Vec::new();
-        // ... all your query building layers remain exactly the same ...
-        // ===============================
-        // LAYER 1: PREFIX MATCHING (HIGHEST PRIORITY, Boost: 4.0)
-        // ===============================
-        {
-            let mut must_clauses: Vec<(Occur, Box<dyn Query>)> = Vec::new();
-            for word in &words {
-                let edge_term = Term::from_field_text(prefix_edge_field, word);
-                let full_term = Term::from_field_text(prefix_full_field, word);
-
-                let per_word_query = BooleanQuery::new(vec![
-                    (
-                        Occur::Should,
-                        Box::new(TermQuery::new(edge_term, IndexRecordOption::Basic)),
-                    ),
-                    (
-                        Occur::Should,
-                        Box::new(TermQuery::new(full_term, IndexRecordOption::Basic)),
-                    ),
-                ]);
-                must_clauses.push((Occur::Must, Box::new(per_word_query) as Box<dyn Query>));
-            }
-
-            if !must_clauses.is_empty() {
-                let prefix_query = BooleanQuery::new(must_clauses);
-                let boosted_query = BoostQuery::new(Box::new(prefix_query), 4.0);
-                query_layers.push((Occur::Should, Box::new(boosted_query)));
-            }
-        }
-
-        // ===============================
-        // LAYER 2: FUZZY MATCHING (HIGH PRIORITY, Boost: 3.0)
-        // ===============================
-        {
-            let last_word = words.last().unwrap();
-            let fuzzy_term = Term::from_field_text(prefix_full_field, last_word);
-            let fuzzy_query = FuzzyTermQuery::new(fuzzy_term, 2, true);
-            let boosted_query = BoostQuery::new(Box::new(fuzzy_query), 3.0);
-            query_layers.push((Occur::Should, Box::new(boosted_query)));
-        }
-
-        // ===============================
-        // LAYER 3: PHRASE MATCHING WITH SLOP (MEDIUM PRIORITY, Boost: 2.0)
-        // ===============================
-        if words.len() > 1 {
-            let slop_parser = QueryParser::for_index(&index, vec![prefix_full_field]);
-            let slop_query_str = format!("\"{}\"~3", normalized_query);
-            if let Ok(slop_query) = slop_parser.parse_query(&slop_query_str) {
-                let boosted_query = BoostQuery::new(slop_query, 2.0);
-                query_layers.push((Occur::Should, Box::new(boosted_query)));
-            }
-        }
-
-        // ===============================
-        // LAYER 4: NGRAM SUBSTRING MATCHING (LOWEST PRIORITY, Boost: 1.0)
-        // ===============================
-        {
-            let ngram_parser = QueryParser::for_index(&index, vec![text_ngram_field]);
-            if let Ok(ngram_query) = ngram_parser.parse_query(&normalized_query) {
-                let boosted_query = BoostQuery::new(ngram_query, 1.0);
-                query_layers.push((Occur::Should, Box::new(boosted_query)));
-            }
-        }
-        let master_query = BooleanQuery::new(query_layers);
-        // --- End of Query Building Logic ---
+async fn run_search(
+    pool: &PgPool,
+    profile: &ProfileIndex,
+    profile_name: &str,
+    table_filter: Option<&str>,
+    free_query: &str,
+    must: &[SearchConstraint],
+    limit: usize,
+) -> Result<Vec<Hit>, Status> {
+    let master_query =
+        build_master_query(&profile.index, &profile.fields, free_query, must, table_filter)?;
 
+    let searcher = profile.reader.searcher();
     let top_docs = searcher
-            .search(&master_query, &TopDocs::with_limit(100))
+        .search(&*master_query, &TopDocs::with_limit(limit))
         .map_err(|e| Status::internal(format!("Search failed: {}", e)))?;
 
     if top_docs.is_empty() {
-            return Ok(Response::new(SearchResponse { hits: vec![] }));
+        return Ok(vec![]);
     }
 
