Local & self-hosted appsCommunity / local-first stack — the opencode-voice Rust CLI (independent), the Olbrasoft/opencode-voice-plugin, and any local TTS; demoed end-to-end by the YouTube channel "No place like localhost".
opencode + voice
Decision matrix
opencode + voice app decision matrix
Overview
opencode-voice is a small community tool that lets you talk to OpenCode: hold the space bar, speak, and a local Whisper model types your request. No audio leaves your machine. It runs on macOS and Linux, not Windows, and was built for OpenCode 1; OpenCode 2 changed the server API it relies on.
Other good fits
Keeping speech private: transcription runs on your own computer with whisper.cpp, with no cloud service or API key.
+ Pros
Read pros
opencode-voice is an independent command-line tool (Apache 2.0) by a community developer, not part of OpenCode itself. It listens to your microphone, turns speech into text with Whisper running locally through whisper.cpp, and sends the text to a running OpenCode session over its HTTP interface. It doesn't speak back; answers stay on screen.
Pick it when typing is the bottleneck. Hold space to record, release to transcribe and send. A toggle mode and global hotkeys are also available.
Pick it when audio must stay local. The project states that no audio leaves your device. Whisper models range from about 75 MB (tiny) to 466 MB (small).
Pick it when you want to check before sending. You can switch off auto-submit and read the transcript first.
− Cons
It may not work with OpenCode 2. OpenCode's migration guide says the server API has new contracts in version 2 (September 2026). Stay on OpenCode's 1.x line for this tool (1.19 in October 2026), or wait for an update.
No Windows support.
Pick something else when you only need dictation. Your operating system's built-in dictation types into OpenCode Desktop's input box without extra software.
Pick something else when you want to build a voice app, not dictate code.Pipecat is a framework for voice agents, and open-source Whisper is the speech model itself.
Avoid if
You've moved to OpenCode 2 and need it to work today. The tool was last released in March 2026 for OpenCode 1's server API.
Workflow
Install the community CLI -> connect it to an already configured OpenCode server -> speak a non-sensitive sample prompt -> review the transcribed text and any resulting agent action.
Cost
Prices checked 23 September 2026: free software (Apache 2.0); you pay for the machine. Any recent Mac or Linux laptop can run the small Whisper models on its CPU. The model behind OpenCode is billed as usual.
Effort is moderate for a non-developer: a Rust toolchain to build it, microphone permissions, and two programs to start. The project is small: version 0.1.4 from March 2026 is the latest release.
Restrictions
Treat microphone access, transcripts and any voice-output component as separate data boundaries. Do not infer offline operation, a particular TTS engine or hardware support unless the chosen component documents it.
First project
Documentation-based suggestion — not run here.
Use a project folder under Git with OpenCode already connected to a model.
Install with cargo install opencode-voice (needs Rust, cmake and a C compiler).
Start OpenCode with its HTTP server on a fixed port, then start opencode-voice on the same port in a second terminal.
Switch off auto-submit. Say: "Explain what the main script in this folder does. Don't change anything."
Compare the transcript with what you said, then send it.
You're done when five spoken requests in a row arrive as the words you meant. Only then try requests that change files, and keep reviewing the diff as usual.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Based on the opencode-voice repository README and releases, its crates.io entry and OpenCode's server and v1-to-v2 migration docs, checked 23 September 2026. We did not install or run it for this review. The catalogue record also mentions a separate text-to-speech plugin (Olbrasoft/opencode-voice-plugin, last updated December 2025); this review covers the voice-input CLI.
Drive opencode hands-free by speaking commands into a local voice pipeline.
Whisper transcribes speech on your CPU and forwards it to opencode.
Add community text-to-speech so the agent can answer aloud.
Community / local-first stack — the opencode-voice Rust CLI (independent), the Olbrasoft/opencode-voice-plugin, and any local TTS; demoed end-to-end by the YouTube channel "No place like localhost". · free · self-host
voice coding
speech recognition
local text-to-speech
hands-free terminal
privacy-first
cpu whisper
plugin integration
offline processing
talk-think
write-code
automate
View sourced imageOpenCode Voice 0.1.4 transcribed synthetic speech locally with Whisper tiny.en and forwarded its partial transcription into OpenCode's native terminal prompt (right). The input remains unsubmitted; no language-model response was requested. This capture demonstrates the integration, not transcription accuracy.
Isolated Ubuntu capture of mathew-cf/opencode-voice with native OpenCode 1.18.25
Owner-requested local interface capture using exact official binaries and a public synthetic phrase generated by the installed Windows speech capability. No microphone, private session or provider credentials. Original full viewport retained without overlays or transcript editing; upstream product names retained for attribution.
