Skip to content

Fastest and only macOS Dictation app with on-device STT and custom trained AI enhancement model. Windows, iOS testflight! Android, Linux waitlist open! A local Wispr Flow alternative. Follow https://x.com/fluidvoiceapp for more

fix: avoid typing AI errors

altic-deve64bfaaee0ea
diff --git a/Sources/Fluid/ContentView.swift b/Sources/Fluid/ContentView.swift
index 0d53b10b..6f70183b 100644
--- a/Sources/Fluid/ContentView.swift
+++ b/Sources/Fluid/ContentView.swift
@@ -13,6 +13,22 @@ import CoreGraphics
import Security
import SwiftUI
+// MARK: - AI Processing Errors
+
+enum AIProcessingError: LocalizedError {
+ case missingAPIKey(provider: String)
+ case emptyResponse
+
+ var errorDescription: String? {
+ switch self {
+ case let .missingAPIKey(provider):
+ return "API key not set for \(provider)"
+ case .emptyResponse:
+ return "AI returned an empty response"
+ }
+ }
+}
+
// MARK: - Sidebar Item Enum
enum SidebarItem: Hashable {
@@ -69,6 +85,7 @@ enum ShortcutRecordingTarget: String, Hashable {
// NOTE: Streaming and AI response parsing is now handled by LLMClient
+// swiftlint:disable type_body_length
struct ContentView: View {
private enum ActiveRecordingMode: String {
case none
@@ -1499,7 +1516,7 @@ struct ContentView: View {
_ inputText: String,
overrideSystemPrompt: String? = nil,
dictationSlot: SettingsStore.DictationShortcutSlot? = nil
- ) async -> String {
+ ) async throws -> String {
// CRITICAL FIX: Read current settings from SettingsStore, not stale @State copies
// This ensures AI provider/model changes in AISettingsView take effect immediately
let currentSelectedProviderID = SettingsStore.shared.selectedProviderID
@@ -1595,7 +1612,7 @@ struct ContentView: View {
self.logDictationPromptTrace("Selected context text", value: "<none (dictation mode)>")
}
DebugLogger.shared.debug("Using Apple Intelligence for transcription cleanup", source: "ContentView")
- let output = await provider.process(systemPrompt: systemPrompt, userText: userMessageContent)
+ let output = try await provider.process(systemPrompt: systemPrompt, userText: userMessageContent)
if self.shouldTracePromptProcessing {
self.logDictationPromptTrace("Model answer (A)", value: output)
}
@@ -1611,7 +1628,7 @@ struct ContentView: View {
if !isLocal {
guard !apiKey.trimmingCharacters(in: CharacterSet.whitespacesAndNewlines).isEmpty else {
- return "Error: API Key not set for \(derivedCurrentProvider)"
+ throw AIProcessingError.missingAPIKey(provider: derivedCurrentProvider)
}
}
@@ -1704,26 +1721,24 @@ struct ContentView: View {
DebugLogger.shared.info("Using LLMClient for transcription (streaming=\(enableStreaming))", source: "ContentView")
- do {
- let response = try await LLMClient.shared.call(config)
-
- // Log thinking if present (for debugging)
- if let thinking = response.thinking {
- DebugLogger.shared.debug("LLM thinking tokens extracted (\(thinking.count) chars)", source: "ContentView")
- if self.shouldTracePromptProcessing {
- self.logDictationPromptTrace("Model thinking", value: thinking)
- }
- }
+ let response = try await LLMClient.shared.call(config)
+ // Log thinking if present (for debugging)
+ if let thinking = response.thinking {
+ DebugLogger.shared.debug("LLM thinking tokens extracted (\(thinking.count) chars)", source: "ContentView")
if self.shouldTracePromptProcessing {
- self.logDictationPromptTrace("Model answer (A)", value: response.content)
+ self.logDictationPromptTrace("Model thinking", value: thinking)
}
+ }
- return response.content.isEmpty ? "<no content>" : response.content
- } catch {
- DebugLogger.shared.error("AI API error: \(error.localizedDescription)", source: "ContentView")
- return "Error: \(error.localizedDescription)"
+ if self.shouldTracePromptProcessing {
+ self.logDictationPromptTrace("Model answer (A)", value: response.content)
+ }
+
+ guard !response.content.isEmpty else {
+ throw AIProcessingError.emptyResponse
}
+ return response.content
}
// MARK: - Streaming Response Handler (DEPRECATED - Now handled by LLMClient)
@@ -1794,9 +1809,13 @@ struct ContentView: View {
promptTest.isProcessing = false
}