-        // --- NEW LOGIC: Fetch from DB and combine results ---
-
-        // Step 1: Extract (score, pg_id) from Tantivy results.
-        let mut scored_ids: Vec<(f32, u64)> = Vec::new();
+    let mut candidates: Vec<(f32, i64, String)> = Vec::with_capacity(top_docs.len());
     for (score, doc_address) in top_docs {
         let doc: TantivyDocument = searcher
             .doc(doc_address)
             .map_err(|e| Status::internal(format!("Failed to retrieve document: {}", e)))?;
-            if let Some(pg_id_value) = doc.get_first(pg_id_field) {
-                if let Some(pg_id) = pg_id_value.as_u64() {
-                    scored_ids.push((score, pg_id));
-                }
-            }
+        let Some(pg_id) = doc
+            .get_first(profile.fields.pg_id)
+            .and_then(|value| value.as_u64())
+        else {
+            continue;
+        };
+        let Some(table_name) = doc
+            .get_first(profile.fields.table_name)
+            .and_then(|value| value.as_str())
+        else {
+            continue;
+        };
+        candidates.push((score, pg_id as i64, table_name.to_string()));
     }
 
-        // Step 2: Fetch all corresponding rows from Postgres in a single query.
-        let pg_ids: Vec<i64> = scored_ids.iter().map(|(_, id)| *id as i64).collect();
-        let qualified_table = format!("gen.\"{}\"", table_name);
-        let query_str = format!(
-            "SELECT id, to_jsonb(t) AS data FROM {} t WHERE id = ANY($1)",
-            qualified_table
+    if candidates.is_empty() {
+        return Ok(vec![]);
+    }
+
+    let mut ids_by_table: HashMap<String, Vec<i64>> = HashMap::new();
+    for (_, pg_id, table_name) in &candidates {
+        ids_by_table
+            .entry(table_name.clone())
+            .or_default()
+            .push(*pg_id);
+    }
+
+    let mut content_map: HashMap<(String, i64), String> = HashMap::new();
+    for (table_name, pg_ids) in ids_by_table {
+        validate_identifier(&table_name, "table_name")?;
+        let sql = format!(
+            "SELECT id, to_jsonb(t) AS data FROM {} t WHERE deleted = FALSE AND id = ANY($1)",
+            qualify_profile_table(profile_name, &table_name)
         );
 
-        let rows = sqlx::query(&query_str)
+        let rows = sqlx::query(&sql)
             .bind(&pg_ids)
-            .fetch_all(&self.pool)
+            .fetch_all(pool)
             .await
             .map_err(|e| Status::internal(format!("Database query failed: {}", e)))?;
 
-        // Step 3: Map the database results by ID for quick lookup.
-        let mut content_map: HashMap<i64, String> = HashMap::new();
         for row in rows {
-            let id: i64 = row.try_get("id").unwrap_or(0);
-            let json_data: serde_json::Value =
-                row.try_get("data").unwrap_or(serde_json::Value::Null);
-            content_map.insert(id, json_data.to_string());
+            let id: i64 = row.try_get("id").unwrap_or_default();
+            let json_data: serde_json::Value = row.try_get("data").unwrap_or_default();
+            content_map.insert((table_name.clone(), id), json_data.to_string());
+        }
     }
 
-        // Step 4: Build the final response, combining Tantivy scores with PG content.
-        let hits: Vec<Hit> = scored_ids
+    Ok(candidates
         .into_iter()
-            .filter_map(|(score, pg_id)| {
-                content_map.get(&(pg_id as i64)).map(|content_json| Hit {
-                    id: pg_id as i64,
+        .filter_map(|(score, pg_id, table_name)| {
+            content_map
+                .get(&(table_name.clone(), pg_id))
+                .map(|content_json| Hit {
+                    id: pg_id,
                     score,
                     content_json: content_json.clone(),
+                    table_name,
                 })
         })
-            .collect();
+        .collect())
+}
 
