Adding Devices
Every managed machine has a config record, an authenticated Yokai agent, and a local daemon tunnel. The Devices route owns the full lifecycle.
Choose a source
Section titled “Choose a source”Open Devices with 2, press A, and select one of the available sources.
SSH config
Section titled “SSH config”Yokai reads ~/.ssh/config through the daemon and normalizes each usable host alias. Hostname, user, port, and identity file are prefilled, including encrypted-key detection.
Use this path when SSH already works from your terminal:
ssh gpu-box-01Tailscale
Section titled “Tailscale”Yokai calls the local Tailscale CLI, reports whether the daemon is installed, running, or awaiting login, and lists peers with hostname, DNS name, IPs, OS, online state, and ACL tags.
Peers carrying tag:ai-gpu are highlighted as recommended. The tag is helpful, not mandatory.
For the complete setup—including tag ownership, Admin Console and CLI workflows, least-privilege access, and verification—see Tailscale GPU Nodes.
Manual
Section titled “Manual”Enter a hostname or IP, SSH user, port, and optional key path. This is the right path for a host that does not appear in SSH config or Tailscale.
SSH authentication
Section titled “SSH authentication”Connection resolution supports normal SSH-key and agent workflows. If a configured key is encrypted, Yokai can use the running SSH agent or request the required credential in the setup flow.
The selected user must be able to:
- write the agent binary and service configuration;
- run Docker commands;
- inspect GPU and system state;
- create monitoring containers when monitoring is selected.
Bootstrap sequence
Section titled “Bootstrap sequence”The bootstrap operation streams progress through the daemon:
- connect over SSH;
- check the OS, architecture, Docker, GPU runtime, disk, and ports;
- generate a 64-character agent bearer token when needed;
- copy the matching Yokai binary;
- write the agent configuration and systemd unit;
- start and verify
yokai agenton port7474; - open the local SSH tunnel;
- optionally install Prometheus, Grafana, node_exporter, and dcgm-exporter.
Device operations
Section titled “Device operations”| Action | Key | Result |
|---|---|---|
| Add | A | Open source picker and setup flow |
| Edit | E | Change connection and display details |
| Test | T | Test the selected device |
| Upgrade | U | Replace and restart the selected remote agent |
| Test all | Shift+T | Run a fleet-wide connectivity check |
| Upgrade all | Shift+U | Upgrade all configured remote agents |
| Remove | X | Remove device state and associated service records |
CLI workflow
Section titled “CLI workflow”yokai devices add \ --host 100.64.0.2 \ --label "GPU box 01" \ --ssh-user ubuntu \ --ssh-key ~/.ssh/id_ed25519 \ --type tailscale
yokai devices bootstrap 100.64.0.2yokai devices test 100.64.0.2yokai devices listTroubleshooting onboarding
Section titled “Troubleshooting onboarding”- Verify direct SSH first with the same user, port, and key.
- Verify
docker infosucceeds for that user. - Verify the NVIDIA container runtime can run
nvidia-smiinside a container. - Check
~/.config/yokai/daemon.logfor bootstrap or tunnel errors. - If the host changed, edit it in Devices and test again before upgrading.
See Troubleshooting for command-level checks.