If you let Hermes Agent perform its own gateway restart in Telegram, and the gateway dies and doesn't come up, this is not an illusion. Public issue #6666 describes this scenario: in the 0.8.0, WSL2, systemd user service environment, triggering a restart from a Telegram session, it may not wake up because the gateway stopped due to PID race.
The current most stable approach
- Don't let it self-restart in a Telegram conversation that is carrying its own gateway.
- If you are a systemd user service, you should prefer to use the
systemctl --user restart hermes-gatewaymanually. - If it doesn't get up after restarting, use
systemctl --user start hermes-gatewayto pull it up first.
Why is this happening?
The core of the problem is not Telegram, but "a running gateway process trying to restart itself from within." As of April 12, 2026, the phenomenon given in the public issue is that the old process has just stopped, and the new process detects that the old PID is still there, so it withdraws directly, and the last two do not survive.
So the short-term pitfall avoidance idea is simple: let the external service manager restart it, not let it restart itself in its own chat session. If you still rely on proxy environment variables, remember to check the systemd override as well to avoid losing the proxy configuration after restarting.
Official open source address: https://github.com/NousResearch/hermes-agent; Official document entry: https://hermes-agent.nousresearch.com/.
