Your own AI machine
in the cloud.
Your agent already knows how to reach the tools you run on
From sign-in to shipped
Four steps between you and an agent doing real work on a real computer.
A dedicated cloud computer, just for you
Sign in and Hoshi provisions your machine — the place your agent lives. It is isolated from everyone else’s, always on, and never shared between runs.
Start from your organization’s boilerplates
Teams publish project boilerplates; you pull one as a checkout and it lands on your machine ready to run — dependencies, config, and all.
Conversations that build their own UI
Sessions are long-lived and agentic. When the agent needs input it answers with real interface — forms, choices, charts, tables — generated mid-conversation.
The agent asks with real UI — forms, choices, charts — not walls of text.
Schedules keep your machine working
Put recurring work on a schedule and your agent runs it unattended — daily digests, weekly audits, whatever your projects need. Come back to finished results.
Your machine keeps working while you sleep — on your schedule.
Built like infrastructure, feels like chat
Real machines, real isolation, real work — behind a surface as simple as a conversation.
A machine per person
Every user gets a dedicated cloud machine — its own runtime, filesystem, and workspace. Nothing shared, nothing leaked.
Everything authenticated
Clients never touch the agent runtime directly — every call goes through an authenticated proxy with an owner lock.
Organizations & boilerplates
Teams publish project templates once; every member pulls them onto their own machine and starts from working code.
Generative UI
The agent answers with interactive widgets — forms, choices, charts, tables — not just prose to copy out of.
Live previews
When the agent starts a dev server, the open port becomes a live preview you can click through from the chat.
Scheduled triggers
Cron for your agent: recurring sessions run on your machine unattended and leave finished results behind.
Isolation and control, not just a chat window
Every seam a multi-tenant agent platform needs — scoped access, audited action, infrastructure you can actually own.
A machine per person
Every user gets an isolated cloud machine — its own runtime and filesystem, never shared between runs.
Authenticated proxy, always
Clients never talk to the agent runtime directly. Every call carries a session or owns a lock, checked on the machine itself.
Provider keys stay server-side
Organization AI provider keys are stored and synced centrally — never shipped to the browser or the model.
Organizations & roles
Orgs, groups, and invites scope who can publish boilerplates, provision machines, and manage a team.
Activity you can audit
Sessions, machine actions, and provisioning events land in an activity trail — not a black box.
Bring your own infrastructure
The orchestrator is pluggable — run on Docker Swarm today, Kubernetes tomorrow, on your own hardware.
Give your agent a home
Your machine is a sign-in away — provisioned in moments, yours for as long as you need it.