You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The nousresearch/hermes-agent:latest Docker image (v0.15.0 / v2026.5.28) crashes on startup with three init failures. The container exits with code 127.
Error Log
/etc/cont-init.d/01-hermes-setup: 2: exec: /opt/hermes/docker/stage2-hook.sh: not found
/opt/hermes/.venv/bin/python: No module named hermes_cli.container_boot
/run/s6/basedir/scripts/rc.init: 91: /opt/hermes/docker/main-wrapper.sh: not found
Root Cause
1. Missing init scripts — Dockerfile line 229-232 creates 01-hermes-setup which execs /opt/hermes/docker/stage2-hook.sh. ENTRYPOINT (line 290) references /opt/hermes/docker/main-wrapper.sh. Both files exist in the source tree at docker/stage2-hook.sh and docker/main-wrapper.sh but are not present at those paths in the built image. The COPY --chown=hermes:hermes . . (line 159) should include them — likely a build-context or multi-stage issue.
I've had to roll back to nousresearch/hermes-agent:v2026.5.16 to get this working. I'm not on unraid, I'm running in a docker container on omv OS, which is Debian based
We're so back. Apparently, the issue can be caused by updating via the UI instead of updating the container's tag.
In any case, here's what solved it for me on Unraid. In short:
Remove all Hermes-related containers
Removed the shared volume. This shouldn't impact anything to do with your custom data but caveat emptor
In Unraid, go to Apps > Previous Apps > Docker then find + reinstall the Hermes containers. Doing it this way should preserve any custom settings you had
Additionally, I needed to add a new variable called HERMES_DASHBOARD_INSECURE with a value of 1 in order to access my dashboard locally
I've had to roll back to nousresearch/hermes-agent:v2026.5.16 to get this working. I'm not on unraid, I'm running in a docker container on omv OS, which is Debian based
I just tested on latest which was released 4 hours ago. Issue still persists.
We're so back. Apparently, the issue can be caused by updating via the UI instead of updating the container's tag.
In any case, here's what solved it for me on Unraid. In short:
Remove all Hermes-related containers
Removed the shared volume. This shouldn't impact anything to do with your custom data but caveat emptor
In Unraid, go to Apps > Previous Apps > Docker then find + reinstall the Hermes containers. Doing it this way should preserve any custom settings you had
Additionally, I needed to add a new variable called HERMES_DASHBOARD_INSECURE with a value of 1 in order to access my dashboard locally
Never used unraid, but I'm trying to translate your steps into the docker native commands for each step you have here. I've been doing step 1 already, so deleting containers didn't help me. Ill try translating your steps and see how far I get.
I apologize for the Unraid issue. It is caused by the use of a shared Docker volume.
This is required because Hermes WebUI must have access to the Hermes source code to install its own dependencies.
The only way to do this is to start Hermes Agent with a non-existent shared Docker volume. When the container starts, it will create the volume and copy the contents of the container's path into that volume. This same volume is, in turn, used by the WebUI to work.
The problem is that once the volume is created, updating the Hermes Agent container to a new version does NOT update the shared volume's content. As such, you are NOT using the latest version as expected UNLESS you first delete the shared volume.
Unfortunately, there is nothing I can do to make this more seamless unless Hermes WebUI removes this dependency. This process is based on the two-container deployment method described on their repo.
I note that the shared volume is not required for the Workspace template, so you can delete the hidden parameter with the shared volume if you are not using the WebUI.
Note that I am considering dropping the WebUI template because of this issue.
Quick root-cause recap since there's been more traction here:
What's missing from the v0.15.0 image:
stage2-hook.sh — referenced in the Dockerfile at RUN chmod +x stage2-hook.sh
main-wrapper.sh — missing from the image layer
container_boot module — removed from the source tree entirely (directory exists in v2026.5.16, gone in latest)
Cascade: Without stage2-hook.sh, the entrypoint fails. On Unraid's template (and likely any system using named volumes for the Hermes-Dashboard shared volume), the volume stays empty since the failing container never populates it.
Reproducers so far: arm64 (szmania), Unraid amd64 (subpanic, stonebert-png), OMV/Debian (szmania rolled back). The Unraid template maintainer (mmartial) also has users reporting the same crash in mmartial/unraid-templates#8 — currently advising a volume-nuke workaround not knowing it's a broken-image issue.
Workaround for anyone hitting this: pin to nousresearch/hermes-agent:v2026.5.16 until the next release.
We're so back. Apparently, the issue can be caused by updating via the UI instead of updating the container's tag.
In any case, here's what solved it for me on Unraid. In short:
Remove all Hermes-related containers
Removed the shared volume. This shouldn't impact anything to do with your custom data but caveat emptor
In Unraid, go to Apps > Previous Apps > Docker then find + reinstall the Hermes containers. Doing it this way should preserve any custom settings you had
Additionally, I needed to add a new variable called HERMES_DASHBOARD_INSECURE with a value of 1 in order to access my dashboard locally
thanks this worked for me. deleted containers, deleted share to /opt/hermes and then set that env var to 1.
Confirmed — v0.15.2 resolves this. I updated to v2026.5.29.2 today and the container starts cleanly. Both stage2-hook.sh and main-wrapper.sh are present in the docker/ directory now, and the container_boot module has been properly replaced by s6-overlay supervision.
