Invite-only beta · macOS · iPad & Slack

Autonomous AI engineering teams.
On your Mac.

Stop juggling Claude windows. Spin up a team of agents instead.

They plan, build, and verify together, routing work by role through a shared board, handing off to each other, proving features actually work, and pinging you only when they're truly blocked.

macOS 13+ · Apple Silicon & Intel · Requires Claude subscription

Parallel agents made you the bottleneck.

Ten Claude windows don't make a team, they make ten things to babysit. You become the router, the reviewer, and the memory. Claudeverse moves that coordination into the software.

Window-juggling
waiting for input…
approve edit? (y/n)
which one had the bug?
idle, you forgot this one
  • · You hand-route every task between agents
  • · You review everything, even what's obviously fine
  • · "Done" means tests passed, not that it works
A coordinated team
Backendbuilding
Frontendbuilding
Verifierverifying
Mobileverified ✓
  • The coordinator routes work by role, on its own
  • A Verifier drives the real app and signs off
  • You only see what genuinely needs a human
Claudeverse, a Tag Team of Claude agents working in one window

One window. The whole team, their status, and what they need from you, at a glance.

Tag Teams

A team that runs the loop without you.

Give a Tag Team a spec. It breaks the work down, assigns it by role, hands off through a shared task board, and verifies the result, the same loop a human engineering team runs, minus the standups.

1

Decompose

Hand the team a PRD or spec. A Project Manager agent breaks it into labeled tasks on a shared board, with dependencies ordered API-first.

2

Route by role

A coordinator dispatches each task to the agent that owns that role (Backend, Frontend, Mobile) only when its blockers are done. No hand-routing.

3

Hand off

Agents pass work to each other through the board with explicit contracts, so the front end builds against the API the back end actually shipped.

4

Verify & close

A Verifier exercises the real feature and emits a verdict. Pass → done. Fail → a bug routed back to the owner. The loop continues on its own.

Staff the team with the roles you need

Backend Frontend Mobile Verifier Project Manager Architect QA

Every agent has a stable identity and its own role. Pick a model per role, Opus where it matters, a lighter model where it doesn't.

The Verifier

Green tests aren't proof.
The Verifier drives the real app.

Passing unit tests while the feature is broken is exactly the failure a Verifier exists to catch. It opens the actual UI, a real browser for the web and a real iOS simulator for mobile, proves each acceptance criterion with a deterministic assertion, and emits one honest verdict.

verified failed → files a bug blocked → asks a human
  • A failed verdict files a bug routed straight back to the role that owns the code.
  • Each team gets its own dedicated iOS simulator, two teams never collide.
  • "Couldn't test it" is reported as blocked, never a false failure.
verify-mobile · iOS 26 simulator
AC-1: empty feed shows pull-to-refresh
AC-2: follow triggers feed reload
AC-3: new_follower notification renders
verdict verified

3/3 acceptance criteria proven on a live build. Screenshots attached. Signed off, no human needed.

Your Tasks 2
Provide RecipePal test login Explain

Requested by Verifier · secure field, stored 0600 on your Mac

Approve API contract change Explain

Requested by Backend · decision unblocks 3 tasks

Human-in-the-loop

It only pings you when it's truly stuck.

When an agent hits something only a human can resolve (a login, a key, a decision), it doesn't stall silently. It surfaces a card in Your Tasks, then auto-resumes the moment you answer. Everything else, it handles itself.

  • Secure credentials. Secrets go into a masked field, are stored 0600 on your Mac, and injected as env vars at launch, never to the board, never to disk in plaintext.
  • Plain-language Explain. Not sure what a task is asking? One click and Claude walks you through it, step by step, like you're non-technical.
  • Auto-resume. Answer a card and the blocked work picks back up on the next poll, no babysitting.
Slack relay

Run your team from Slack.

Connect a workspace and your Tag Team comes to you. Agents post progress, ask questions, and relay approval prompts straight into a channel, answer from your phone and the work keeps moving. Connected over Socket Mode; you stay in control of the tokens.

Approvals to your pocket

A blocking question becomes a Slack message. Reply and the agent unblocks instantly.

Live progress feed

Dispatches, hand-offs, and verdicts post to a channel as they happen, no need to be at the Mac.

Your tokens, your control

Bot and app tokens are stored locally and scoped to the connection. Disconnect any time.

Remote oversight

Stay in command from the couch.

