Release v1.6.0

build.sbt

@@ -120,7 +120,7 @@ libraryDependencies ++= Seq(
   "org.tpolecat" %% "doobie-specs2"   % doobieVersion,
   "org.tpolecat" %% "doobie-hikari"   % doobieVersion,
   "io.getquill"  %% "quill-doobie"    % "4.8.4",
-  "org.postgresql" % "postgresql"     % "42.7.3",
+  "org.postgresql" % "postgresql"     % "42.7.4",
   "io.getquill" %% "quill-cassandra-monix" % "4.8.4",
 
   // search

js/css/main.css

@@ -940,6 +940,9 @@ popover-content a:hover {
 .search-options {
     width: 100%;
     font-size: 12px;
+    display: flex;
+    flex-direction: row;
+    flex-wrap: wrap;
 }
 
 .search-options * {
@@ -952,6 +955,12 @@ popover-content a:hover {
 
 .search-options select {
     margin-right: 20px;
+    padding-right: 60px !important;
+}
+
+.search-option {
+    display: flex;
+    flex-direction: row;
 }
 
 /* audio player */
@@ -1009,3 +1018,14 @@ popover-content a:hover {
 .nsfw-show .nsfw .article-desc-nsfw {
     display: none;
 }
+
+.text-with-icon {
+    display: flex;
+    flex-direction: row;
+    align-items: center;
+    gap: 10px;
+}
+
+.show-comment-btn {
+    font-size: 18px;
+}

js/src/match-id.js

@@ -182,6 +182,9 @@ window.getPositionBefore = getPositionBefore;
 
 function getNextFolderPosition() {
     const folders = getFoldersFromDom();
+    if (folders.length == 0) {
+        return POS_STEP;
+    }
     const lastFolder = folders[folders.length - 1];
     return lastFolder.position + POS_STEP;
 }

src/main/protobuf/grpc-api.proto

@@ -311,6 +311,8 @@ message User {
     bool subscribed = 16;
     NSFWSetting nsfwSetting = 17;
     SearchEngine searchEngine = 18;
+    optional LLMEngine llmEngine = 19;
+    optional string llmApiKey = 20;
 }
 
 // Define UserInfo
@@ -328,6 +330,7 @@ message UserInfo {
     bool subscribed = 10;
     NSFWSetting nsfwSetting = 11;
     SearchEngine searchEngine = 12;
+    optional LLMEngine llmEngine = 13;
 }
 
 // Define UserUpdater
@@ -368,6 +371,16 @@ message UserUpdater {
     optional string username = 13;
     optional NSFWSetting nsfwSetting = 14;
     optional SearchEngine searchEngine = 15;
+
+    message LlmEngineOption {
+        optional LLMEngine llmEngineOption = 1;
+    }
+    optional LlmEngineOption llmEngine = 16;
+
+    message LlmApiKeyOption {
+        optional string llmApiKeyOption = 1;
+    }
+    optional LlmApiKeyOption llmApiKey = 17;
 }
 
 // Define UserSession
@@ -572,6 +585,12 @@ message ImportFailedSource {
     string error = 2;
 }
 
+// Define LLMEngine
+
+enum LLMEngine {
+    OpenAI = 0;
+}
+
 // Define me.binwang.rss.service.ArticleService
 
 message GetArticlesBySourceRequest {
@@ -1275,6 +1294,16 @@ message UpdateUserSettingsRequest {
     string token = 1;
     optional NSFWSetting nsfwSetting = 2;
     optional SearchEngine searchEngine = 3;
+
+    message LlmEngineOption {
+        optional LLMEngine llmEngineOption = 1;
+    }
+    optional LlmEngineOption llmEngine = 4;
+
+    message LlmApiKeyOption {
+        optional string llmApiKeyOption = 1;
+    }
+    optional LlmApiKeyOption llmApiKey = 5;
 }
 
 message RemoveCurrentFolderAndSourceRequest {

src/main/resources/application.conf

@@ -115,8 +115,7 @@ search {
 }
 
 open-ai {
-    apiKey = ""
-    model = "gpt-3.5-turbo"
+    model = "gpt-4o-mini"
 }
 
