Attach Desktop / private beta
Give every AI worker its own computer.
Run Claude, ChatGPT, Codex, and OpenClaw sessions in isolated machines. Host those machines in your own DigitalOcean or cloud account, watch their screens, and require approval before deploys, credentials, or paid actions.
Private-cloud pilots start with a fit call and a small beta reservation before monthly infrastructure or managed support.
Your agents should not borrow your laptop.
The more capable agents become, the worse it feels to give them your real browser, shell, logged-in accounts, and card-backed tools. Attach Desktop moves that work into machines you can supervise.
Agents share your browser, terminal, and logged-in accounts.
Long workflows disappear into separate chats with no clear owner.
Most hosted bots put sensitive work inside someone else's SaaS boundary.
Deploys, paid tools, and external sharing need supervision before they happen.
Research a market, draft a page, review it in a browser, then ask for approval.
Run a coding worker and a browser reviewer without sharing your laptop session.
Give an operator scoped credentials while finance and production require approval.
The work can run in your cloud, not ours.
Team chat is familiar. Attach Desktop makes the private computer the product: each worker can live inside infrastructure you control. In private-cloud mode, your data does not need to pass through Attach-hosted machines.
Private cloud
Run the worker computers inside your own DigitalOcean, AWS, GCP, or other cloud account. Your repos, browser sessions, credentials, generated files, and machine logs stay in your infrastructure.
Attach managed
Use Attach to set up the control plane, policies, updates, and supervision workflows while keeping risky actions behind approval gates.
Community edition
Run the core worker console yourself when you want maximum control, a lower-cost internal deployment, or a starting point for your own agent stack.
The page opens cleanly in a browser. I saved the review before anyone touches production.
The launch package is ready. Asking for approval before any external action.
Prepare the final checklist only. Publishing, sharing, and paid actions stay locked.
The controls that make an AI team usable.
Persistent coworkers
Named agents keep role, context, and work history instead of starting as disposable chats.
Isolated computers
Each worker gets a controlled remote machine with reset, snapshot, and watch/takeover paths.
Visible handoffs
Worker-to-worker messages become a first-class activity stream you can inspect, pause, and redirect.
Scoped credentials
Different workers can have different accounts, budgets, approvals, and expiration windows.
Guardrails by default
Attach Guard policy, audit trails, and spend limits sit in front of the work, not after it.
Human supervision
You stay above the workflow: observe, intervene, approve, revoke, or reset the machine.
Start self-hosted, or let Attach help operate it.
Pilot pricing keeps the buyer decision simple: community edition for builders, private-cloud setup for early customers, and managed support for teams that want help running the worker fleet.
For technical teams that want to run the core worker console themselves.
For founders and agencies that want a private-cloud pilot scoped before they commit to a monthly plan.
For teams that want Attach to help operate the control plane, policies, and worker rollout.
A thin slice that can grow into the real control plane.
The first paid pilot should connect one worker record, one remote computer in the customer's cloud, one visible handoff stream, one credential scope, and one approval gate.
Start with one supervised worker.
The first build should connect one named worker to one isolated computer, record handoffs, and require approval before external spend or production access.