RVGE
Early access

RVGEBRAIN

Formerly RVGE CODE · Flagship

One workbench for every coding agent you already use.

A Windows development environment built around AI coding agents. Run Claude Code, Codex and other CLI agents side by side, see at a glance which one needs you, review their edits in a real editor, and keep every session alive across restarts.

Request early access Watch the 62 s tour
Status
Private alpha
Version
0.3.0
Platform
Windows 10 / 11 · x64
RVGEBRAIN with five agent panes: three Claude Code chat views, a Gemini CLI terminal and a Codex CLI chat, with the Agent Activity bar and its timeline on the right.
Real capture · release build 0.3.0 · demo project, scripted agents Five agents on one objective, each with its own status.
01 Why it exists The problem

Coding agents got good. Working with three at once did not.

Before

A wall of terminal windows, and no idea which agent stopped to ask you something ten minutes ago.

With RVGEBRAIN

Every agent has one status: working, needs you, ready or idle. A notification arrives the moment one waits.

Before

You type a follow-up while the agent is still busy, and it lands in the middle of a dialog.

With RVGEBRAIN

The composer queues your message and types it only when that agent's own prompt is on screen.

Before

Edits scroll past in a terminal, and a restart takes every session with it.

With RVGEBRAIN

Diffs open in a real editor, permission requests come as cards, and sessions keep running and replay when you return.

02 Tour 62 seconds · with sound

See it run.

Recorded from the 0.3.0 release build. The agents in the recording are scripted stand-ins, so the demo makes no model calls and every run looks the same. Versions shown inside the demo are part of that script.

03 How it works Five steps

From an empty folder to a verified change.

  1. 01

    Open a folder

    RVGEBRAIN checks which agent CLIs are installed and signed in. It never reads their credentials, only whether they exist. Untrusted folders open in Restricted Mode.

  2. 02

    Launch the agents you want

    Pick a preset, Solo, Pair, Trio or Workbench, choose an engine and a role for each pane, review, and launch them side by side.

  3. 03

    Give them work

    Type into each agent’s composer, or start a Swarm with one shared objective. Messages wait in a queue until that agent’s prompt is ready for them.

  4. 04

    Answer only when asked

    The Agent Activity bar tells you which agent needs your answer. Permission requests and plans come as cards you approve or deny.

  5. 05

    Check the result

    Read the diffs in the editor, run the terminal, commit from source control. With ULTRACODE on, your own test command decides when a task is done.

RVGEBRAIN welcome page listing detected agents and the launch presets Solo, Pair, Trio, Workbench and Swarm.
Real capture Detected agents and launch presets on the welcome page.
Claude Code chat view in RVGEBRAIN showing a to-do list, a side-by-side diff and a "Make this edit?" card with Allow and Deny buttons.
Real capture A permission request, answered from a card instead of a scrolling terminal.
04 Supported agents Bring your own

Seven agent CLIs. Two with a full chat view.

RVGEBRAIN detects which agents are installed and signed in, then launches them under your own subscriptions. Claude Code and Codex also run in a structured chat view; the rest run in the integrated terminal.

Agent Terminal view Chat view Tested with the real CLI Notes
Claude CodeAnthropic Yes Yes Yes Permission and plan approval cards, sub-agent steps, pasted images, execution modes, effort, model, resume and fork of a session.
Codex CLIOpenAI Yes Yes Yes One structured turn per message on a resumable thread. Model, effort and sandbox apply from the next message.
Gemini CLIGoogle Yes No Not yet Runs in the terminal view with a model choice. RVGEBRAIN asks before sending, because it cannot read Gemini’s prompt screen yet.
GitHub Copilot CLIGitHub Yes No Not yet Detected and launched in the terminal view.
Cursor AgentCursor Yes No Not yet Detected and launched in the terminal view.
OpenCodeOpen source Yes No Not yet Detected and launched in the terminal view, with a first message.
AiderOpen source Yes No Not yet Launched in the terminal view. It cannot take a first message or brief on start.

Product names belong to their owners. RVGEBRAIN is an independent application and is not affiliated with or endorsed by any of them. You need your own installation and account for each agent.

05 Multi-agent Teams · Swarm · ULTRACODE

Run a team, not a tab collection.

What you actually see and do when several agents work at once, and where the current limits are.

01

Panes, not windows

Agents, terminals, editors, diffs and a localhost browser preview are panes. Split them, stack them as tabs, float them, or tidy them into a grid.

02

Team roster

Save named teammates, each with an engine, a role, a standing brief, a start folder and a model. Every launch is logged with its duration and exit code.

03

Swarm

Several agents receive the same objective and are told who else is working on it. They share one folder today, so they work best on separate parts of a task.

04

Agent Activity

One status per agent, working, needs you, ready or idle, plus a timeline of what happened and a notification when someone waits for you.

05

Composer

Send now, queue or interrupt with Esc. Mention files, attach the git diff or terminal output, and switch execution mode, effort or model per agent.

06

ULTRACODE

Wrap a task in a fixed procedure. When the agent finishes, your verification command runs, and failures go back as corrections for a set number of rounds.

The RVGEBRAIN composer with the execution mode menu open: Default, Manual, Edit automatically, Plan and Auto.
Real capture Execution mode, reasoning effort and model, per agent.
06 Inside Verified in the source

A real development environment underneath.

