Core Spotlightを利用したLLM検索

0:59 - Ask the model with a Foundation Models session

let response = try await session.respond(to: "What are some nice hikes near water?")

4:20 - Set up SpotlightSearchTool

// Set up SpotlightSearchTool
  import CoreSpotlight
  import FoundationModels

  // In one line, the tool is ready to search your app's Core Spotlight index
  let tool = SpotlightSearchTool()

  // Or provide a custom configuration — e.g. search file paths in your app's sandbox
  let fileTool = SpotlightSearchTool(
      configuration: .init(
          sources: [
              .files
          ]
      )
  )

4:50 - Add SpotlightSearchTool to a session

// Add SpotlightSearchTool to a session
  import CoreSpotlight
  import FoundationModels
  
  let tool = SpotlightSearchTool()

  let session = LanguageModelSession(model: model, tools: [tool], instructions: instructions)

  let response = try await session.respond(to: "What hikes have I gone on?")

6:24 - Implement an index delegate

// Implement an index delegate
  import CoreSpotlight

  class IndexDelegate: NSObject, CSSearchableIndexDelegate {

      // Called when the index requests searchable items for the provided identifiers
      func searchableItems(forIdentifiers identifiers: [String]) async -> [CSSearchableItem] {
          let entries = await mystore.fetchEntries(ids: identifiers)
          return entries.map { makeSearchableItem(from: $0) }
      }
  }

7:37 - Track the query token for refresh

// Track the query token for refresh
  import CoreSpotlight
  import FoundationModels

  let tool = SpotlightSearchTool()

  for await reply in tool.searchResults {
  
      if reply.queryToken != currentToken {
          // New query — start a new display section
          currentToken = reply.queryToken
      }

      switch reply.content {
      case .items(let searchItems):
      }
  }

8:42 - Set a dynamic guidance profile

// Set a dynamic guidance profile
  import CoreSpotlight
  import FoundationModels

  let profile = SpotlightSearchTool.GuidanceProfile(
      textMatch: true,
      dates: true,
      people: false,
      attributes: [.title, .altitude, .completionDate]
  )

  let tool = SpotlightSearchTool(
      configuration: .init(
          guide: .init(level: .dynamic(profile))
      )
  )

  // On-device models have smaller context — prefer focused guidance
  let focusedTool = SpotlightSearchTool(
      configuration: .init(
          guide: .init(level: .focused(.items))
      )
  )

9:32 - Implement a ContactResolver

// Implement a ContactResolver
  import CoreSpotlight
  import FoundationModels

  struct MyContactResolver: ContactResolver {
  
      func userIdentity() -> ResolvedContact {
          // Pull from whatever identity source your app has —
          // account profile, Contacts framework, sign-in session, etc.
          var contact = ResolvedContact(displayName: "Jane Doe")
          contact.emailAddresses = ["jane@example.com", "jdoe@work.com"]
          contact.names = ["Jane", "JD"]
          return contact
      }
  }
  
  tool.contactResolver = MyContactResolver()

11:34 - Define a custom stage

// Define a custom stage
  import CoreSpotlight
  import FoundationModels

  @Generable
  struct HappinessStage: CustomStage {
      static var name = "happiness"
      static var description = "Scores hike by how happy the author was"
      static var inputTypes: [SearchPipelineDataType] = [.items]
      static var outputTypes: [SearchPipelineDataType] = [.scoredItems]

      @Guide(description: "Minimum happiness score (0.0-1.0) to include in results")
      var threshold: Double?

      func execute(on input: SearchPipelineData) async throws -> SearchPipelineData {
          return SearchPipelineData(payload: .scoredItems(sorted))
      }
  }

  // Register the stage by adding it to the tool's configuration
  let tool = SpotlightSearchTool(configuration: .init(
      customStages: [.happinessBoost(threshold: 0.5)])
  )

12:10 - Handle a reply data types

// Handle a reply data types
  import CoreSpotlight
  import FoundationModels

  for await reply in tool.searchResults {

      let label = reply.label
      case .items(let searchItems):
      case .scoredItems(let scored):
      case .groupedItems(let groups):
      case .count(let count):
      case .table(let table):
      case .statistic(let statistic):
      case .text(let text):
          continue
      }   
  }

13:47 - Define an evaluation dataset with ModelSampleProtocol

// Evaluations
  import Evaluations
  
  struct TrailRequest: ModelSampleProtocol {
  
      typealias ExpectedValue = String                    // sample response
      typealias Expectation   = TrajectoryExpectation     
      
      var input:  ModelSampleInput
      var output: ModelSampleOutput<String, TrajectoryExpectation>
      
      var expectedIdentifiers: [String]
  }

15:06 - Define the trajectory expectation

// Evaluations
  import Evaluations
  
  TrajectoryExpectation(
      unordered: [
          ToolExpectation("searchSpotlight", arguments: [.keyOnly(argumentName: "query")])
      ]   
  )

15:17 - Run the evaluation test —

@Test("Trail search evaluation meets quality thresholds")
  func trailSearchEval() async throws {
  
      let items = try Self.loadItems()
      let samples = try Self.loadSamples()
      
      try await Self.indexDelegate.indexSearchableItems(items)
      let tool = Self.makeSearchTool()
      
      let evaluation = TrailSearchEvaluation(
          tool: tool,
          dataset: ArrayLoader(samples: samples)
      )   
      
      let result = try await evaluation.run()
      let coverageMean = result.aggregateValue(.mean(of: Metric("ResultCoverage")))
      #expect(coverageMean >= 0.5, "Result coverage should be at least 50% across queries")
  }