XiFan TalkCode
Early Access

XiFan TalkCode

Voice-to-prompt assistant for developers. Speak your coding instructions naturally, TalkCode transcribes and refines them into high-quality prompts — ready to paste into Claude Code, XiFan Coder, Codex, Cursor, or any AI coding agent.

Electron 36 Whisper.cpp TypeScript MIT License

Core Features

Smart Speech Recognition

Supports 58 languages via Whisper.cpp (local) and Parakeet (cloud). Accurate transcription optimized for technical vocabulary, variable names, and code syntax.

AI-Powered Enhancement

Connect GPT, Claude, Gemini, Groq, or any local GGUF model via llama.cpp. The AI layer refines raw transcriptions into well-structured prompts — fixing grammar, adding context, and formatting for optimal AI agent input.

Privacy First

100% local processing mode — audio never leaves your device. API keys stored in SecureCache with OS-level encryption. Zero telemetry, zero analytics.

Custom Dictionary

Train TalkCode on your project's domain terms, function names, and brand words. Custom dictionaries improve transcription accuracy for your specific codebase.

Agent Naming

Name your AI assistant and issue voice commands directly. Say "Hey Kira, refactor the auth module" — TalkCode recognizes the wake word and routes your instruction as a refined prompt.

Cross-Platform

Runs on macOS, Windows, and Linux via Electron 36. A single consistent experience across all platforms with native OS integration for audio capture and key storage.

Product Editions

TalkCode Lite

Free

  • Cloud-based speech recognition
  • 58 language support
  • Basic voice-to-text for coding
  • App Store review in progress
Coming to App Store

TalkCode

Early Access

Full Version

  • Local + cloud speech recognition
  • AI enhancement (GPT, Claude, Gemini, Groq, local GGUF)
  • Custom dictionary & agent naming
  • Cross-platform: macOS, Windows, Linux
Download Now

Supported AI Models

Cloud Inference

  • OpenAI
  • GPT-5.2, GPT-5 Mini
  • GPT-4.1
  • Anthropic
  • Claude Sonnet 4.5
  • Claude Haiku 4.5, Opus 4.5
  • Google
  • Gemini 3 Pro, Flash
  • Gemini 2.5 Flash Lite
  • Groq
  • Qwen3, GPT-OSS, LLaMA

Local Inference

GGUF format via llama.cpp

  • Qwen3 (1.7B – 32B)
  • Qwen2.5-Coder (0.5B – 32B)
  • LLaMA 3.x (8B, 70B)
  • Mistral (7B, 8x7B)

Speech Recognition

Local

  • Whisper.cpp (tiny → large-v3)
  • Parakeet (offline)

Cloud

  • OpenAI Whisper API
  • Groq Whisper

Privacy & Security

TalkCode is designed from the ground up to keep your code and voice data private.

100% Local Processing

In local mode, all audio capture and speech-to-text processing happens on your device. Your voice recordings and transcriptions never leave your machine.

Encrypted Key Storage

API keys are stored in SecureCache with OS-level encryption (Keychain on macOS, Credential Manager on Windows, libsecret on Linux). Keys are never written to plain text files.

Context Isolation (Electron)

Built on Electron 36 with contextIsolation enabled and nodeIntegration disabled. Renderer processes are sandboxed, preventing malicious content from accessing system APIs.

No Telemetry

No analytics SDKs, no crash reporters, no usage tracking. TalkCode communicates only with the LLM and speech recognition endpoints you explicitly configure.

Privacy Policy

XiFan TalkCode supports full local processing mode where all audio capture and speech-to-text conversion happens entirely on your device. In this mode, voice recordings and transcriptions never leave your machine. When using cloud STT or AI enhancement providers, data is sent only to the endpoints you explicitly configure.

API keys are encrypted via SecureCache using OS-level encryption (Keychain on macOS, Credential Manager on Windows, libsecret on Linux). No analytics, crash reporting, or telemetry SDKs are included.

Contact

If you have questions about this privacy policy, contact us at tech-support01@e2ec.biz

Last updated: April 12, 2026