-        info!(
-            "--- SERVER: Successfully processed search. Returning {} hits. ---",
-            hits.len()
-        );
-
-        let response = SearchResponse { hits };
-        Ok(Response::new(response))
+#[tonic::async_trait]
+impl Searcher for SearcherService {
+    async fn search(
+        &self,
+        request: Request<SearchRequest>,
+    ) -> Result<Response<SearchResponse>, Status> {
+        self.run_rpc(request).await
     }
 }

search/src/query_builder.rs

@@ -0,0 +1,234 @@
+use common::search::{
+    json_path_term, normalize_exact, tokenize_ngram, tokenize_word, SchemaFields,
+};
+use tantivy::query::{
+    BooleanQuery, BoostQuery, EmptyQuery, FuzzyTermQuery, Occur, PhraseQuery, Query, QueryParser,
+    TermQuery,
+};
+use tantivy::schema::{IndexRecordOption, Term};
+use tantivy::Index;
+use tonic::Status;
+
+#[derive(Clone, Copy, Debug, Eq, PartialEq)]
+pub enum ConstraintMode {
+    Fuzzy,
+    Exact,
+}
+
+#[derive(Clone, Debug)]
+pub struct SearchConstraint {
+    pub column: String,
+    pub query: String,
+    pub mode: ConstraintMode,
+}
+
+pub fn build_master_query(
+    index: &Index,
+    fields: &SchemaFields,
+    free_query: &str,
+    must: &[SearchConstraint],
+    table_filter: Option<&str>,
+) -> Result<Box<dyn Query>, Status> {
+    let mut clauses: Vec<(Occur, Box<dyn Query>)> = Vec::new();
+    let mut has_search_clause = false;
+
+    for constraint in must {
+        let predicate = match constraint.mode {
+            ConstraintMode::Exact => exact_predicate(fields, &constraint.column, &constraint.query)?,
+            ConstraintMode::Fuzzy => {
+                fuzzy_predicate_scoped(fields, &constraint.column, &constraint.query)?
+            }
+        };
+        clauses.push((Occur::Must, predicate));
+        has_search_clause = true;
+    }
+
+    let free_words = tokenize_word(free_query);
+    if !free_words.is_empty() {
+        let predicate = fuzzy_predicate_unscoped(index, fields, &free_words)?;
+        clauses.push((Occur::Should, predicate));
+        has_search_clause = true;
+    }
+
+    if let Some(table_name) = table_filter {
+        let term = Term::from_field_text(fields.table_name, table_name);
+        clauses.push((
+            Occur::Must,
+            Box::new(TermQuery::new(term, IndexRecordOption::Basic)),
+        ));
+    }
+
+    if !has_search_clause {
+        return Ok(Box::new(EmptyQuery));
+    }
+
+    Ok(Box::new(BooleanQuery::new(clauses)))
+}
+
+fn exact_predicate(
+    fields: &SchemaFields,
+    column: &str,
+    query: &str,
+) -> Result<Box<dyn Query>, Status> {
+    let normalized_value = normalize_exact(query);
+    if normalized_value.is_empty() {
+        return Err(Status::invalid_argument(
+            "exact query is empty after normalization",
+        ));
+    }
+
+    let term = json_path_term(fields.data_exact, column, &normalized_value);
+    Ok(Box::new(TermQuery::new(term, IndexRecordOption::Basic)))
+}
+
+fn fuzzy_predicate_scoped(
+    fields: &SchemaFields,
+    column: &str,
+    query: &str,
+) -> Result<Box<dyn Query>, Status> {
+    let words = tokenize_word(query);
+    if words.is_empty() {
+        return Err(Status::invalid_argument(
+            "fuzzy query has no searchable tokens",
+        ));
+    }
+
+    let mut layers: Vec<(Occur, Box<dyn Query>)> = Vec::new();
+
+    let mut per_word_clauses: Vec<(Occur, Box<dyn Query>)> = Vec::new();
+    for word in &words {
+        let term = json_path_term(fields.data_word, column, word);
+        let mut alternates: Vec<(Occur, Box<dyn Query>)> = Vec::new();
+
+        alternates.push((
+            Occur::Should,
+            Box::new(BoostQuery::new(
+                Box::new(TermQuery::new(term.clone(), IndexRecordOption::WithFreqs)),
+                4.0,
+            )),
+        ));
+
+        alternates.push((
+            Occur::Should,
+            Box::new(BoostQuery::new(
+                Box::new(FuzzyTermQuery::new_prefix(term.clone(), 0, false)),
