Most developers I know have the same problem: their best thinking happens away from the desk, but actually doing anything with that thinking requires sitting down, opening a terminal, and typing. The commute, the walk, the gym session — all of it is dead time. DashVox is a voice-first app that tries to fix that. It lets you start and steer AI coding agents like Claude Code and Codex over SSH, by voice, from your phone, your Apple Watch, or your car’s CarPlay display.
The concept sounds almost too niche at first. But then you think about how many senior engineers sit in Bay Area traffic for two hours a day, or how many developers deal with repetitive strain injuries that make heavy keyboard use painful. Suddenly a tool that keeps your agents running while your hands stay on the wheel sounds less like a novelty and more like something genuinely useful.
It’s also completely free. Both the self-hosted and cloud versions cost nothing. That alone makes it worth a closer look.
Key Features of DashVox
Voice-First AI Chat Across Every Screen
You can talk to Claude, GPT, or Gemini through the phone app, or go fully hands-free via CarPlay, iOS, or Wear OS. An Android version supporting equivalent devices is now in beta, and they are now exploring integrations with smartglasses such as Meta Raybans. Your voice gets transcribed in real time, the request goes to your chosen LLM, and the response is read back through your earbuds. You can even switch models mid-conversation without touching anything. Responses are formatted for spoken delivery, so you’re not hearing a wall of markdown read aloud.
Remote SSH Session Management
This is the part that sets DashVox apart from every other voice AI app. It uses JSch with tmux session management to SSH into your actual development machine. From your watch or car, you can start, stop, monitor, and approve coding sessions running on your own hardware. Built-in profiles exist for Claude Code and Codex, and you can create custom profiles for any CLI tool you use regularly.
Agent Mode: One Command, Multiple Sessions
Tell DashVox to spin up three sessions, frontend, backend, and tests, and the meta-agent handles the entire session lifecycle. It starts each one, monitors progress, routes approval prompts to you, and sends a push notification when something needs attention or finishes. You stay informed without micromanaging anything. It’s the kind of orchestration that usually requires a second monitor and constant tab-switching.
Smart Connectivity and Security
Dead zones happen, especially on commutes. DashVox uses exponential backoff and idempotent message delivery so you don’t get duplicate commands or dropped sessions when your connection blips. On the security side, SSH keys are encrypted at rest. The self-hosted version requires no account at all: no Firebase, no data leaving your network. It supports connection through a device-to-computer VPN ensuring no external exposure to your workflow. The cloud version adds Firebase Auth with Google or Apple sign-in, but the security-first architecture carries through either way.
Marcus at 7:45 AM, Stuck on the 101
Marcus is a backend engineer in San Francisco. He commutes to Redwood City three days a week, roughly 50 minutes each way on a good day, closer to 80 when traffic stacks up near San Mateo. He started using DashVox two weeks ago after a bout of wrist tendinitis made even light keyboard use uncomfortable in the mornings.
At 7:45 AM on a Tuesday, sitting in the usual crawl past Millbrae, he clicked ‘speak’ on DashVox CarPlay with his steering wheel’s D-pad and said: “Start two sessions, one for the auth service refactor and one to run the integration test suite.” DashVox opened SSH connections to his home server, spun up both Claude Code sessions inside tmux, and confirmed each one was running. He heard the confirmation through his AirPods while keeping both hands on the wheel.
Twenty minutes later, a push notification landed on his watch, and his music briefly paused to speak a status update aloud through CarPlay: the integration tests had finished with two failures. He pressed speak on DashVox CarPlay, and responded: “Tell me what failed,” all while keeping his attention on the road. By the time he pulled into the parking garage at 8:55 AM, one session had already applied a fix and was waiting for his approval. He said “approve” out loud. Done.
That morning, Marcus arrived at his desk with 70 minutes of agentic coding already behind him. His wrists hadn’t touched a keyboard.
How DashVox Actually Works
Step 1: Get the App and Backend Running
Download the iOS app from the App Store. For self-hosting, you grab the server JAR (currently v1.0.101), run it with Java 21, and a small control window shows you the LAN address to point the app at. The whole setup takes as little as one command. Download and run. No complex infrastructure, no cloud account required unless you want one.
Step 2: Configure Your SSH Profiles
Add your development machine as an SSH target inside the app. If self-hosting on your development machine, that would be localhost. Set up agent profiles for Claude Code, Codex, or any other CLI tool you use, and an LLM provider api for the agent itself to use. DashVox stores SSH keys encrypted at rest and manages tmux sessions on your behalf.
Step 3: Speak from Wherever You Are
Open the app on your phone, flip to CarPlay on the car display, or raise your wrist. Speak your request. DashVox transcribes it, routes it to your chosen LLM, executes any session commands over SSH, and reads the response back to you. Approvals, status checks, and new session requests all work the same way regardless of which surface you’re on.
Pricing: Both Plans Are Free
DashVox offers two tiers, and both are $0:
- Self-Hosted ($0 forever): Run the server on your own hardware, no account needed, no data collection, unlimited conversations and SSH sessions.
- Cloud ($0 to start): Hosted at api.dashvox.ai, sign in with Google or Apple, cloud sync across devices, push notifications, automatic updates.
The project is built by a solo developer (though he recently brought on a former engineering coworker as a co-founder), so if DashVox saves you real time, the ask is simple: use it, share it, and send feedback.
Worth Trying If You Code Away from Your Desk
DashVox is a focused tool for a specific type of developer: someone already using agentic coding workflows who wants to keep that momentum going outside of desk hours. If you commute, deal with repetitive strain, or just want your agents running while you do other things, this is the most practical implementation of voice-controlled development I’ve seen. Free, self-hostable, and available right now on iOS.
