zealous widget
Follow up on my case of last night
https://discord.com/channels/1053877538025386074/1494347780574150876
it is a "confirmed bug" i spoke too fast yesterday it broke again today i added one aegnt and all the other broke
all of them are on the same host
zealous widget
seems to be independent coincidence ... tokens exipred... not sure why they didn't refresh
zealous widget
Ok actually it is a bug working on a fix will keep updated
zealous widget
ok bug fix PR submitted #developers message
robust garnet
Thanks for digging into it and for opening a PR.
The root cause you described does look real. The separate CLI path hermes auth add nous --type oauth was not persisting Nous state the same way as the normal login flow, so things could appear fine until the Nous runtime key needed refresh, and then fail later with a generic 401. That lines up with the “worked, then broke again later” pattern you were seeing.
There is one important issue in the PR as it stands, though. Right now it writes a pool entry with source manual:device_code, while the singleton seed path still imports providers.nous back into the pool as device_code. That means the PR starts creating duplicate Nous pool entries instead of keeping one canonical device-code entry. So the direction looks right, but I would not merge it yet without fixing that source mismatch.
@velvet spirethe diagnosis looks mostly correct, but the current implementation on PR #11858 appears to duplicate Nous credentials in the pool because of the manual:device_code vs device_code split. I think the right next step is to normalize Nous device-code persistence to one canonical source name before merge.
zealous widget
Thanks for digging into it and for opening a PR.
The root cause you described d...
Thanks for the reply
If you need / want me to review anything in the pr or do more tests let me know