DashVox

Voice control for coding sessions on another machine

DashVox is a voice-first interface for working with AI coding agents remotely. It connects over SSH to the machine running an agent such as Claude Code or Codex, lets you give instructions, and reads back summarized output. Its purpose is to keep a development session accessible when you step away from the workstation, rather than move the whole project into a new mobile editor.

Developers can start, stop and switch sessions, ask about progress, and respond to approval prompts. Spoken responses make short status checks practical without reading a terminal transcript. The iOS application includes CarPlay support; the website currently labels Android as coming soon. Treat platform announcements separately from supported clients you can install.

Sessions, speech and hosting choices

DashVox combines speech input, text-to-speech replies and remote session management. Agent Mode is designed to coordinate multiple coding sessions, such as separate frontend and test tasks. These controls are useful for directing work, but hearing a summary does not provide the same detail as inspecting the complete diff and command output.

The self-hosting guide describes a Java 21 backend for Windows, macOS or Linux. You run the server, point the phone application at its address, and configure access to the development machine. Remote use requires network connectivity to that backend. The guide also describes optional API-token authentication, which is off by default, so remote access needs deliberate configuration and compatible client support.

A concrete development workflow

A useful starting task is investigating a failing test while away from the desk. Connect to a development environment, select its coding session, and ask the agent to summarize the failure and suggest a small fix. Request a test run, listen to the result, then return to a full display to inspect the patch before merging it.

This fits developers who already understand their repository and can describe bounded work clearly. Separate investigation from production changes: use appropriate SSH permissions, name the intended environment, and require review for commands that modify live services. Those are operating practices, rather than guarantees supplied by a voice interface.

Data handling and important limitations

The privacy policy distinguishes local hosting from the managed service. Self-hosting keeps DashVox conversation and session storage on your hardware, but requests to external model providers still go to those providers. Local installation therefore does not automatically make every AI interaction offline.

For cloud mode, the policy says SSH keys and model-provider API keys are stored server-side and are not yet encrypted at rest. Review that limitation before connecting sensitive environments. Voice recognition, summaries and agent-generated edits also need human checking. CarPlay support does not establish that complex code review is appropriate while driving.

Pricing and getting started

The pricing page offers a free self-hosted backend. Cloud Ultimate uses a subscription plus prepaid usage credits for managed models and compute, with current amounts shown in the application. Enterprise access is arranged through sales. Check the hosting mode, provider costs, supported client and credential requirements together before choosing a setup.