 image-proxy {

src/main/scala/me/binwang/rss/cmd/Services.scala

@@ -3,7 +3,7 @@ package me.binwang.rss.cmd
 import cats.effect.{IO, Resource}
 import cats.implicits._
 import com.typesafe.config.ConfigFactory
-import me.binwang.rss.llm.OpenAILLM
+import me.binwang.rss.llm.{LLMModels, OpenAILLM}
 import me.binwang.rss.model.ImportLimit
 import me.binwang.rss.service._
 import me.binwang.rss.sourcefinder.{HtmlSourceFinder, MultiSourceFinder, RegexSourceFinder}
@@ -37,7 +37,7 @@ object Services {
   def apply(baseServer: BaseServer): Resource[IO, Services] = {
 
     val authorizer: Authorizer = new Authorizer(throttler, baseServer.userSessionDao, baseServer.folderDao)
-    val llm = new OpenAILLM(baseServer.sttpBackend)
+    val llmModels = LLMModels(openAI = new OpenAILLM(baseServer.sttpBackend))
 
     val importLimit = ImportLimit(
       paidFolderCount = Try(config.getInt("import.limit.paid-user-folders")).toOption,
@@ -54,7 +54,7 @@ object Services {
 
       new Services(
         new ArticleService(baseServer.articleDao, baseServer.articleContentDao, baseServer.articleUserMarkingDao,
-          baseServer.articleSearchDao, llm, authorizer),
+          baseServer.articleSearchDao, baseServer.userDao, llmModels, authorizer),
         new FolderService(baseServer.folderDao, baseServer.folderSourceDao,
           baseServer.sourceDao, baseServer.importSourcesTaskDao, authorizer, importLimit),
         new SourceService(baseServer.sourceDao, baseServer.folderSourceDao, baseServer.folderDao, baseServer.fetcher,

src/main/scala/me/binwang/rss/dao/sql/ArticleEmbeddingTaskSqlDao.scala

@@ -52,6 +52,7 @@ class ArticleEmbeddingTaskSqlDao(implicit val connectionPool: ConnectionPool) ex
         .filter(_.status == lift(EmbeddingUpdateStatus.PENDING))
         .sortBy(_.scheduledAt)(Ord.asc)
         .take(lift(size))
+        .forUpdate()
     }
 
     val q = for {

src/main/scala/me/binwang/rss/dao/sql/BaseSqlDao.scala

@@ -13,6 +13,7 @@ import me.binwang.rss.model.ArticleOrder.ArticleOrder
 import me.binwang.rss.model.EmbeddingUpdateStatus.EmbeddingUpdateStatus
 import me.binwang.rss.model.FetchStatus.FetchStatus
 import me.binwang.rss.model.ID.ID
+import me.binwang.rss.model.LLMEngine.LLMEngine
 import me.binwang.rss.model.MoreLikeThisType.MoreLikeThisType
 import me.binwang.rss.model.NSFWSetting.NSFWSetting
 import me.binwang.rss.model._
@@ -95,6 +96,13 @@ trait BaseSqlDao {
   protected implicit val encodeNsfwSetting: MappedEncoding[NSFWSetting, String] =
     MappedEncoding[NSFWSetting, String](_.toString)
 
+  protected implicit val decodeLLMEngine: MappedEncoding[String, LLMEngine] =
+    MappedEncoding[String, LLMEngine](LLMEngine.withName)
+
+  protected implicit val encodeLLMSetting: MappedEncoding[LLMEngine, String] =
+    MappedEncoding[LLMEngine, String](_.toString)
+
+
   protected implicit val mediaGroupsEncoder: Encoder[MediaGroups] = encoder(java.sql.Types.OTHER, (index, mediaGroups, row) => {
     val value = io.circe.syntax.EncoderOps(mediaGroups).asJson.toString()
     val pgObj = new PGobject()

src/main/scala/me/binwang/rss/dao/sql/SourceSqlDao.scala

@@ -107,6 +107,7 @@ class SourceSqlDao(implicit val connectionPool: ConnectionPool) extends SourceDa
         .filter(_.fetchStatus == lift(FetchStatus.SCHEDULED))
         .sortBy(source => source.fetchScheduledAt) (Ord.asc)
         .take(lift(size))
+        .forUpdate()
     }
     /*
     This doesn't work because it returns String instead of List[String]

