One company, three products

Say it once, from request to handoff

CubeLV, ParallelSandbox and JarvisVoice are all built by Runvo LLC. Each covers one part of the work: you hand it off by talking, AI follows the process every day, and code goes to several AI agents working in parallel, tested before it reaches you. Each works on its own; together they make one workflow from idea to release.

Three products, one part each

Ask: by talking

JarvisVoice

A macOS menu bar app. It listens while it is on and writes what you say, cleaned up, into the app you are using: Claude Code, Codex, ChatGPT or CubeLV. Say a command and it operates the app for you. Open source and free; transcription can stay entirely on your Mac.

Go to jarvis-voice.com

Every day: AI follows the process

CubeLV

Build your own automation app from one sentence. Agents do each step on schedule and save the results for the next step; where you need to decide, a decision card stops and waits for you.

See how CubeLV works

Build: in parallel, tested before handoff

ParallelSandbox

Lets Claude Code, Codex and other coding agents work in parallel: each task runs in its own box, a real Linux machine, where the agent finishes and tests it first. Finished versions line up for you to try, and each opens as the running product, not a diff.

Go to parallelsandbox.com

How one piece of work gets done

Take a common request: "When a user reports a bug, open a ticket, fix it and run the tests. Ask me before it goes live."

  1. JarvisVoice

    Say it

    You say it to your Mac. JarvisVoice drops the filler words, spells your terms the way you do and writes it into CubeLV. No typing.

  2. CubeLV

    It becomes a daily process

    CubeLV draws the blueprint: an agent that collects reports, a ticket plugin, an agent that fixes bugs, and the "before release" decision card. You review the drafts and press start before anything is built.

  3. ParallelSandbox

    Fixed in parallel

    Each report gets its own ParallelSandbox box for the coding agent that fixes bugs: three reports, three boxes working in parallel. Each runs your project, reproduces the problem, fixes it, runs end-to-end tests in a real browser and fixes what breaks.

  4. ParallelSandbox

    Handed to you

    The fixed version is handed off and opens as the running product, so you click through it yourself. Steps that need you, like signing in or paying, it hands the screen over to you.

  5. CubeLV

    Live only when you say so

    The "ask me before release" card waits for you in CubeLV. Press it and the next agent ships; leave it and nothing moves.

  6. JarvisVoice

    Say what to change

    If something is off, point at the screen and say it. The feedback goes back to the same agent, which keeps working and hands off again.

Why connect them

  • No typing to hand work off: talking is faster, and JarvisVoice works with any AI tool.
  • The process does not depend on you remembering it: CubeLV runs on schedule and only stops where you decide.
  • Several things at once: ParallelSandbox gives each task its own box so work runs in parallel instead of queuing. Finished versions line up for you to try, as the running product rather than a diff.
  • Built by one team: voice input in the CubeLV and ParallelSandbox apps uses the same voice core as JarvisVoice, and this is how we build CubeLV and ParallelSandbox every day.

Where to start

CubeLV also has an MCP server: Claude Code and Codex can connect to CubeLV and ParallelSandbox at the same time, reading and writing your CubeLV and running code in a box in the same conversation.