- let result = await self.processTextWithAI(transcribedText, overrideSystemPrompt: promptTest.draftPromptText)
- let finalText = ASRService.applyGAAVFormatting(result)
- promptTest.lastOutputText = finalText
+ do {
+ let result = try await self.processTextWithAI(transcribedText, overrideSystemPrompt: promptTest.draftPromptText)
+ promptTest.lastOutputText = ASRService.applyGAAVFormatting(result)
+ } catch {
+ DebugLogger.shared.error("Prompt test AI call failed: \(error.localizedDescription)", source: "ContentView")
+ promptTest.lastError = error.localizedDescription
+ }
return
}
@@ -1849,11 +1868,21 @@ struct ContentView: View {
// Ensure the status label becomes visible immediately.
await Task.yield()
- finalText = await self.processTextWithAI(
- transcribedText,
- overrideSystemPrompt: promptOverride,
- dictationSlot: activeDictationSlot
- )
+ do {
+ finalText = try await self.processTextWithAI(
+ transcribedText,
+ overrideSystemPrompt: promptOverride,
+ dictationSlot: activeDictationSlot
+ )
+ } catch {
+ // Fall back to the raw transcription so the user still gets
+ // their words typed instead of an error string.
+ DebugLogger.shared.error(
+ "AI post-processing failed, falling back to raw transcription: \(error.localizedDescription)",
+ source: "ContentView"
+ )
+ finalText = transcribedText
+ }
let postProcessingLatencyMs = Int((Date().timeIntervalSince(postProcessingStart) * 1000).rounded())
AnalyticsService.shared.capture(
.dictationPostProcessingCompleted,
@@ -2129,7 +2158,15 @@ struct ContentView: View {
var finalText = transcribedText
let shouldUseAI = DictationAIPostProcessingGate.isConfigured()
if shouldUseAI {
- finalText = await self.processTextWithAI(transcribedText)
+ do {
+ finalText = try await self.processTextWithAI(transcribedText)
+ } catch {
+ DebugLogger.shared.error(
+ "AI reprocess failed, falling back to raw transcription: \(error.localizedDescription)",
+ source: "ContentView"
+ )
+ finalText = transcribedText
+ }
}
NotchOverlayManager.shared.updateTranscriptionText("")
@@ -2889,8 +2926,13 @@ extension ContentView {
await MainActor.run { self.isCallingAI = true }
defer { Task { await MainActor.run { isCallingAI = false } } }
- let result = await processTextWithAI(aiInputText)
- await MainActor.run { self.aiOutputText = result }
+ do {
+ let result = try await processTextWithAI(aiInputText)
+ await MainActor.run { self.aiOutputText = result }
+ } catch {
+ DebugLogger.shared.error("callOpenAIChat failed: \(error.localizedDescription)", source: "ContentView")
+ await MainActor.run { self.aiOutputText = "Error: \(error.localizedDescription)" }
+ }
}
private func getModelStatusText() -> String {
@@ -3027,6 +3069,8 @@ extension ContentView {
}
}
+// swiftlint:enable type_body_length
+
private extension ContentView {
func reloadSettingsStateAfterBackupRestore() {
self.hotkeyShortcut = SettingsStore.shared.hotkeyShortcut
diff --git a/Sources/Fluid/Networking/AppleIntelligenceProvider.swift b/Sources/Fluid/Networking/AppleIntelligenceProvider.swift
index 94aa934f..59cebfef 100644
--- a/Sources/Fluid/Networking/AppleIntelligenceProvider.swift
+++ b/Sources/Fluid/Networking/AppleIntelligenceProvider.swift
@@ -44,25 +44,17 @@ enum AppleIntelligenceService {
@available(macOS 26.0, *)
final class AppleIntelligenceProvider {
/// Process text with a system prompt (for transcription cleanup)
- func process(systemPrompt: String, userText: String) async -> String {
- do {
- let session = LanguageModelSession()
-
- let fullPrompt = """
- \(systemPrompt)
-
- \(userText)
- """
-
- let response = try await session.respond(to: fullPrompt)
- return response.content
- } catch {
- DebugLogger.shared.error(
- "Apple Intelligence error: \(error.localizedDescription)",
- source: "AppleIntelligenceProvider"
- )
- return "Error: \(error.localizedDescription)"
- }
+ func process(systemPrompt: String, userText: String) async throws -> String {
+ let session = LanguageModelSession()
+
+ let fullPrompt = """
+ \(systemPrompt)
+
+ \(userText)
+ """
+
+ let response = try await session.respond(to: fullPrompt)
+ return response.content
}
/// Process rewrite/write requests with conversation history