Skip to content
Install

Troubleshooting Greenhouse

Use this page to fix common Greenhouse failures. Entries are grouped by what you see, such as an error message or a machine that does not start, and each one lists the cause and the fix. Many fixes need an operator, so the entries say when to contact one.

Cause: The greenhouse graft is not installed on this instance. /sandboxes shows a not found page.

Fix: Ask an operator to install the graft. Instance admin > Grafts shows whether the graft loaded, and the reason if it failed.

Sandbox feature is not enabled for your account

Section titled “Sandbox feature is not enabled for your account”
Sandbox feature is not enabled for your account

Cause: The instance has PostHog configured, and your account does not have the greenhouse feature flag. Greenhouse is also missing from the sidebar in this case.

Fix: Ask an operator to turn on the flag for your account.

Cause: On an image without the required tools, the first boot installs them. Later starts skip what is installed. The workspace shows the current step:

Step What happens
Starting the machine The runner brings the machine up.
Installing packages Installs sshd, git, and tmux, if missing.
Installing the runtime Installs Node.js, if missing.
Installing coding agents Installs the agents, if missing.
Preparing storage Formats the home disk on first boot, and mounts it.
Setting up your account Creates the dev user, SSH host keys, and SSH configuration.
Starting services Starts SSH and the agent server.

Fix: Wait for the steps to finish. If the machine does not start, refer to the next entry.

These logs are on the machine. Read them from a terminal or over SSH.

File Contents
~/.setup.log The machine setup script
~/.clone.log The repository named at creation
~/.opencode-serve.log The agent server
/var/log/orchard-bridge.log The SSH bridge, on Vercel machines
Sandbox did not become ready in time
Sandbox resource never appeared in the cluster

Cause: The machine stayed in Creating or Starting for 10 minutes, or 30 minutes for a fork. The second message means the runner did not create the resource.

Fix: Start the machine again. If it fails the same way, contact your operator.

Cause: Orchard carries out starts, stops, deletes, and credential syncs through a queue. A worker retries a failed change up to 10 times, with increasing delays. Transient errors clear without action. After the last attempt, the error stays on the machine.

Fix: Fix the cause, then queue the failed changes again. The host or an organization admin can select Retry in the machine menu, or run the following command:

orchard sandbox retry my-machine

If the error returns, or the machine shows Applying for more than a few minutes, contact your operator.

The API returns one of these messages when you create a machine on an unavailable runner:

Message Cause
Kubernetes sandboxes are turned off on this server The operator turned the runner off.
Vercel sandboxes are turned off on this server The operator turned the runner off.
Kubernetes is not configured on this server The server has no cluster access.
Vercel sandboxes need vercelToken (SANDBOX_VERCEL_TOKEN), ... Vercel credentials are incomplete. The message names what is missing.
Vercel sandboxes need SECRETS_ENCRYPTION_KEY so their public ports are password protected The instance has no encryption key.
No sandbox runner is available Every runner is off or misconfigured.

Fix: Contact your operator. Machines already on a runner that is turned off keep working.

The New machine sheet offers only available runners.

Message Cause
None Idle stop. There was no terminal or SSH activity for the idle limit, and no agent was working.
Stopped after reaching max lifetime The machine reached the maximum time since its last start.
Stopped by provider timeout Vercel’s timeout stopped the machine.

Fix: Start the machine again. Files on disk remain. Running processes, including dev servers and agent turns in progress, do not.

Message Cause and fix
Sandbox resource disappeared from the cluster Something outside Orchard deleted the resource. Start the machine to recreate it. If its disk is also gone, it starts empty.
The Vercel sandbox no longer exists The sandbox is gone on Vercel. Start the machine to recreate it.
The Vercel sandbox failed Vercel reported a failure. Start the machine again.
Failed to create sandbox: ... Creation failed. The rest of the message states why.

Cause: Host keys stay the same across stops and starts. Deleting a machine and creating another with the same name changes them. The stale key is in a known_hosts file that Orchard does not manage.

Fix: Run the installer and remove the stale entry:

orchard sandbox ssh-config --install
ssh-keygen -R orchard-<name>
Permission denied (publickey)

Cause: None of your SSH identities matches the key your client offered.

Fix: Run orchard sandbox ssh-config --install, or add your key in Account settings > Greenhouse > SSH identities.

The pull request button reads Link GitHub to push, or an action fails with github_not_linked.

Cause: Your account has no linked GitHub account. Pushes, pull requests, and merges use your own account.

Fix: Link it in Account settings > Connections. You can commit without it.

No GitHub account is linked to read checks

Section titled “No GitHub account is linked to read checks”
No GitHub account is linked to read checks

Cause: Neither you nor the host has linked GitHub.

Fix: Link it in Account settings > Connections.

Message Fix
"<repo>" is not a git repository in this sandbox The repository is not cloned yet, or the clone failed. Check ~/.clone.log.
Could not find the base branch in that repository Select another base branch.
This session is archived Restore it with orchard sandbox session archive --restore.
Sandbox is not running Start the machine.
Message Cause and fix
The sandbox is still setting up. Try again in a moment. The dev user does not exist yet. Wait for the machine to finish booting.
Interactive shells are not available for Vercel sandboxes The installed Vercel SDK cannot open shells. Your operator must upgrade it.
Message Cause and fix
Previews need a preview domain; ... Your operator has not set a preview domain, so Kubernetes previews are off.
That port cannot be previewed Use 3000, 3001, 4173, 5173, 8000, 8080, or 8787, or declare the port in .orchard/greenhouse.json.
This preview link has expired or was already used Links are single use and last 60 seconds. Open the preview again from Greenhouse.
Nothing is answering on this port in the sandbox Nothing listens on the port, or it listens on localhost only. Bind to 0.0.0.0.
Cross-site requests to previews are not allowed Another site called your preview. Open the preview directly.
You no longer have access to this sandbox You were removed from the machine.
Message Cause and fix
This cluster has no VolumeSnapshotClass for the sandbox disk's storage class, ... The cluster cannot snapshot the storage, so Kubernetes forks are off.
This machine has no home volume to copy yet Start the machine once before forking it.
Only the machine's owner can fork it: ... The copy contains the host’s home directory and credentials.
Forking needs Vercel drive forks, which this team's plan doesn't have The operator’s Vercel plan does not include drive forks.
Sandbox limit reached (...) A fork counts against your machine limit. Ask a superadmin to raise it.

Collect the machine’s name, runner, status message, and the relevant log above. Then contact your instance operator.