docs(readme): add hermes-telemetry to Community section - #56441
Conversation
teknium1
left a comment
There was a problem hiding this comment.
Thanks for surfacing a standalone plugin that uses Hermes' existing plugin path.
Problems
README.md:256overstates the enforcement guarantee. The linked plugin documents that soft status injects a notice; hard status blocks subsequent tool calls and pauses future cron runs; an in-flight model response remains billable (plugin README:1171-1185).
Suggested changes
- Replace “pauses a cron run before it breaches a soft/hard budget cap” with wording that describes the documented post-hard-breach guardrail.
Automated hermes-sweeper review.
| - 🐛 [Issues](https://github.com/NousResearch/hermes-agent/issues) | ||
| - 🔌 [computer-use-linux](https://github.com/avifenesh/computer-use-linux) — Linux desktop-control MCP server for Hermes and other MCP hosts, with AT-SPI accessibility trees, Wayland/X11 input, screenshots, and compositor window targeting. | ||
| - 🔌 [HermesClaw](https://github.com/AaronWong1999/hermesclaw) — Community WeChat bridge: Run Hermes Agent and OpenClaw on the same WeChat account. | ||
| - 🔌 [hermes-telemetry](https://github.com/nujovich/hermes-telemetry) — Local-first cost observability with budget enforcement: hooks into the Hermes runtime to record per-session and per-cron token, cost, and latency in SQLite, and pauses a cron run before it breaches a soft/hard budget cap. Adds `/stats` and `/budget`, with OpenRouter pricing auto-synced. |
There was a problem hiding this comment.
This overstates the linked plugin's documented guarantee: soft status only injects a notice, while a hard breach blocks subsequent tool calls and pauses future cron runs; an in-flight model call still completes and is billed. Please describe that post-breach behavior rather than claiming prevention before a soft/hard breach.
Describe the guardrail as post-hard-breach (blocks further tool calls, pauses future cron runs) instead of a pre-breach cron pause. Matches the plugin's documented Enforcement Levels; the in-flight model response is still billable.
|
Good catch — the previous wording implied a pre-breach cut, which the plugin doesn't do. Fixed in 6258a3d. The entry now describes the documented post-hard-breach guardrail: on a hard breach it blocks further tool calls and pauses future cron runs, and it's explicit that the in-flight model response still completes and is billed. This matches the plugin's Enforcement Levels table. |
What & why
Adds a one-line entry under Community for
hermes-telemetry, an MIT,runtime-integrated plugin (installed into
~/.hermes/plugins/) that hooks intothe model-call pipeline to record per-session and per-cron token, cost, and
latency into local SQLite — and, unlike report-after-the-fact dashboards, can
pause a cron run before it breaches a soft/hard budget cap.
The gap it fills
Observability tools already report what was spent. What multiple users have
asked for is a guardrail that stops spend before it happens, for unattended
(cron/gateway) operation:
the model call, returning a structured error on breach
Evidence the pain is real: #44771 (91M-token runaway loop), #45782
(unexpected $145 in API costs), #45783 (credit spikes on session resume).
hermes-telemetryimplements exactly this enforcement path via the pluginhooks, plus
/statsand/budget, with OpenRouter pricing auto-synced.Relationship to native work
This complements — it does not replace — the first-class telemetry effort in
#51714 and the budget-enforcement hook proposal in #52648. It's an opt-in,
available-today option for users who need a spend guardrail now.
Links
hermes plugins install nujovich/hermes-telemetryHow to test
Markdown-only change — verify the rendered Community section.
Platforms
N/A (docs).