GuidesHosting choices

Hermes "Another gateway already owns this host": what the PID is

Hermes Agent refuses to start a gateway because another holds the host's one slot. What the PID and "profiles: unknown" mean, the boot race, and the fixes.

November 19, 2026The Everpod team
The short answer

Hermes Agent allows one gateway per machine (per Linux user), which serves every profile, and the PID in the message is the process that already holds that slot. “No bound port; profiles: unknown” means it had claimed the slot but had not yet said what it serves, which is what a gateway looks like in its first seconds. Under systemd this is usually a start-up race: the losing unit exits with code 75 and systemd retries it. If it keeps happening, look at the PID with ps -fp <PID>, then either let that gateway serve all your profiles (hermes gateway migrate --multiplex --dry-run shows the plan), or stop it and start yours. --force starts a second gateway anyway, and can connect one bot twice.

The message in full

In Hermes v0.23.2 (released November 16, 2026), the gateway’s source prints:

❌ Another gateway already owns this host: PID 3184 (no bound port; profiles: unknown)
   Exactly one gateway per host serves every profile, so this process will not start a
   second one (it would double-bind this profile's platforms).
   Fold every profile onto the owner:  hermes gateway migrate --multiplex
   Or stop the other gateway first, then start this one.
   Or start one anyway (skips the host-lock check):  hermes gateway run --force
   (--replace does not skip this check; it only replaces an owner that serves this profile.)
Copied

When the gateway is at its normal work, the part in brackets carries an address and a list, such as 127.0.0.1:<port>; profiles: default, coder. A gateway that has just won the slot records itself with an empty list, and fills it in once it is running, so “unknown” usually means the other process was still starting. If even the record is missing, the line reads “owner unknown (its record is gone)”. A near relative, “⏳ A gateway already owns this host and has not published its served set yet”, is the same situation met on a different code path.

Why there can only be one

Since multiplexing became the default, Nous’s multi-profile gateways page says “Exactly one gateway per host is the inbound process for every profile.” Each profile keeps its own bot tokens, sessions and memory, but one process connects them all. The slot is a lock file held for as long as the owning process lives, one per Linux user, under ~/.local/state/hermes/gateway-locks/ (or $XDG_STATE_HOME/hermes/gateway-locks). The operating system releases it when that process dies, so a crashed gateway does not leave the slot taken.

Running a profile’s gateway is meant to join the existing one, not start a second. When the owner already serves that profile, Hermes prints “✓ The host gateway already serves profile ‘coder’ — nothing to start.” and exits 0. The refusal comes when the owner cannot be asked, which is the case while it is still starting.

Which process holds the slot

The PID is an ordinary Linux process ID:

ps -fp 3184
systemctl --user status 3184
Copied

The first shows its command line, including any -p <profile>; the second names the systemd unit it belongs to, if any (drop --user for a gateway installed with sudo hermes gateway install --system). The default profile’s unit is hermes-gateway.service and a profile’s is hermes-gateway-<profile>.service. From Hermes itself, hermes status --full on the default profile reports the gateway and the profiles it serves, and hermes -p coder gateway status says whether that profile is running through it.

At boot, under systemd

This is where most people meet the message. The refusal exits with code 75, which systemd, s6 and launchd treat as “try again shortly”, and the docs describe the race outright: “Two units started at once can both see no host process yet; the host lock decides which one runs.” The loser joins on its retry, so one line in the journal followed by a working bot needs nothing doing. To see how many times it went round:

journalctl --user -u hermes-gateway.service -b
Copied

Two open bug reports describe it not settling. In issue #152451, with some profiles still set to run their own gateway (gateway.standalone: true), a standalone gateway can win the slot at boot and the main gateway exits 75 because it asks too early; the reporter says the retry succeeds in practice, and a fix is in an open pull request (#152873) as of November 20, 2026. In issue #157832, after an update that adds a database index, the gateway holding the slot spends a long time migrating one profile’s store before it says what it serves, and every other gateway keeps exiting 75 until it finishes; the reporter saw up to 25 minutes on an 8.3 GB store. In that case the cure is waiting for the migration to finish. Retries like these do not stop on their own: according to issue #150328, filed against v0.23.0, the unit Hermes writes turns off systemd’s start-rate limit, and the issue cites a loop of 9,152 restarts over 33 hours.

The choices, and which to take

Installing a service per profile is not a way out either: in the multiplexed setup, hermes -p coder gateway install refuses with exit 78 and “Profile ‘coder’ does not get a gateway of its own.” The single service is the default profile’s, set up as in installing Hermes Agent on a Linux server. If the message started right after an update, the gateway steps in when hermes update fails are worth checking too.

Hermes installed on an always-on developer pod, which you run yourself.

A developer pod is a permanent cloud computer (VPS), set up and secured for you, with your pick of Claude Code, Codex, OpenCode, Pi, Hermes and OpenClaw installed. Run more agents at once than your laptop can hold, and keep them working when it’s closed. From $24 a month, built in about ten minutes.