Troubleshooting

/stats cron week shows “No cron runs in the last 7 days”

The gateway loaded before the plugin was enabled. Restart it and run a cron job again:

hermes gateway restart

Budget limits are missing or out of date

/budget shows (no limit set), or an old limit, after you edited budget.yaml by hand.

The plugin caches the budget configuration in memory. It reloads the file automatically when the watchdog package is installed. If it is not, use /budget set global daily <amount>, which writes the file and reloads at once, or restart the gateway. Also check telemetry.log: a YAML syntax error disables budgets rather than crashing, and logs a warning.

Cost is $0.00 for every session

The model is not in the pricing table. Look in telemetry.log for a one-time warning such as:

hermes-telemetry: unknown model 'openrouter/some-model' — cost recorded as $0.00

or, when the call carried a provider name:

hermes-telemetry: no price for model 'some-model' under provider 'nous' — cost recorded as $0.00

Add the model to pricing.yaml under the provider’s native model id. Running /setup pricing auto also works, but it overwrites the whole file and discards hand edits, so back it up first. See Pricing. If /stats models marks the row free tier or subscription/free-tier, the zero is intentional.

Est% is above zero for my provider

Your provider returns usage=None for some or all calls, so tokens are estimated. Check /stats providers to see which providers are affected. If Est% is 100% for your main provider, all of its spend is estimated, and under the default warn_only mode hard budget verdicts degrade to soft. See Provider probe.

A hard breach did not stop the session

Budgets block tool calls, not model calls. A session that only generates text never reaches the gate, and the response already in flight when a limit is crossed still completes. Also check whether the verdict was degraded by estimated data (~est in /budget), and whether the plugin is running in observe mode. See Budgets.

The plugin is not loading at all

Check telemetry.log for errors. Common causes:

• pyyaml is missing from the gateway’s Python environment: pip install pyyaml.
• The plugin is not listed in plugins.enabled in config.yaml: hermes plugins enable hermes-telemetry.
• A syntax error in pricing.yaml or budget.yaml.
• A second copy of the plugin, such as a backup directory, sits in ~/.hermes/plugins/. Hermes indexes plugins by the name in their manifest, and when two directories share a name, one silently shadows the other, often the backup over the active copy. Keep backups outside ~/.hermes/plugins/.

hermes-telemetry stats efficiency fails with “invalid choice”

The standalone CLI does not include the efficiency, smells, and raw views. Use the in-chat /stats efficiency and /stats smells, or the dashboard. See Standalone CLI.

The Telemetry tab does not appear in the Hermes dashboard

Reload the page after the dashboard process has started. The dashboard discovers plugins at startup from ~/.hermes/plugins/<name>/dashboard/manifest.json; if you installed the plugin while the dashboard was running, restart the dashboard process. See Installation.

The standalone dashboard is unreachable from another machine

By default it binds to 127.0.0.1 only. Use an SSH tunnel, as described in Dashboard.