ROMA runs multiple coding agents simultaneously – Claude, Codex, etc.
Git worktree isolation lets agents vote on code before merging anything.

Audio-first agent workspace trying to unify Claude, Gemini, Grok, and Cursor in one GUI.
Developers wanting a unified interface for multiple AI coding assistants
Cursor · Continue · Windsurf
It’s now available for download. I’m updating the demo video and docs next.
Conceptually, it’s a unified GUI (+ optional TUI) for working with multiple coding/agent tools in one place — similar space as Claude Code, Codex CLI, Gemini CLI, Cursor CLI, etc., but designed to mix providers and workflows instead of locking into one.
Some things it does:
Natural, two-way conversational interface at the core (text + optional private audio capture/transcription) Multi-provider support — switch models mid-session or combine them Native Telegram integration (text, audio, commands) Built-in browser automation (headless Chromium), plus bundled tools (bun, ffmpeg) for scripting and session recording MCP + Skills system for extending capabilities Drag-and-drop files with preprocessing before LLM use Context management + caching (auto + manual control) Environment variable management with selective (“surgical”) injection into authorized commands only Local-first / privacy-focused (except for chosen LLM + web queries)
Still early, but usable today. Curious how it compares for people already using Claude Code / Cursor / etc.
Happy to answer questions.
Git worktree isolation lets agents vote on code before merging anything.
Turns agent terminal output into ambient radio—Global Mix monitors all sessions at once.
Actual Kubernetes operator for agent lifecycle, but orchestrating agents is still a niche use case.
Unified CLI for coding agents, but shell aliases or a simple script could do the same.
tmux-based agent topology you can save and restore like a Docker compose file.
Cross-agent workstreams with exact transcript binding beat native /resume.