Open source Grok Bot alternative, Bring your own agent and subscription: message Claude Code, Codex, Gemini CLI, Cursor, Copilot and 40+ AI agents as bots on your own computer, from iPhone, Mac, Windows, Linux or terminal. Agent-agnostic and free. Also a Muse and Dots alternative. Built in Rust.
Rust acpagent-client-protocolai-agentsai-assistantclaudeclaude-codecodexcoding-agentscursorgeminigemini-cligrok-bot-alternativeiosiphonelinuxmacosmcpopen-sourcerustself-hosted
feat(kit): default voice calls to gpt-transcribe
leepokai20ad6910b699
diff --git a/apps/ios/WhatsNew.md b/apps/ios/WhatsNew.mdindex 3b098248..2bd477ef 100644--- a/apps/ios/WhatsNew.md+++ b/apps/ios/WhatsNew.md@@ -5,12 +5,12 @@ listing what iPhone users notice. tools/asc-submit.py reads the released versiona version without one gets a generic "Bug fixes and improvements" line. Delete old sections.-->-## 2.6.2+## 2.6.3### zh-Hant-- 輸入框旁的 + 按鈕改成和輸入框一樣高,兩者對齊。+- 語音通話的「語音轉文字」預設改用 OpenAI 最新的 gpt-transcribe,辨識更準,也更認得 bot 的名字。### en-US-- The + button next to the message field now matches the field's height.+- Voice calls now use OpenAI's newest speech-to-text, gpt-transcribe, by default: more accurate, and better at your bots' names.diff --git a/docs/features/voice-call.md b/docs/features/voice-call.mdindex c145ffcf..7c2eba43 100644--- a/docs/features/voice-call.md+++ b/docs/features/voice-call.md@@ -95,8 +95,12 @@ Each cloud engine has two **modes**:speech-to-text and is sent as an ordinary message; replies are read with its text-to-speech. Bothrun on the computer (`voiceTranscribe`, `voiceSpeak`), so the key never leaves it; utterances stopat 20 s and replies are spoken in clips (OpenAI 600 characters as AAC, Gemini 150 as WAV) so every- request and answer fits one 1 MiB channel message. Defaults: `gpt-4o-transcribe` +- `gpt-4o-mini-tts`, `gemini-3.8-flash` + `gemini-3.8-flash-tts`.+ request and answer fits one 1 MiB channel message. Defaults: `gpt-transcribe` ++ `gpt-4o-mini-tts`, `gemini-3.8-flash` + `gemini-3.8-flash-tts`. Speech-to-text gets the app's and+ the bots' names as hints (gpt-transcribe also as `keywords`). gpt-transcribe returns Chinese in+ Simplified characters whatever the prompt says, so the client turns it into Traditional+ (`Hans-Hant`) for a reader whose first Chinese language, or region, is Traditional. Only a+ model the user picks is saved; the default isn't, so a newer default reaches everyone.Voice settings (`VoiceChatSettingsView`) open from the call bar's gear and from settings: on theiPhone, **Computers & settings → Voice chat**; on the Mac, **Account → Voice chat** (this Mac'sdiff --git a/host/src/voice.rs b/host/src/voice.rsindex 3b9ba836..d884a8a9 100644--- a/host/src/voice.rs+++ b/host/src/voice.rs@@ -273,11 +273,29 @@ fn classify(p: Provider, models: &[Model]) -> (Vec<String>, Vec<String>, Vec<Str(realtime, transcribe, speech)}+/// Names a speaker is likely to say that speech-to-text would otherwise misspell: the app and+/// the bots on this computer.+fn spoken_names(store: &Store) -> Vec<String> {+ let mut names = vec!["Codync".to_owned()];+ for bot in store.bots().unwrap_or_default() {+ if !bot.deleted && !names.contains(&bot.config.name) {+ names.push(bot.config.name);+ }+ }+ names.truncate(50);+ names+}++fn names_hint(names: &[String]) -> String {+ format!