+                3.0,
+            )),
+        ));
+
+        if let Some(distance) = fuzzy_distance(word.chars().count()) {
+            alternates.push((
+                Occur::Should,
+                Box::new(BoostQuery::new(
+                    Box::new(FuzzyTermQuery::new(term.clone(), distance, true)),
+                    2.0,
+                )),
+            ));
+        }
+
+        per_word_clauses.push((Occur::Must, Box::new(BooleanQuery::new(alternates))));
+    }
+    layers.push((Occur::Should, Box::new(BooleanQuery::new(per_word_clauses))));
+
+    if words.len() > 1 {
+        let phrase_terms: Vec<(usize, Term)> = words
+            .iter()
+            .enumerate()
+            .map(|(offset, word)| (offset, json_path_term(fields.data_word, column, word)))
+            .collect();
+        let phrase = PhraseQuery::new_with_offset_and_slop(phrase_terms, 3);
+        layers.push((
+            Occur::Should,
+            Box::new(BoostQuery::new(Box::new(phrase), 2.0)),
+        ));
+    }
+
+    let ngrams = tokenize_ngram(query);
+    if !ngrams.is_empty() {
+        let ngram_clauses: Vec<(Occur, Box<dyn Query>)> = ngrams
+            .into_iter()
+            .map(|gram| {
+                let term = json_path_term(fields.data_ngram, column, &gram);
+                (
+                    Occur::Must,
+                    Box::new(TermQuery::new(term, IndexRecordOption::Basic)) as Box<dyn Query>,
+                )
+            })
+            .collect();
+        layers.push((
+            Occur::Should,
+            Box::new(BoostQuery::new(Box::new(BooleanQuery::new(ngram_clauses)), 1.0)),
+        ));
+    }
+
+    Ok(Box::new(BooleanQuery::new(layers)))
+}
+
+fn fuzzy_predicate_unscoped(
+    index: &Index,
+    fields: &SchemaFields,
+    words: &[String],
+) -> Result<Box<dyn Query>, Status> {
+    let mut layers: Vec<(Occur, Box<dyn Query>)> = Vec::new();
+
+    {
+        let parser = QueryParser::for_index(index, vec![fields.data_word]);
+        let query_string = words
+            .iter()
+            .map(|word| format!("+{}*", word))
+            .collect::<Vec<_>>()
+            .join(" ");
+        if let Ok(query) = parser.parse_query(&query_string) {
+            layers.push((Occur::Should, Box::new(BoostQuery::new(query, 4.0))));
+        }
+    }
+
+    {
+        let parser = QueryParser::for_index(index, vec![fields.data_word]);
+        let query_string = words
+            .iter()
+            .map(|word| match fuzzy_distance(word.chars().count()) {
+                Some(distance) => format!("+{}~{}", word, distance),
+                None => format!("+{}", word),
+            })
+            .collect::<Vec<_>>()
+            .join(" ");
+        if let Ok(query) = parser.parse_query(&query_string) {
+            layers.push((Occur::Should, Box::new(BoostQuery::new(query, 2.0))));
+        }
+    }
+
+    if words.len() > 1 {
+        let parser = QueryParser::for_index(index, vec![fields.data_word]);
+        let query_string = format!("\"{}\"~3", words.join(" "));
+        if let Ok(query) = parser.parse_query(&query_string) {
+            layers.push((Occur::Should, Box::new(BoostQuery::new(query, 2.0))));
+        }
+    }
+
+    {
+        let parser = QueryParser::for_index(index, vec![fields.data_ngram]);
+        let query_string = words
+            .iter()
+            .map(|word| format!("+{}", word))
+            .collect::<Vec<_>>()
+            .join(" ");
+        if let Ok(query) = parser.parse_query(&query_string) {
+            layers.push((Occur::Should, Box::new(BoostQuery::new(query, 1.0))));
+        }
+    }
+
+    if layers.is_empty() {
+        return Ok(Box::new(EmptyQuery));
+    }
+
+    Ok(Box::new(BooleanQuery::new(layers)))
+}
+
+fn fuzzy_distance(word_len: usize) -> Option<u8> {
+    match word_len {
+        0..=3 => None,
+        4..=6 => Some(1),
+        _ => Some(2),
+    }
+}