01
Write
Claude Code writes the change.
It reads the code, makes the edit and runs the tests in its own Git worktree.
claudewriter
> add rate limiting to /api
● Edit middleware/limit.ts +18 −2
● Bash npm test ✓ 47
Open source · MIT · Beta
Your AI agents, working as a crew
Vulgr puts your coding agents in one terminal and gives each a role. They hand off work, check each other, and loop until it's right — you just lead the crew.
Works with
Runs the agents you already use
Vulgr is not a new agent. It drives Claude Code, Codex, Cursor, Gemini CLI, OpenCode and AGY — the same CLIs and sessions you use every day, wired together into one crew.
The crew
Three agents. Three roles.
Vulgr gives every agent a job. One writes, one reviews, one tests — and they pass the work along.
↺ loop until approved
02
Review
Codex reviews it.
A second agent reads the diff with fresh eyes. If something's off, the work goes straight back.
codexreviewer
> review claude's changes
1. 429 lacks Retry-After
→ handed back to claude
03
Test
AGY tests it end to end.
The tester runs the real flow. Nothing reaches you until it passes.
agytester
> run e2e on /api
✓ Retry-After header set
✓ 48 passed (3.2s)
claude writer
> add rate limiting to /api, 100 req/min per key ● Read src/routes/api.ts ● Read src/lib/redis.ts ● Edit src/middleware/limit.ts +18 −2 ● Edit src/routes/api.ts +4 −1 ● Bash npm test ✓ 47 passed (2.1s) Done — limiter uses a sliding window in Redis.
codex reviewer
> review claude's changes • reading middleware/limit.ts • reading limit.test.ts 1. 429 response lacks Retry-After 2. window resets on Redis restart — acceptable for now → handed back to claude next: agy runs e2e…
Agent chat · new
Talk to one agent — or the whole team
Chat with Claude Code, Cursor, Codex or Gemini/AGY. Pick the model and effort, watch file changes live — or type two @mentions and the agents start working it out together.
SoloTeam Claude #1 opus · high Claude #2 sonnet Cursor gpt-5 Fast
@claude @claude @cursor add rate limiting to /api — 100 req/min per key
Team mode · 3 agents · splitting the work
Claude #1
I'll take the middleware. Plan: fixed window in Redis, reset every 60s.
Claude #2
Pushback — a fixed window lets a client burst 200 requests across the boundary. Sliding window instead?
Claude #1
Fair point, switching to a sliding window. Cursor, can you cover the 429 path?
Cursor
On it — writing limit.test.ts.
⠋ Claude #2 is reading limit.test.ts…
Cursor wants to run npm test -- limit
AllowAlways allow in this chatDeny
@Message the team…↵
Solo mode
Chat with Claude Code, Cursor, Codex or Gemini/AGY. Choose the model and effort level and watch file changes live in the side panel.
Team mode
Several agents genuinely debate in the same chat: they split the work, review each other's code and change their minds when they should. Two @mentions — like @claude @cursor — start it.
A model per agent
Give every agent in the team its own model.
Claude #1, Claude #2
Add the same agent more than once. Each copy thinks on its own and questions the other.
Fast mode
Runs team rounds on fast models so you aren't left waiting.
Command approval
When an agent wants to run a command, you decide: allow, always allow in this chat, or deny. Or set the default to ask, auto-allow or always deny.
History: past chats sit in the sidebar, grouped by day and searchable. Open an old one and Claude picks up where it left off.
Crew
Agents that work together
In Vulgr, agents don't just run next to each other. Each has a role, work moves from one to the next, and a result can be checked by another agent before it reaches you.
Duo Loop01
Two real terminals side by side. The checker writes tests from your goal, the writer codes until they pass, then the checker reviews the diff and approves or sends feedback.
Agent Swarm02
Three roles in the background, optionally in an isolated Git worktree: a verifier writes tests, a builder codes until they pass, an auditor reviews the diff.
Orchestrator03
A planner splits a big goal into a task graph, workers run each task in parallel in their own worktree, and a reviewer checks the merged result.
Tip: pair different agents — Claude Code writing, Codex or AGY checking — so each catches what the other misses. Works with Claude Code, Codex, AGY / Gemini CLI, OpenCode and Cursor.
Workspace
A terminal first, then a crew
Everything you expect from your terminal, plus the pieces that keep a crew organized.
Real terminals
Every pane is a real terminal on xterm.js and node-pty. Your shell, aliases and tools work as usual, and agents run in the same environment you do.
Sessions & groups
Keep each piece of work separate. Name sessions, group them by project, and pick up where you left off.
Command box
Start with # to describe a command in plain words and get the shell command back. Start with ? to ask Claude without leaving the terminal.
Changes, report, memory
See which files the crew touched, get a summary of the session, and keep context the crew carries across sessions.
Download
Get Vulgr
Download the build for your system — the one that matches your computer is highlighted.
For you macOS Apple Silicon .dmg For you macOS Intel .dmg For you Windows x64 .exe For you Linux x64 · AppImage .AppImage
Or install with npm
$
npm install -g vulgr@beta Latest: checking latest release…
Unsigned beta — how to open
Beta builds aren’t code-signed or notarized yet, so your system warns you the first time you open Vulgr. That’s expected — here’s how to run it.
macOS
Move Vulgr to Applications and open it once. When macOS blocks it, go to System Settings → Privacy & Security and click Open Anyway. Or clear the quarantine flag:
$
xattr -dr com.apple.quarantine /Applications/Vulgr.app Windows
Run the installer. When SmartScreen says “Windows protected your PC”, click More info, then Run anyway.
Linux
Make the AppImage executable, then run it:
$
chmod +x Vulgr-*.AppImage && ./Vulgr-*.AppImage