# Vulgr

> Vulgr is a free, open-source (MIT) desktop terminal for running AI coding agents — Claude Code, Codex, Gemini CLI, Cursor, OpenCode and Antigravity — side by side. It gives each agent a role so one writes code, another reviews it and a third tests it, looping until the work passes. It runs locally on macOS, Windows and Linux and uses the agent CLIs you already have.

Vulgr is not a new AI model or agent. It is a desktop app (Electron, React, TypeScript) that orchestrates the agent CLIs you already use and keeps everything local. Website: https://vulgr.tech/ · Source: https://github.com/vulgrs/vulgr · License: MIT.

## Key facts

- Platforms: macOS (Apple Silicon and Intel), Windows, Linux. Install: DMG, EXE or AppImage from https://github.com/vulgrs/vulgr/releases, or `npm install -g vulgr@beta`.
- Agents: Claude Code, Codex, Antigravity (AGY), Gemini CLI, Cursor, OpenCode, and any OpenAI-compatible endpoint.
- Multi-agent modes: Duo Loop (writer + checker in two real terminals, looping until approved), Agent Swarm (verifier, builder and auditor in the background, optionally in a Git worktree), Orchestrator (a planner splits a goal into a task graph, workers run tasks in parallel worktrees, a reviewer approves the merge).
- Team chat: mention several agents in one message; a Discussion → Verification (local tests) → Synthesis pipeline runs in the background.
- Workspace: real xterm.js + node-pty terminals, split panes, sessions and groups, a command box (`#` turns plain language into a shell command, `?` asks Claude).
- Project tools: diff viewer, AI commit messages, commit & push, branch menu, pull requests, project memory and shared skills, package manager, dependency optimizer, context optimizer, session reports.
- Privacy: runs locally; Vulgr goes online only for update checks, GitHub sign-in and the agent CLIs themselves.
- Vulgr Agent (CLI preview): an agentic shell inside your existing terminal with a model router, ten tools, fail-closed permissions and a headless mode (`vulgr -p`).

## Docs

- [Welcome to Vulgr](https://vulgr.tech/docs/): Run your AI coding agents as a crew, in one terminal.
- [Installation](https://vulgr.tech/docs/installation/): Install Vulgr on macOS, Windows or Linux and get your agents ready.
- [First steps](https://vulgr.tech/docs/first-steps/): Open your first session, split the pane and give the crew a task.
- [Interface](https://vulgr.tech/docs/interface/): The top bar, sidebar, session tree and profile card.
- [Command box](https://vulgr.tech/docs/command-box/): Shell commands, commands from plain words, and questions for Claude.
- [Keyboard shortcuts](https://vulgr.tech/docs/shortcuts/): Every default shortcut; customize them in Settings → Shortcuts.
- [Project tools](https://vulgr.tech/docs/project-tools/): Git, memory, packages, dependency and context optimizers, custom endpoints, System One and reports.
- [Duo Loop](https://vulgr.tech/docs/duo-loop/): A live two-agent loop — one writes the code, the other tests and reviews it.
- [Agent Swarm](https://vulgr.tech/docs/agent-swarm/): An autonomous network of three agents — Builder, Verifier and Auditor.
- [Orchestrator](https://vulgr.tech/docs/orchestrator/): A planner and merger that splits big goals into parallel tasks.
- [Agent chat and System One](https://vulgr.tech/docs/claude-chat/): Chat with Claude Code and other agents, with live changes, mentions, usage limits — and a strict JSON compiler mode.
- [Team chat](https://vulgr.tech/docs/team-chat/): Mention several agents in one message — they discuss, verify with your tests and hand you one synthesis.
- [Vulgr Agent](https://vulgr.tech/docs/vulgr-agent/): An agentic shell that runs inside your existing terminal — one input line for shell commands and plain-language tasks.
- [Agent architecture](https://vulgr.tech/docs/agent-architecture/): The engine loop and the modules under src/ — UI-independent, event-driven and cache-friendly.
- [Model router](https://vulgr.tech/docs/agent-router/): How Vulgr Agent picks a model and effort for each task, escalates when stuck and classifies cheaply when unsure.
- [Tools and permissions](https://vulgr.tech/docs/agent-tools/): The ten agent tools, the agent's own PTY sessions, and the fail-closed permission policy.
- [Headless mode](https://vulgr.tech/docs/agent-headless/): Run Vulgr Agent without a TTY — for scripts, CI and benchmarks — with a JSONL event stream and exit codes.
- [CLI reference](https://vulgr.tech/docs/cli/): The vulgr and orchestrate commands.
- [Configuration and security](https://vulgr.tech/docs/configuration/): Where and how settings, keys and project memory are stored.
- [Architecture](https://vulgr.tech/docs/architecture/): How the Electron main process, the core engine and the React UI fit together.
- [Privacy and network](https://vulgr.tech/docs/privacy/): What runs locally, and the only reasons Vulgr goes online.

## Guides

- [How to use Claude Code and Codex together](https://vulgr.tech/guides/claude-code-and-codex-together/): Run Claude Code and OpenAI Codex side by side so one writes the code and the other reviews it — a practical cross-review workflow with tests in the loop.
- [Duo Loop vs Agent Swarm vs Orchestrator: which multi-agent mode to use](https://vulgr.tech/guides/multi-agent-coding-modes/): Three ways to make AI coding agents check each other's work — when to pick a writer–checker pair, a three-role swarm, or a planner with parallel workers.
- [Run AI coding agents in parallel with Git worktrees](https://vulgr.tech/guides/parallel-ai-agents-git-worktrees/): Let several AI agents work on the same repository at once without stepping on each other — how Git worktrees isolate each agent, and how to merge the results safely.
- [What is an AI coding agent terminal — and do you need one?](https://vulgr.tech/guides/ai-coding-agent-terminal/): Why developers who use Claude Code, Codex, Gemini CLI or Cursor move to a terminal built for agents, and the features that matter when you run more than one.

## Optional

- [Full text of the docs and guides](https://vulgr.tech/llms-full.txt)
- [Türkçe site](https://vulgr.tech/tr/)