discourse/plugins/discourse-ai/app/controllers/discourse_ai/embeddings/embeddings_controller.rb

# frozen_string_literal: true

module DiscourseAi
  module Embeddings
    class EmbeddingsController < ::ApplicationController
      requires_plugin PLUGIN_NAME

      SEMANTIC_SEARCH_TYPE = "semantic_search"

      MAX_HYDE_SEARCHES_PER_MINUTE = 4
      MAX_SEARCHES_PER_MINUTE = 100

      # Global rate limit shared across all anonymous users. This is deliberately
      # a single shared bucket — it caps the total anonymous load on the semantic
      # search endpoint, mirroring how SearchController applies a global anon limit
      # in addition to per-IP limits.
      MAX_ANON_SEARCHES_PER_MINUTE = 60

      def search
        query = params[:q].to_s
        use_hyde = SiteSetting.ai_embeddings_semantic_search_use_hyde

        if params[:hyde].present? &&
             (params[:hyde].to_s.downcase == "false" || params[:hyde].to_s == "0")
          use_hyde = false
        end

        if query.length < SiteSetting.min_search_term_length
          raise Discourse::InvalidParameters.new(:q)
        end

        grouped_results =
          Search::GroupedSearchResults.new(
            type_filter: SEMANTIC_SEARCH_TYPE,
            term: query,
            search_context: guardian,
            use_pg_headlines_for_excerpt: false,
            can_lazy_load_categories: guardian.can_lazy_load_categories?,
          )

        semantic_search = DiscourseAi::Embeddings::SemanticSearch.new(guardian)

        if use_hyde && !semantic_search.cached_query?(query)
          if current_user.present?
            RateLimiter.new(
              current_user,
              "semantic-search",
              MAX_HYDE_SEARCHES_PER_MINUTE,
              1.minute,
            ).performed!
          else
            rate_limit_anon_semantic_search(MAX_HYDE_SEARCHES_PER_MINUTE)
          end
        else
          if current_user.present?
            RateLimiter.new(
              current_user,
              "semantic-search-non-hyde",
              MAX_SEARCHES_PER_MINUTE,
              1.minute,
            ).performed!
          else
            rate_limit_anon_semantic_search(MAX_SEARCHES_PER_MINUTE)
          end
        end

        hijack do
          begin
            semantic_search
              .search_for_topics(query, _page = 1, hyde: use_hyde)
              .each { |topic_post| grouped_results.add(topic_post) }
          rescue Discourse::InvalidAccess
            return render_json_error(I18n.t("invalid_access"), status: 403)
          rescue Net::HTTPBadResponse => e
            Rails.logger.warn("Semantic search embedding generation failed: #{e.message}")
          end

          render_serialized(grouped_results, GroupedSearchResultSerializer, result: grouped_results)
        end
      end

      def quick_search
        query = params[:q].to_s

        if query.length < SiteSetting.min_search_term_length
          raise Discourse::InvalidParameters.new(:q)
        end

        grouped_results =
          Search::GroupedSearchResults.new(
            type_filter: SEMANTIC_SEARCH_TYPE,
            term: query,
            search_context: guardian,
            use_pg_headlines_for_excerpt: false,
            can_lazy_load_categories: guardian.can_lazy_load_categories?,
          )

        semantic_search = DiscourseAi::Embeddings::SemanticSearch.new(guardian)

        if !semantic_search.cached_query?(query)
          if current_user.present?
            RateLimiter.new(current_user, "semantic-search", 60, 1.minute).performed!
          else
            rate_limit_anon_semantic_search(MAX_SEARCHES_PER_MINUTE)
          end
        end

        hijack do
          begin
            semantic_search
              .search_for_topics(query, _page = 1, hyde: false)
              .each { |topic_post| grouped_results.add(topic_post) }
          rescue Net::HTTPBadResponse => e
            Rails.logger.warn("Quick search embedding generation failed: #{e.message}")
          end

          render_serialized(grouped_results, GroupedSearchResultSerializer, result: grouped_results)
        end
      end

      private

      # Applies both per-IP and global anonymous rate limits, following the same
      # pattern as SearchController#rate_limit_search. The per-IP limit prevents
      # any single anonymous user from monopolizing the endpoint, while the global
      # limit caps the total anonymous load on the system.
      def rate_limit_anon_semantic_search(per_ip_limit)
        RateLimiter.new(
          nil,
          "semantic-search-#{request.remote_ip}",
          per_ip_limit,
          1.minute,
        ).performed!

        RateLimiter.new(
          nil,
          "semantic-search-anon-global",
          MAX_ANON_SEARCHES_PER_MINUTE,
          1.minute,
        ).performed!
      end
    end
  end
end