Remote dev machines

Spin up a managed Linux or Windows machine for a project, sync your local files to it, sleep and resume it, and share a durable link.

A remote dev machine is an always-on workspace that taiku provisions for you, no VMs to manage, no SSH keys to copy, no port forwarding to wire up. Machines run Linux by default; Windows machines are available on paid plans.

Click to scrub

You point taiku at a project, it prepares a machine, syncs your files, and opens a normal taiku session running on that remote machine. Your files live on the machine's persistent disk, so they survive sleeps, restarts, and reconnections.

Local, remote, or GPU

Pick the kind of workspace that fits the job. You can move between them at any time, the same project can run locally today and on a remote machine tomorrow.

	Local session	Remote dev machine	GPU machine
Where it runs	Your own computer	A managed Linux or Windows machine	A managed Linux machine with a GPU
Best for	Day-to-day work on your hardware	Always-available compute, heavier builds, sharing	Model training, rendering, anything that needs a GPU
Files	Your local disk	A persistent disk on the machine, synced to a local folder	Same as a remote dev machine
Stays available when your laptop sleeps	No	Yes	Yes
Setup	None	One command	One command (when enabled for your account)

Create one from the dashboard

You don't need a local install to get a remote machine, you can create one straight from the dashboard, including on your phone.

Open the Cloud panel on your dashboard.
Click Create your cloud workspace, give it a name, and click Create. On paid plans you'll also see an operating system choice, pick Linux (the default) or Windows.

taiku spins up your machine and creates a project folder named after what you typed (for example, workspaces/my-app). The first machine takes a little longer while it's provisioned; after that it's ready to open like any other session.

You get one machine per operating system: a Linux machine on any plan, plus a Windows machine on paid plans. To start another project on a machine, click + New project, name it, and taiku adds another project folder on the same machine, each project keeps its own files, side by side.

Start a project from a GitHub repo

When you create a Linux cloud workspace, you can start it from one of your GitHub repositories instead of an empty folder.

In the create form, keep GitHub repo selected (it is the default).
The first time, click Install taiku on GitHub. GitHub asks which account or organization to install on and which repositories taiku may read, and that list stays under your control. Private repositories and organization repositories (including ones behind single sign-on) work once taiku is installed on that account.
Back in taiku, pick the repository, then choose a branch or leave it on the default branch.
Click Create. taiku clones the repository into the new project so it is ready the moment the machine opens.

To start empty instead, just click Create without choosing a repository. You do not need to install taiku on GitHub for a blank project. taiku only reads the repositories you picked during install, and you can change that selection any time from your GitHub settings.

Start a project from another computer

If you have another computer connected to taiku, you can copy one of its folders into a new Linux project instead of starting from scratch. In the create form choose Another computer, pick the device and the folder, and taiku copies it into the new project.

Start a remote machine

If you work locally, you can also start a workspace from the command line. Open a terminal in your project folder and start a workspace for it. The first run takes a little longer while the machine is provisioned and set up; later runs for the same folder reuse the existing machine.

When the machine is ready, taiku copies the session link to your clipboard and opens it in the desktop app or your browser, exactly like a regular session. Keep the workspace running while you want the remote session and file sync to stay active.

How many remote machines you can keep depends on your plan, see Plans and limits.

Bind a local folder

Each remote machine is tied to a local folder on your computer. taiku keeps that folder in sync with the machine, so you can edit files locally and run them remotely, or vice versa, and both sides stay current.

Click to scrub

You can use the same machine from more than one computer. From a new computer, attach to an existing machine and choose the local folder you want it to sync there. The files all live on the machine's persistent disk, each computer just keeps its own local-folder binding.

Work with git worktrees

If your project uses git worktrees, they come along. Any worktrees under .claude/worktrees/ sync to the machine as real, working worktrees, each on its own branch with your uncommitted changes intact, so you pick up right where you left off. Worktrees you create on the machine (for example with taiku switch) sync back to your computer the same way, and your commits flow in both directions.

One detail worth knowing: changes you had staged but not yet committed arrive as ordinary unstaged edits. Nothing is lost, you just re-stage before committing if you want them grouped the same way.

One narrow case: if the folder you start a machine from is itself a git worktree of a repository that has no remote, the cloud copy carries your branch tips with their current state, and the full commit history stays on your computer. Starting from a normal repository, or any repository with a remote, brings the full history along. taiku prints a note when this applies.

Storage that grows with you

A machine's disk starts small and grows automatically as your project needs more room, up to your plan's limit (see Plans and limits). You never have to pre-size anything. If you want headroom ahead of a big checkout or build, you can also grow it yourself with taiku cloud resize.

GPU machines

When GPU machines are available on your account, you can start one the same way you start any remote machine. Use them for work that needs a graphics card, model training, rendering, simulation, and sleep them between runs so you only pay for compute while you're actually using it. Everything else (file sync, sharing, the dashboard) works the same as a standard remote machine.

Manage from the dashboard

You can control a workspace's whole lifecycle from the dashboard or the terminal:

Sleep pauses the machine and stops compute charges. Your files stay intact, and resuming brings them right back.
Resume wakes a sleeping machine.
Restart reboots the machine without erasing anything.
Detach removes a workspace's binding from the current computer but leaves the machine running for your other computers.
Delete destroys the machine and its disk. This cannot be undone, and the files do not come back.

The dashboard lists every workspace on your account, when each was last used, and its current state, so you can sleep idle machines or clean up ones you no longer need.

A machine you haven't opened in a long while is archived for you automatically, moved to cheaper storage with nothing to configure. Everything is kept, and the next time you open it taiku brings it back on its own, that first open just takes a little longer while it comes back. Sleeping a machine yourself is the quick, always-ready pause; archiving is the hands-off, long-term one, and you never lose anything either way.

You can hand someone a stable link to a remote machine so they can open it without a taiku account. From a workspace session, open its Share panel and create a pinned link. Pick how long it should stay valid, anywhere from a day to a year, and copy the link to send it.

The link is the access grant, so share it only with people you trust. taiku keeps the secret that unlocks the session in the link itself and never sends it to the server, which is how someone can open the session without an account. See how access links work for what that link does and does not reveal.

A pinned link is durable: it points at the machine, not at one live session. When someone opens it, taiku wakes the machine if it's asleep, starts a fresh session scoped to the project folder, and sends them straight in. They can read and edit files inside that folder, but the session is sandboxed and cannot reach the rest of the machine.

The same panel shows the links you've created, when each expires, and when it was last opened. To cut off access, revoke a link, it stops working immediately, and expired or revoked links send recipients to a friendly error page instead of a broken one.

Bring your tools along

When you start a new remote machine, taiku makes a best-effort, one-time transfer of your supported local tool and agent setup so your agent workflows are ready to go right away. You can skip that transfer, run it again after rotating credentials, or keep general shell and tool config while blocking a specific agent's saved login. Your taiku preferences, including which files to leave out of sync, carry over too and refresh on later runs.