`apus connect`

Connect the current machine to an Apus Gateway as a headless node. The command redeems a short-lived bootstrap credential, saves the Gateway endpoint in the existing node-host state, and runs the same runtime as apus node run.

Create a join command

On the Gateway host, use admin credentials to mint a single-use join URL:

apus devices join-code

The command prints the URL and a pasteable command that installs a background node service and enables it to run agent sessions:

npx -y apus connect https://gateway.example/j/<shortcode> --service --session-host

Only use session hosting on a machine you trust as shared Gateway infrastructure. The printed command makes that consent explicit with --session-host; the node runtime's default remains non-hosting. See Session hosting. For a command-only node, omit --session-host and keep --service. The -y flag skips npm's package-install confirmation, not Apus pairing or session-hosting consent.

The shortcode has 128 bits of entropy, expires with the setup credential after about 10 minutes, and can be fetched exactly once. Mint another code if it expires or has already been used.

Connect in the foreground

To connect without installing a service, omit --service and --session-host:

npx apus connect https://gateway.example/j/<shortcode>

Set the device name during enrollment when useful:

npx apus connect https://gateway.example/j/<shortcode> --display-name "Build Node"

The node stays in the foreground until you stop it.

To expose only selected commands, pass a comma-separated list of exact command IDs. For a Session Share node:

apus connect <join-url> \
  --commands apus.sessions.list.v1,apus.sessions.read.v1

The flag is repeatable. The allowlist is durable node state and also applies after --service installation; omitting it preserves a saved list. It filters available commands and their required capabilities, and disables computer use, skills, plugin-tool publication, MCP servers, and worker hosting. Startup fails when no requested command is available. The Gateway pairing approval shows the resulting declared commands.

To restore the full default surface, use apus node run --all-commands for a foreground node or apus node install --force --all-commands for an installed service. When enrolling again, use apus connect <join-url> --all-commands (add --service for a service). This forgets the saved allowlist; --all-commands cannot be combined with --commands.

To let that foreground process host full worker sessions, give explicit local consent with --session-host:

npx apus connect https://gateway.example/j/<shortcode> --session-host

Foreground consent applies only to that process. It does not change apus.json, so the next normal node-host start remains non-hosting.

Reconnect a paired node

Join URLs and setup codes are single-use, so rerunning the original apus connect <join-url> command after the node stops reports that the join code was not found or has expired. The node keeps its paired device token and Gateway endpoint in node-host state. Reconnect with apus node run, repeating any process-scoped flags:

apus node run --session-host

Running apus connect without a target does not connect. When node-host state has a saved Gateway endpoint and a node device token, it exits with an error that prints the matching apus node run command for the flags you passed, to use if that pairing is still current, and the apus connect command to use with a new join URL otherwise. With --service, it prints apus node install --force instead, preceded by apus config set nodeHost.workerRuns.enabled true when you also passed --session-host. If the first enrollment never completed, it only points to a new join URL. The device token is not tied to one endpoint: after a failed enrollment with a different Gateway, the reconnect command can fail, so enroll again instead. To enroll the machine again, mint a new join URL with apus devices join-code.

Environment-managed cloud nodes

Worker providers use --ephemeral for disposable cloud machines:

npx apus connect <setup-code> --ephemeral

This process hosts worker sessions even when the machine's durable node config has worker hosting disabled. It does not install a service and cannot be combined with --service or --session-host. The Gateway owns the setup identity and paired-node lifetime: provider replay resumes the persisted device token after the one-shot setup credential is consumed, and environment teardown removes the node role after releasing the cloud lease.

--ephemeral is intended for provider-managed state directories on throwaway machines, not as a shortcut for enrolling a personal device.

Install as a service

Pass --service to redeem the bootstrap credential and install the node host as the platform user service:

npx apus connect https://gateway.example/j/<shortcode> --service

Apus completes the first authenticated connection before installing the service. The short-lived bootstrap token is never stored in the service command or node-host configuration; later starts use the durable paired-device token. When restarting against that saved endpoint, config credentials for a co-located Gateway do not override the paired token. Explicit APUS_GATEWAY_TOKEN or APUS_GATEWAY_PASSWORD environment credentials still take precedence. Use apus node status to inspect the installed service.

The service does not host worker sessions by default. To consent to full worker-session hosting, add --session-host:

npx apus connect https://gateway.example/j/<shortcode> --service --session-host

The one-shot bootstrap connection authenticates and saves the durable device identity without advertising worker hosting. Only after that connection succeeds does Apus persist nodeHost.workerRuns.enabled=true, preserving the rest of the config, and install the service. If the config write fails, service installation does not start. The installed service advertises worker hosting and exact capacity from this durable consent when it starts.

Accepted targets

Option	Purpose
`--commands <ids>`	Save an exact comma-separated node command allowlist (repeatable). Applies to foreground runs and installed services.
`--all-commands`	Advertise the full default command surface and forget any saved `--commands` allowlist. Conflicts with `--commands`.
`--display-name <name>`	Set the node display name.
`--service`	Pair first, then install the node as a user service.
`--session-host`	Consent to worker hosting. An explicit command allowlist disables hosting.
`--ephemeral`	Run a provider-managed disposable worker node.
`--target-file <path>`	Read a join target from a file and consume the handoff after a successful read.

apus connect <target> accepts:

an https://<gateway>/j/<shortcode> join URL;
an oc-pair://<setup-code> URL;
a bare base64url setup code.

--target-file <path> accepts a regular file up to 64 KiB. It removes the path only after reading a non-empty target. If the file is empty, too large, unreadable, or not a regular file, Apus leaves it in place. A symlink is allowed; Apus reads its target, removes the symlink after a successful read, and keeps the backing file. The dormant installer wrapper uses this handoff to keep the single-use target out of child-process arguments.

Join URLs must use HTTPS. Plain HTTP is accepted only for loopback Gateway URLs such as http://127.0.0.1/j/<shortcode>. Direct setup codes can carry the Gateway TLS certificate fingerprint, which lets the node host pin a self-signed Gateway certificate after decoding the payload.

The payload determines the saved host, port, TLS mode, WebSocket context path, and ordered fallback endpoints. Normal and foreground connections do not add apus.json keys; --service --session-host explicitly persists the worker hosting consent described above.

Revocation behavior

A join code and a paired device have separate lifecycles:

Burning or expiring a join code prevents another enrollment with that code.
It does not disconnect or remove a node that already redeemed it.
To revoke a normal enrolled machine, remove its paired device with apus devices remove <deviceId>.
Environment-managed --ephemeral nodes are removed automatically when their owning cloud environment is destroyed.

Troubleshooting

If the join URL reports that it is missing or expired, mint a new one with apus devices join-code. A used code intentionally returns the same result as an unknown code. If this machine already redeemed it, reconnect with the saved pairing instead; see Reconnect a paired node.

If an HTTPS join URL uses a certificate the local machine does not trust, use the direct oc-pair:// or bare setup-code form that includes the TLS pin.

See Node for service management, explicit connection flags, node state, and exec approval behavior.