("Names that may come up: {}.", names.join(", "))+}+/// Speech to text for one utterance (`audio`: base64 WAV, 16 kHz mono).pub async fn transcribe(store: &Store, p: Provider, model: &str, audio: &str) -> Result<Value> {let key = key(store, p)?;let model = checked(model)?;let wav = STANDARD.decode(audio).map_err(|_| anyhow!("audio must be base64"))?;+ let names = spoken_names(store);let text = match p {Provider::OpenAi => {let boundary = format!("codync-{}", uuid::Uuid::new_v4().simple());@@ -286,10 +304,19 @@ pub async fn transcribe(store: &Store, p: Provider, model: &str, audio: &str) ->format!("--{boundary}\r\nContent-Disposition: form-data; name=\"model\"\r\n\r\n{model}\r\n").as_bytes(),);// Keeps Traditional Chinese from coming back Simplified, and code terms intact.- body.extend_from_slice(- format!("--{boundary}\r\nContent-Disposition: form-data; name=\"prompt\"\r\n\r\n{TRANSCRIBE_HINT}\r\n")- .as_bytes(),- );+ let mut field = |name: &str, value: &str| {+ body.extend_from_slice(+ format!("--{boundary}\r\nContent-Disposition: form-data; name=\"{name}\"\r\n\r\n{value}\r\n")+ .as_bytes(),+ );+ };+ field("prompt", &format!("{TRANSCRIBE_HINT} {}", names_hint(&names)));+ // Only gpt-transcribe takes keywords; the older models reject the field.+ if model.starts_with("gpt-transcribe") {+ for name in &names {+ field("keywords[]", name);+ }+ }body.extend_from_slice(format!("--{boundary}\r\nContent-Disposition: form-data; name=\"file\"; filename=\"speech.wav\"\r\nContent-Type: audio/wav\r\n\r\n"@@ -307,7 +334,7 @@ pub async fn transcribe(store: &Store, p: Provider, model: &str, audio: &str) ->}Provider::Gemini => {let body = json!({"contents": [{"parts": [- {"text": format!("Transcribe this audio exactly as spoken. {TRANSCRIBE_HINT} Reply with the transcript only; reply with nothing if there is no speech.")},+ {"text": format!("Transcribe this audio exactly as spoken. {TRANSCRIBE_HINT} {} Reply with the transcript only; reply with nothing if there is no speech.", names_hint(&names))},{"inlineData": {"mimeType": "audio/wav", "data": audio}},]}]});let v = send(p, gemini_generate(&key, model).json(&body)).await?;diff --git a/kit/Sources/CodyncKit/Models/Voice.swift b/kit/Sources/CodyncKit/Models/Voice.swiftindex d8b064c8..b86cf4ea 100644--- a/kit/Sources/CodyncKit/Models/Voice.swift+++ b/kit/Sources/CodyncKit/Models/Voice.swift@@ -39,7 +39,7 @@ public enum VoiceProvider: String, CaseIterable, Codable, Sendable {/// Speech-to-text and text-to-speech defaults for the non-realtime mode.public var transcribeModel: String {switch self {- case .openAI: "gpt-4o-transcribe"+ case .openAI: "gpt-transcribe"case .gemini: "gemini-3.8-flash"}}diff --git a/kit/Sources/CodyncUI/Call/CallView.swift b/kit/Sources/CodyncUI/Call/CallView.swiftindex 191dcb1e..a7a2fb99 100644--- a/kit/Sources/CodyncUI/Call/CallView.swift+++ b/kit/Sources/CodyncUI/Call/CallView.swift@@ -74,7 +74,7 @@ struct CallView: View {return client}engine = CloudSpeechEngine(provider: provider, speech: .init(- transcribe: { try await client().voiceTranscribe(provider, model: transcribe, wav: $0) },+ transcribe: { try await VoiceSettings.inReadersScript(client().voiceTranscribe(provider, model: transcribe, wav: $0)) },speak: { try await client().voiceSpeak(provider, model: speechModel, voice: voice, text: $0) },send: { [model, botId] in model.send($0, to: botId) }))diff --git a/kit/Sources/CodyncUI/Call/VoiceChatSettingsView.swift b/kit/Sources/CodyncUI/Call/VoiceChatSettingsView.swiftindex 06b41561..bfa50f23 100644--- a/kit/Sources/CodyncUI/Call/VoiceChatSettingsView.swift+++ b/kit/Sources/CodyncUI/Call/VoiceChatSettingsView.swift@@ -144,9 +144,13 @@ private struct ProviderVoiceSettings: View {Color.clear.frame(height: 0).task { await load() }.onChange(of: voice) { _, v in UserDefaults.standard.set(v, forKey: VoiceSettings.voiceKey(provider)) }- .onChange(of: model) { _, m in UserDefaults.standard.set(m, forKey: VoiceSettings.modelKey(provider)) }- .onChange(of: transcribeModel) { _, m in UserDefaults.standard.set(m, forKey: VoiceSettings.transcribeKey(provider)) }- .onChange(of: speechModel) { _, m in UserDefaults.standard.set(m, forKey: VoiceSettings.speechKey(provider)) }+ .onChange(of: model) { _, m in VoiceSettings.save(m, default: provider.model, forKey: VoiceSettings.modelKey(provider)) }+ .onChange(of: transcribeModel) { _, m in+ VoiceSettings.save(m, default: provider.transcribeModel, forKey: VoiceSettings.transcribeKey(provider))+ }+ .onChange(of: speechModel) { _, m in+ VoiceSettings.save(m, default: provider.speechModel, forKey: VoiceSettings.speechKey(provider))+ }}/// The key is on the computer: say so, and offer test / replace / remove.diff --git a/kit/Sources/CodyncUI/Call/VoiceEngine.swift b/kit/Sources/CodyncUI/Call/VoiceEngine.swiftindex 7f19b5bb..5c8990ec 100644--- a/kit/Sources/CodyncUI/Call/VoiceEngine.swift+++ b/kit/Sources/CodyncUI/Call/VoiceEngine.swift@@ -63,7 +63,7 @@ enum VoiceSettings {UserDefaults.standard.string(forKey: modeKey(provider)) == "speech"}- static func transcribeKey(_ provider: VoiceProvider) -> String { "callTranscribeModel.\(provider.rawValue)" }+ static func transcribeKey(_ provider: VoiceProvider) -> String { "callSpeechToText.\(provider.rawValue)" }static func speechKey(_ provider: VoiceProvider) -> String { "callSpeechModel.\(provider.rawValue)" }static func transcribeModel(_ provider: VoiceProvider) -> String {@@ -74,6 +74,26 @@ enum VoiceSettings {UserDefaults.standard.string(forKey: speechKey(provider)) ?? provider.speechModel}+ /// Keeps a model the user picked; the default isn't stored, so a newer default reaches them.+ static func save(_ model: String, default fallback: String, forKey key: String) {+ if model.isEmpty || model == fallback {+ UserDefaults.standard.removeObject(forKey: key)+ } else {+ UserDefaults.standard.set(model, forKey: key)+ }+ }++ /// gpt-transcribe writes Chinese in Simplified characters whatever it's told: turn it back+ /// into Traditional for someone who reads Traditional (their first Chinese language, or a+ /// Traditional region when none is listed).+ static func inReadersScript(_ text: String) -> String {+ let chinese = Locale.preferredLanguages.lazy.map(Locale.Language.init(identifier:)).first { $0.languageCode == .chinese }+ let traditional = if let chinese { chinese.script == .hanTraditional }+ else { ["TW", "HK", "MO"].contains(Locale.current.region?.identifier ?? "") }+ guard traditional else { return text }+ return text.applyingTransform(StringTransform("Hans-Hant"), reverse: false) ?? text+ }+static func voiceKey(_ provider: VoiceProvider) -> String { "callVoice.\(provider.rawValue)" }static func voice(_ provider: VoiceProvider) -> String {