The team runs on your Mac. You don't have to be sitting at it.

Claudeverse mirrored to an iPad over local WiFi

iPad command center, not desk anxiety.

Mirror the whole team to an iPad over your local WiFi, scan a QR code in Safari, no app to install. Watch progress, answer a blocker, approve a step from the kitchen. Nothing leaves your network.

  • Live mirror, no app install, just Safari + a QR code
  • macOS notifications + dock badge when a team needs you
  • Streams over local WiFi, nothing leaves your network

Built like a control plane, not a wrapper.

The things that make a team trustworthy at scale.

Stable agent identity

Every agent has a permanent ID that survives restarts, so prompts and history always bind to the right worker, never the wrong row.

Handoff guardrails

A round cap stops verify↔fix ping-pong from running away. If a thread loops too long, the team pauses and asks you.

Confidence-graded status

Awaiting-input and awaiting-approval are detected with a confidence level, notifications only fire when it's sure.

MCP profile management

Per-agent MCP config without touching your global setup. Imported credentials are extracted to a 0600 file and injected at launch.

System health at a glance

Coordinator, board, disk, and Claude's own status, surfaced live, so you know when an outage (not your team) is the problem.

Native macOS craft

A real Mac app with full terminal fidelity, compaction control, and four themes, including a warm Sunset. Apple Silicon & Intel.

Invite-only beta

Join the private beta.

Claudeverse is invite-only while we polish the first release.
Drop your email and we'll send you an invite code.

Claudeverse

Autonomous teams · macOS · iPad & Slack

Beta access

We'll send an invite code once a spot opens. No spam, no marketing, just your code.

What's inside

  • Autonomous Tag Teams, role routing & hand-offs
  • Verifier that drives the real browser & iOS simulator
  • Human inbox with secure credentials & Explain
  • Slack relay + iPad command center
  • Unlimited agents & workspaces · 100% local
Have an invite code?

Questions

What is Claudeverse, in one sentence?
It's a macOS app that runs autonomous teams of Claude agents. You give a team a spec; it decomposes the work onto a shared board, routes tasks by role, hands off between agents, verifies the result by driving the real app, and only asks you when something genuinely needs a human.
How is this different from running a few Claude windows in parallel?
Parallel windows still make you the router, the reviewer, and the memory. A Tag Team moves that into the software: a coordinator assigns work by role and only when blockers are done, agents hand off through a shared board, and a Verifier proves the feature works before anything is called "done." You step in for decisions, not dispatch.
What does the Verifier actually do?
It's a QA specialist agent that exercises the real feature in a real browser for web work and a real iOS simulator for mobile, and proves each acceptance criterion with a deterministic assertion. It emits one of three verdicts: verified (done), failed (files a bug back to the code owner), or blocked (an environment problem it asks a human to resolve). It never signs off on a hunch.
Does it run on its own, or do I have to babysit it?
It runs the build → verify → fix loop on its own. When an agent hits something only you can resolve (a login, an API key, a product decision), it surfaces a card in Your Tasks (or pings you in Slack / on the iPad) and auto-resumes the moment you answer. A round cap pauses any thread that loops too long. You set the autonomy.
Do I need a Claude subscription?
Yes. Claudeverse wraps the Claude CLI, so you need either a Claude Pro / Max subscription or Anthropic API access. Claudeverse itself doesn't make API calls, it doesn't add to your Claude bill.
What Mac do I need?
macOS 13 Ventura or later on Apple Silicon (M1/M2/M3/M4) or Intel. Mobile verification additionally needs Xcode's iOS simulators. Claudeverse is macOS-only for now, Windows and Linux are on the roadmap.
Where does my data live?
Claudeverse itself adds no telemetry, your prompts, Claude's responses, and app state stay local in SQLite, and secrets are written to a 0600 file on your Mac, never in plaintext to the database. Integrations you explicitly connect (a shared task board, Slack) only ever receive what you route through them.
Is Claudeverse affiliated with Anthropic?
No. Claudeverse is an independent product built on top of the Claude CLI. It is not made by, endorsed by, or affiliated with Anthropic.

The shift is already happening

Stop managing windows.
Start running a team.

Claudeverse is how a team of Claude agents plans, builds, and verifies real work, together. Request a beta invite while the program is still open.

Claude Code gave developers agents. Claudeverse gives them a team.

Request beta invite →