Home / Companies / Axiom / Blog / Post Details
Content Deep Dive

Hermes agent plugin: Where did all my tokens go?

Blog post from Axiom

Post Details
Company
Date Published
Author
-
Word Count
2,059
Company Posts That Month
4
Language
English
Hacker News Points
-
Post removed?
No
Summary

Axiom’s hermes-metrics is an Apache-2.0/MIT-licensed observability plugin for the open-source Hermes agent that converts Hermes observer hooks into OpenTelemetry traces, metrics, and limited failure logs, allowing users to track sessions, model calls, tools, approvals, subagents, token usage, latency, and estimated spending. To avoid slowing agent execution, the plugin places hook events into a bounded asynchronous queue and drops excess events rather than blocking, while exporting its own health counters and queue-depth metrics so data loss or reporting failures can trigger alerts. It calculates costs from provider pricing routes while avoiding double-counting reasoning tokens and treating subscription-covered usage as zero cost, exports pricing rates to reveal provider changes, and seeds tool metrics from installed tools so unused tools remain visible. Setup can provision Axiom datasets, restricted write tokens, dashboards, and seven monitors for telemetry drops, errors, token consumption, spending, and subagent concurrency. Privacy controls use cumulative redaction levels that default to metadata only, never transmit prompts or conversation histories, and mask credential-like field names, though secrets embedded in arbitrary values are not detected. The plugin is designed to fail without disrupting the agent, handling missing dependencies, export failures, and schema mismatches through warnings and diagnostic events.

Trends Found in this Post
Trend Post Mentions Total Month Mentions Posts Companies MoM
Subagents 4 15 10 7 -95%
OpenTelemetry 2 125 18 15 -83%
Secrets Management 1 451 99 43 -80%
Use This Data

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.