src/main/scala/me/binwang/rss/dao/sql/UserSqlDao.scala

@@ -34,7 +34,9 @@ class UserSqlDao(implicit val connectionPool: ConnectionPool) extends UserDao wi
             subscribeEndAt timestamp not null,
             subscribed boolean not null default false,
             nsfwSetting varchar not null default 'BLUR',
-            searchEngine jsonb not null
+            searchEngine jsonb not null,
+            llmEngine varchar default null,
+            llmApiKey varchar default null
           )
          """)
       .update
@@ -94,6 +96,8 @@ class UserSqlDao(implicit val connectionPool: ConnectionPool) extends UserDao wi
         setOpt(_.username, updater.username),
         setOpt(_.nsfwSetting, updater.nsfwSetting),
         setOpt(_.searchEngine, updater.searchEngine),
+        setOpt(_.llmEngine, updater.llmEngine),
+        setOpt(_.llmApiKey, updater.llmApiKey),
       )
     run(q).transact(xa).map(_ > 0)
   }

src/main/scala/me/binwang/rss/grpc/generator/GenerateGRPC.scala

@@ -7,6 +7,7 @@ import me.binwang.rss.model.ArticleListLayout.ArticleListLayout
 import me.binwang.rss.model.ArticleOrder.ArticleOrder
 import me.binwang.rss.model.FetchStatus.FetchStatus
 import me.binwang.rss.model.ID.ID
+import me.binwang.rss.model.LLMEngine.LLMEngine
 import me.binwang.rss.model.MoreLikeThisType.MoreLikeThisType
 import me.binwang.rss.model.NSFWSetting.NSFWSetting
 import me.binwang.rss.model._
@@ -88,6 +89,7 @@ object GenerateGRPC extends GRPCGenerator {
     typeOf[SearchTerms],
     typeOf[ImportSourcesTask],
     typeOf[ImportFailedSource],
+    typeOf[LLMEngine],
   )
 
   override val serviceClasses: Seq[Type] = Seq(

src/main/scala/me/binwang/rss/llm/LLMModels.scala

@@ -0,0 +1,12 @@
+package me.binwang.rss.llm
+
+import me.binwang.rss.model.LLMEngine
+import me.binwang.rss.model.LLMEngine.LLMEngine
+
+case class LLMModels(
+  openAI: OpenAILLM,
+) {
+  def getModel(llmEngine: LLMEngine): LargeLanguageModel = llmEngine match {
+    case LLMEngine.OpenAI => openAI
+  }
+}

src/main/scala/me/binwang/rss/llm/LargeLanguageModel.scala

@@ -2,6 +2,7 @@ package me.binwang.rss.llm
 
 import cats.effect.IO
 import me.binwang.rss.llm.LargeLanguageModel.initMessage
+import me.binwang.rss.model.Article
 import org.typelevel.log4cats.LoggerFactory
 
 case class ChatMessage(
@@ -22,27 +23,26 @@ trait LargeLanguageModel {
   private val logger = LoggerFactory.getLoggerFromClass[IO](this.getClass)
 
   // return multiple choices based on previous message
-  def chat(messages: Seq[ChatMessage]): IO[Seq[ChatMessage]]
+  def chat(messages: Seq[ChatMessage], apiKey: String): IO[Seq[ChatMessage]]
 
-  def getRecommendSearchQueries(likedArticleTitles: Seq[String], size: Int): IO[Seq[String]] = {
-    // TODO: replace "videos" in prompt based on the type of articles
+  def getRecommendSearchQueries(articles: Seq[Article], size: Int, apiKey: String): IO[Seq[String]] = {
     val interestQueryStr =
-      """Here are some recent videos in a user's subscription feed, what is his interest?
+      """Here are some recent posts in a user's subscription feed, what is his interest?
         | Which languages the user speak?""".stripMargin + "\n\n" +
-        likedArticleTitles.map(t => s"* $t").mkString("\n")
+        articles.map(a => s"* ${a.title}, posted at ${a.postedAt.toLocalDateTime}").mkString("\n")
     val interestQuery = ChatMessage(role = "user", content = interestQueryStr)
     val initReq = Seq(initMessage, interestQuery)
     for {
-      interestChoices <- chat(initReq)
+      interestChoices <- chat(initReq, apiKey)
       recommendQueryStr = s"""Based on that, could you help me to come with $size search queries so that I can find more
-        | interesting videos for him/her on the Internet? Better to use all the languages the user speaks.
-        | Show only the search queries without any explanation. One query per line.""".stripMargin
+        | interesting articles or videos for him/her on the Internet? Better to use all the languages the user speaks.
+        | Show only the search queries without any explanation. One query per line. Do not include number or quotes""".stripMargin
       recommendQuery = ChatMessage(role = "user", content = recommendQueryStr)
       nextQueries = initReq :+ interestChoices.head :+ recommendQuery
-      recommendChoices <- chat(nextQueries)
+      recommendChoices <- chat(nextQueries, apiKey)
       // TODO: change to debug (or output to a seperate log file) after tuning results
       _ <- logger.info(s"LLM prompts and results for recommend search. Query: $nextQueries, result: $recommendChoices")
-      result = recommendChoices.head.content.split('\n').map(removeLeadingListNumber)
+      result = recommendChoices.head.content.split('\n').map(removeLeadingListNumber).filter(_.nonEmpty)
     } yield result
   }
 