For anyone doing the upgrade: if you're on Unraid with the shared Docker volume setup, you'll need to delete the old hermes_shared_volume (or whatever your template names it) before recreating the container — Docker named volumes don't auto-populate on tag updates, so stale files from the old volume will cause the same crash.
Also note the new env var requirement: HERMES_DASHBOARD_INSECURE=1 is now required for the dashboard to bind, and with v0.15.x you can run the dashboard alongside the gateway in the same container via s6-overlay (set HERMES_DASHBOARD=1) instead of a separate dashboard container.
Thanks everyone for confirming reproductions — and @mmartial for keeping the Unraid template updated.
Activity
szmania commentedon May 29, 2026
I can reproduce this issue as well in latest docker image, running on arm64.
subpanic commentedon May 29, 2026
Also reproduced with latest docker image in Unraid, amd64
kieraneglin commentedon May 29, 2026
+1
Out of curiosity, are we all running Unraid? I wonder if that's the common denominator.
stonebert-png commentedon May 29, 2026
@kieraneglin I am on unraid as well.
szmania commentedon May 29, 2026
I've had to roll back to nousresearch/hermes-agent:v2026.5.16 to get this working. I'm not on unraid, I'm running in a docker container on omv OS, which is Debian based
kieraneglin commentedon May 29, 2026
We're so back. Apparently, the issue can be caused by updating via the UI instead of updating the container's tag.
In any case, here's what solved it for me on Unraid. In short:
HERMES_DASHBOARD_INSECUREwith a value of1in order to access my dashboard locallyszmania commentedon May 29, 2026
I just tested on latest which was released 4 hours ago. Issue still persists.
szmania commentedon May 29, 2026
Never used unraid, but I'm trying to translate your steps into the docker native commands for each step you have here. I've been doing step 1 already, so deleting containers didn't help me. Ill try translating your steps and see how far I get.
mmartial commentedon May 29, 2026
I apologize for the Unraid issue. It is caused by the use of a shared Docker volume.
This is required because Hermes WebUI must have access to the Hermes source code to install its own dependencies.
The only way to do this is to start Hermes Agent with a non-existent shared Docker volume. When the container starts, it will create the volume and copy the contents of the container's path into that volume. This same volume is, in turn, used by the WebUI to work.
The problem is that once the volume is created, updating the Hermes Agent container to a new version does NOT update the shared volume's content. As such, you are NOT using the latest version as expected UNLESS you first delete the shared volume.
I explained the process here mmartial/unraid-templates#8 (comment)
Unfortunately, there is nothing I can do to make this more seamless unless Hermes WebUI removes this dependency. This process is based on the two-container deployment method described on their repo.
I note that the shared volume is not required for the Workspace template, so you can delete the hidden parameter with the shared volume if you are not using the WebUI.
Note that I am considering dropping the WebUI template because of this issue.
stonebert-png commentedon May 29, 2026
Quick root-cause recap since there's been more traction here:
What's missing from the v0.15.0 image:
stage2-hook.sh— referenced in the Dockerfile atRUN chmod +x stage2-hook.shmain-wrapper.sh— missing from the image layercontainer_bootmodule — removed from the source tree entirely (directory exists in v2026.5.16, gone in latest)Cascade: Without
stage2-hook.sh, the entrypoint fails. On Unraid's template (and likely any system using named volumes for the Hermes-Dashboard shared volume), the volume stays empty since the failing container never populates it.Reproducers so far: arm64 (szmania), Unraid amd64 (subpanic, stonebert-png), OMV/Debian (szmania rolled back). The Unraid template maintainer (mmartial) also has users reporting the same crash in mmartial/unraid-templates#8 — currently advising a volume-nuke workaround not knowing it's a broken-image issue.
Workaround for anyone hitting this: pin to
nousresearch/hermes-agent:v2026.5.16until the next release.mmartial commentedon May 29, 2026
@stonebert-png doesn't the 0.15.2 release solve this?
Note that due to recent changes in Hermes, you need to set new variable to
true.Per https://hermes-agent.nousresearch.com/docs/user-guide/docker
you now must add
HERMES_DASHBOARD_INSECURE=1for the system to start up.I updated my template last night to support this feature.
Users of the current template need to manually add the variable for it to work (also described in my small blurb on the issue listed earlier)
FYSA, I followed the steps I described: I updated, deleted my shared volume, restarted and the WebUI is accessible to me.
szmania commentedon May 29, 2026
thanks this worked for me. deleted containers, deleted share to /opt/hermes and then set that env var to 1.
stonebert-png commentedon May 30, 2026
Confirmed — v0.15.2 resolves this. I updated to
v2026.5.29.2today and the container starts cleanly. Bothstage2-hook.shandmain-wrapper.share present in thedocker/directory now, and thecontainer_bootmodule has been properly replaced by s6-overlay supervision.For anyone doing the upgrade: if you're on Unraid with the shared Docker volume setup, you'll need to delete the old
hermes_shared_volume(or whatever your template names it) before recreating the container — Docker named volumes don't auto-populate on tag updates, so stale files from the old volume will cause the same crash.Also note the new env var requirement:
HERMES_DASHBOARD_INSECURE=1is now required for the dashboard to bind, and with v0.15.x you can run the dashboard alongside the gateway in the same container via s6-overlay (setHERMES_DASHBOARD=1) instead of a separate dashboard container.Thanks everyone for confirming reproductions — and @mmartial for keeping the Unraid template updated.