Voice in.
Voice out.

A voice plugin for OpenCode. Dictate prompts, check transcripts, and hear coding-agent replies without leaving your terminal.

Read setup guide Explore the project

Animated terminal capture: configure voice, dictate a prompt, receive an answer, and play speech. — The real workflow, captured in the terminal.Dictate

One continuous session. Your eyes stay on the code while speech moves through the same prompt-and-response loop.

Dictate the whole thought.

Local whisper.cpp turns speech into text. Streaming mode shows words as you speak, with the unfinished tail marked separately.

/stt-record · start recording

Read it. Edit it. Then send.

By default, the transcript lands at your cursor as editable text. Submit it when ready; auto-submit and conversation mode are optional.

Transcript → prompt editor → you decide

Let the reply speak.

Choose Piper, Chatterbox, or VieNeu for speech output. Pause, replay, seek, and change playback speed from the terminal.

Piper · Chatterbox · VieNeu

Talk through a whole turn.

Conversation mode listens, waits for the completed agent reply, speaks it, then listens again. Interrupt or pause whenever you need.

listen → think → speak → listen

Built for your terminal.

Voice controls live beside your OpenCode session. Choose the pieces that fit how you work; you do not need to use every mode.

Live dictation

Watch a live waveform and transcript. Pause recording, use auto-end, or submit when you choose.

Speech that handles code

Normalize code-heavy replies into speakable phrasing, then choose a local voice engine for playback.

Vietnamese + English

VieNeu-TTS supports Vietnamese and English code-switching in a single utterance. Pick from 25 voices.

Your phone can be the mic

Route capture and playback through your own SSH audio commands when the computer's mic is not convenient.

Keep your model setup.

Text normalization uses the endpoint and model you configure. The plugin does not replace your coding model or require a specific provider.

OpenAI

Claude

Gemini

DeepSeek

Meta AI

OpenRouter

Provider names are examples, not endorsements or partnership claims. Configure an endpoint compatible with the plugin's OpenAI Chat Completions client; compatibility varies by provider.

Set up your voice.

Install the plugin in OpenCode, then configure a compatible text endpoint and your local speech tools. Full OS-specific steps and both host versions are in the setup guide.

Read the installation guide

OpenCode v1 · partial config example

{
  "plugin": [
    ["@bojackduy/opencode-voice", {
      "endpoint": "YOUR_OPENAI_COMPATIBLE_ENDPOINT",
      "model": "YOUR_MODEL"
    }]
  ]
}

Example only. Add the API key environment variable and local audio prerequisites from the guide.

Questions, answered.

Does it send my transcript automatically?

Not by default. Dictation inserts editable text at your cursor. Auto-submit and conversation mode can submit as part of their configured behavior.

Does audio leave my machine?

Default speech recognition and synthesis use local tools. Text normalization calls your configured LLM endpoint; hosted speech recognition is optional.

Can I use it hands-free?

Conversation mode listens, waits for the completed agent turn, speaks the reply, and listens again. You can pause or interrupt it.

Does it support Vietnamese?

VieNeu-TTS supports Vietnamese and English code-switching. Available language support depends on the selected speech engine and voice.

Which OpenCode hosts work?

The plugin supports OpenCode v1 and v2. See the installation guide for the config format and prerequisites for your host.

Keep your hands in flow.

Open source voice tools for your OpenCode session. Read the setup, inspect the code, make it yours.

Open the GitHub project

Voice in.Voice out.