src/main/scala/me/binwang/rss/llm/OpenAILLM.scala

@@ -22,10 +22,9 @@ case class OpenAIChatResponse(
 
 class OpenAILLM(backend: SttpBackend[IO, _])(implicit val loggerFactory: LoggerFactory[IO]) extends LargeLanguageModel {
 
-  private val apiKey = ConfigFactory.load().getString("open-ai.apiKey")
   private val model = ConfigFactory.load().getString("open-ai.model")
 
-  override def chat(messages: Seq[ChatMessage]): IO[Seq[ChatMessage]] = {
+  override def chat(messages: Seq[ChatMessage], apiKey: String): IO[Seq[ChatMessage]] = {
     val req = OpenAIChatRequest(model, messages)
     basicRequest
       .post(uri"https://api.openai.com/v1/chat/completions")

src/main/scala/me/binwang/rss/model/Errors.scala

@@ -55,6 +55,9 @@ final case class UserCannotBeActivated(email: String)
 final case class UserDeleteCodeInvalidException(code: String)
   extends ServerException(code = 20009, msg = s"User delete verification code $code is not valid")
 
+final case class LLMEngineNotConfigured(userID: String)
+  extends ServerException(code = 20010, msg = s"LLM engine not configured for user $userID")
+
 // Source service
 
 final case class SourceNotFound(sourceID: String)

src/main/scala/me/binwang/rss/model/User.scala

@@ -1,8 +1,9 @@
 package me.binwang.rss.model
 
+import me.binwang.rss.model.LLMEngine.LLMEngine
+
 import java.time.ZonedDateTime
 import java.util.UUID
-
 import me.binwang.rss.model.NSFWSetting.NSFWSetting
 
 object NSFWSetting extends Enumeration {
@@ -14,6 +15,13 @@ object NSFWSetting extends Enumeration {
   = Value
 }
 
+object LLMEngine extends Enumeration {
+  type LLMEngine = Value
+  val
+  OpenAI
+  = Value
+}
+
 case class SearchEngine(name: Option[String], urlPrefix: String)
 
 object SearchEngine {
@@ -45,6 +53,8 @@ case class User (
   subscribed: Boolean = false,
   nsfwSetting: NSFWSetting = NSFWSetting.BLUR,
   searchEngine: SearchEngine = SearchEngine.DEFAULT,
+  llmEngine: Option[LLMEngine] = None,
+  llmApiKey: Option[String] = None,
 ) {
 
   def toInfo: UserInfo = {
@@ -61,6 +71,7 @@ case class User (
       subscribed = subscribed,
       nsfwSetting = nsfwSetting,
       searchEngine = searchEngine,
+      llmEngine = llmEngine,
     )
   }
 
@@ -78,7 +89,8 @@ case class UserInfo (
   currentSourceID: Option[String] = None,
   subscribed: Boolean = false,
   nsfwSetting: NSFWSetting = NSFWSetting.BLUR,
-  searchEngine: SearchEngine = SearchEngine.DUCKDUCKGO,
+  searchEngine: SearchEngine = SearchEngine.DEFAULT,
+  llmEngine: Option[LLMEngine] = None,
 )
 
 case class UserUpdater (
@@ -97,4 +109,6 @@ case class UserUpdater (
   username: Option[String] = None,
   nsfwSetting: Option[NSFWSetting] = None,
   searchEngine: Option[SearchEngine] = None,
+  llmEngine: Option[Option[LLMEngine]] = None,
+  llmApiKey: Option[Option[String]] = None,
 )