Agents in real terminals

Every agent runs in a Windows ConPTY session owned by the backend. Sessions keep running in the background, replay their output when you come back, and stop as a whole process tree.

Structured chat for Claude Code and Codex

Plans, to-do lists, diffs and tool calls render as cards. Permission requests arrive as Allow, Always allow or Deny, and plan approvals get their own card.

One status per agent

The Agent Activity bar shows who is working, who needs your answer and who is ready, with a timeline and a notification the moment an agent waits for you.

A composer that waits for the prompt

Send now, queue or interrupt. The queue survives restarts, and text is typed into an agent only when its own prompt is on screen.

Teams and Swarm

Save named teammates with an engine, a role and a standing brief. Launch presets from Solo to Workbench, or give several agents one shared objective.

ULTRACODE verification

When an agent says it is done, RVGEBRAIN runs your own verification command and sends failures back as corrections, round after round, until it passes or you stop it.

A real editor

Monaco panes with autosave, diff view and external-change detection, plus language servers for TypeScript, Rust, Python, Go and C/C++ when they are installed.

Workspace jail

Every file operation is checked against the open folder, junction escapes included. Untrusted folders open in Restricted Mode, without language servers.

RVGEBRAIN with the Claude Code chat next to a Monaco editor tab and a PowerShell terminal.
Real capture Chat, editor and terminal in one layout.
A finished Claude Code turn in RVGEBRAIN with the applied diff and a summary of the change.
Real capture A finished turn: the diff that was applied and what changed.
07 Engineering and status Where it stands

Tested, measured, and honest about the gaps.

201 passed Vitest 37 files · project test log, 17.10.2026
191 passed Rust 7 ignored · project test log, 17.10.2026
96 passed End to end Playwright against the release build · 17.10.2026
1.28 s Startup release build, measured on the development PC

Test counts come from the project's own test log, including work not yet committed on that date. Timing was measured on the development PC.

Built with
  • Tauri 2
  • Rust
  • React 19
  • TypeScript
  • Monaco
  • xterm.js
  • ConPTY
  • SQLite

Works today

  • Workbench with editor, terminals, source control, search and language servers
  • Claude Code and Codex in the structured chat view
  • Seven agent CLIs in the terminal view
  • Teams, Swarm, composer queue and ULTRACODE
  • Themes, compact density, command palette and full keyboard access
  • Open VSX themes, icon themes, snippets and grammars

Missing before a public release

  • Code signing. Windows SmartScreen warns about an unknown publisher.
  • Auto-update, licence and trial.
  • Git branches, push and pull. Status, staging, commit, amend, log and diff work today.
  • A separate git worktree per agent. Agents in a Swarm share one folder today.
  • A debugger.
  • Clean-machine QA and a smoke suite against the real agent CLIs.
  • macOS and Linux builds.

Next on the roadmap · no dates

  • A git worktree per agent, so agents in a Swarm never share files
  • A task graph that hands work from one agent to the next
  • Git branches, push and pull, and a debugger
  • Code signing, auto-update, licence and a clean-machine test

RVGEBRAIN does not sandbox the agents. They run with your user rights, exactly as they would in your own terminal.

You bring the agents and your own subscriptions. RVGEBRAIN stores no API keys and contains no telemetry code.

Windows 10 and 11, x64 only.

08 Who it is for

Windows developers who already use Claude Code, Codex or other CLI agents and run more than one at a time.

If you already pay for Claude Code or Codex and find yourself managing agents instead of code, RVGEBRAIN is built for exactly that afternoon. If you use a single agent in an editor you love, you may not need it yet.

09 Recent changes From git
  1. Release

    RVGEBRAIN · v0.3.0

    Codex joins the chat view

  2. Release

    RVGEBRAIN · v0.2.0

    Plan approval card

  3. Milestone

    RVGEBRAIN

    RVGE CODE becomes RVGEBRAIN

  4. Feature

    RVGEBRAIN

    Teams, Swarm, the composer and ULTRACODE

All updates
10 Questions

Asked before you ask.

Can I download RVGEBRAIN today?

Not publicly. It is in private alpha. Ask for early access through the contact page and describe how you work with agents today.

Do I need an API key?

No. RVGEBRAIN launches the agent CLIs you already have installed and signed in to, under your own subscriptions. It stores no API keys and sends nothing to RVGE.

Is it a fork of VS Code?

No. It is its own application built on Tauri, Rust and React, with the Monaco editor at its core. It reads the look of VS Code where that helps, and it does not run VS Code extensions. Themes, snippets and grammars from Open VSX do work.

Are the agents sandboxed?

No. RVGEBRAIN controls what it does itself, every file operation is jailed to the open folder, but the agents run with your user rights, exactly as they do in your own terminal. Use their own permission modes.

What happened to RVGE CODE?

RVGEBRAIN is the same product line. RVGE CODE was rebuilt from scratch in October 2026 and renamed when the agent workbench became the centre of it.

Will there be a Mac or Linux version?

Not yet. RVGEBRAIN relies on Windows ConPTY and Job Objects today. Other platforms come after a stable Windows release.

What will it cost?

No price has been set, and nothing is for sale before the company is registered. When a price exists it is published here with its anchor price.

Run your agents
side by side.

RVGEBRAIN is in private alpha on Windows. Tell us which agents you use and how many at once, and we will be in touch when access opens.

Request early access Follow the updates