I build small, local plugins for Hermes Agent, the open-source agent from Nous Research. Each one hooks into a single point in the agent, keeps a short local ledger, and answers one question with one slash command: which models are slow, which calls fail, how long the agent waits on you, which files it touched. No network, no subprocesses, no credentials; nothing leaves your machine.
All ten live in one repo, olric3015/hermes-plugins, and install by name from the Hermes catalog:
hermes plugins install <name> --enable
| Plugin | Command | What it tells you |
|---|---|---|
| aux-ledger | /aux |
Every auxiliary LLM call (titles, compression, vision, approvals) by task, model and provider |
| stream-speed | /speed |
Time to first text and characters per second for each model |
| error-ledger | /errors |
Failed provider requests by model, provider and failure reason |
| approval-ledger | /approval-log |
Approval prompts by rule, with how often each was approved, denied or left unanswered |
| command-ledger | /command-log |
Which slash commands you actually use, on which surfaces |
| wait-ledger | /waits |
How long the agent sat waiting on you: approvals, questions, password prompts |
| stop-ledger | /stop-log |
Turns cut short by /stop or /new, per platform |
| voice-ledger | /voice-log |
Voice recordings sent to speech-to-text, by provider, model and language |
| interim-ledger | /interim-log |
Progress messages the agent sends mid-turn, even where the display hides them |
| edit-ledger | /edit-log |
Files the agent changed per turn, by file type |
Every plugin is tested on Linux, macOS and Windows, and end to end against both the oldest supported Hermes release and current main in CI.
![]() |
![]() |
![]() |
![]() |




