Problem or Use Case

Hermes Agent adoption is accelerating rapidly, and users often employ it to orchestrate the activity of a multitude of other AI agents. Today, we already have basic analytics in Hermes Dashboard but there is no consolidated visibility across multiple agents: token usage, model costs, reasoning traces, and tool-call patterns across sessions.

TokenTelemetry is open source and offers just that with one command — no signup, no cloud, nothing leaves your machine — local, read-only observability for Claude Code, Claude Code, Codex, Gemini CLI, Cursor, Copilot, Antigravity, OpenCode.

Proposed Solution

Develop of a Hermes plugin that integrates with TokenTelemetry project to expose a unified multi-agent telemetry panel inside the Hermes Dashboard.

TokenTelemetry is:

• designed to unify logs from multiple agents into a single view
• agent‑log–driven, i.e. no SDK or instrumentation
• 100% local operation with self‑hosted dashboard
• written in Python & TypeScript (like Hermes itself)
• open source with MIT license (star their repo)

Additional Context

The TokenTelemetry project is very young (less than a month old) and does NOT yet support Hermes Agent. It is, however, fairly straighfoward to implement by reading the ~/.hermes/state.db following the pattern of their existing parser for OpenCode (also based on a SQLite database); in fact, it's even simpler because Hermes Agent already has token counts pre‑aggregated per session (in the sessions table). See feature request in their repo: (VasiHemanth/tokentelemetry#20)

Raising awareness here to attract contributors interested in building a Hermes plugin and/or a parser to enable support in TokenTelemetry. Happy to pick it up if there is sufficient interest from this community (NB: however I'm not much of a front-end dev, so Hermes & Co would have to do most of the heavy-lifting ;-)

Alternatives Considered

• External dashboards (Langfuse, Helicone, LangSmith) → Require cloud accounts or instrumentation; not aligned with Hermes’ local‑first philosophy.

• Manual log inspection or scripted analytics → Not scalable or user‑friendly. Reinventing the wheel.

• Here, I also found issue feat(observability): unified telemetry + analytics for latency, cost, and completion/failure rates #6642 but the scope doesn't seem to include multiple agents, even if lays the groundwork for it... Nonetheless, that one looks like a much more ambitious effort than my proposal above.

Feature Type

New tool

Scope

Small (single file, < 50 lines)

Contribution

• I could implement this myself and submit a PR (see note above)