The Hermes Agent issue queue, read live. Labelling is already automated by the team's own sweeper bots; what these numbers track is whether a person has answered.
Reading
Labels on the 100 newest open issues
Deep sample: who answers, and how fast
Issues opened vs closed per week
Time to first team reply
By component (6+ issues in the sample)
| Label | Issues | Team reply | Median close | Still open |
|---|
Live counters above refresh on every visit; this sample needs about 400 authenticated API calls and is regenerated by a script, dated above.
Must-touch: open P0 / P1 with no comment, oldest first
| Issue | Title and labels | Age |
|---|
Enter any issue number from NousResearch/hermes-agent (or paste an issue). The checks run in seconds; the output is a first response to post and a packet for engineering.
First response
Escalation packet for engineering
Portal API tickets usually arrive as a status code and one line of JSON. This tab checks model ids against the live list and maps each error the API returns to its real cause.
Check a model id
Checks the live list, then asks the API about the id without a key (it answers model errors before auth).
Diagnose an error
How the API answers (observed 30 Sep 2026)
| Request | Answer | Support read |
|---|---|---|
| Listed model, no Authorization header | 400 "Unknown model" | A missing key is reported as a bad model name. Expect tickets saying "the model in your list does not work". |
| Listed model, bad key | 401 "invalid, blocked or out of funds" | Three causes in one message; ask which before debugging. |
nousresearch/hermes-4-405b, any auth | 404 "retired" | Model errors come before auth. Hermes models are absent from the public list. |
hermes-4-405b (no org) | 404 "not found ... OpenRouter catalog" | The error names OpenRouter, and list entries use OpenRouter's schema. |
| Malformed JSON body | 422 | Payload problem; auth not reached. |
What I would measure from week one, and how the numbers above map to it.
| Metric | Definition | First goal |
|---|---|---|
| Human first response | Share of new issues with a maintainer or support comment within 24 h (bots excluded). | P0/P1: 100% within 4 business hours. All issues: 80% within 48 h. |
| Evidence completeness | Share of bug reports with version, OS, repro steps and a debug report that is still readable. | Capture debug output into the issue before the 6-hour paste expiry. |
| needs-repro age | Median age of open needs-repro issues. | Under 14 days; auto-close with a reopen path after 30. |
| Duplicate rate | Share of new issues closed as duplicate. | Tracked weekly; top duplicate clusters get a pinned issue or a doc fix. |
| Reopen rate | Share of closed issues reopened within 14 days. | Below 5%. |
| Portal ticket mix | Auth, billing, model availability, rate limits, upstream incidents. | Weekly trend note to engineering with counts and one proposed fix each. |
Escalation rubric
- P0: secret exposure, data loss, security, or a core flow down for most users. Page engineering now.
- P1: core flow broken for one platform or provider with no workaround. Same-day engineering owner.
- P2: broken with a workaround. Document the workaround; schedule the fix.
- P3 and P4: minor or cosmetic. Batch weekly.
An escalation is actionable when it has: exact version, environment, minimal repro, the failing output, evidence that survives (not an expired paste), impact count, and what support already tried.