<- all tokdocs
2026.09.16@aiclawbots

Four Community Plugins for Hermes Agent: Cost Guard, Model Override, Goals, and a Claude Code Bridge

AI
Watch on TikTok

View on TikTok ->

The four plugins in this video are real and live in one MIT-licensed GitHub repo, 42-evey/hermes-plugins, but each one is narrower than the video describes. Hermes Agent is Nous Research's open-source, self-improving agent (MIT license, about 246k GitHub stars). The video runs 44 seconds, names four plugins with animated mockups, and closes with a comment-to-get-a-guide call to action.

What Hermes Agent is

Hermes Agent is built by Nous Research and published at github.com/NousResearch/hermes-agent under the MIT license. It runs on a single gateway process that connects to Telegram, Discord, Slack, WhatsApp, Signal, and the CLI. It supports multiple model providers (Nous Portal, OpenRouter, OpenAI, custom endpoints) and already ships a hermes model command for switching models. Plugins live in ~/.hermes/plugins/, register tools through a register(ctx) function, and are disabled by default until you add them to plugins.enabled in config.yaml. There is also a hermes plugins install owner/repo command that pulls straight from GitHub.

Plugin one: cost control

The video shows a spending bar climbing from $5.71 to $13.12 toward "your limit," with the line "you find out before the bill does." The matching plugin is evey-cost-guard (v2.0.0). Its manifest describes "budget enforcement + cost analytics via Langfuse" and exposes three tools: cost_check, cost_set_budget, and cost_analytics. The detail the video skips: it pulls cost numbers from a Langfuse instance, so you need Langfuse running and configured before the plugin reports anything.

Plugin two: model picker

On screen, a "locked into whatever you set at setup" line gets struck through, followed by toggles for "a local model" and "a big frontier model," with "cheap for grunt work" and "expensive only when needed." The closest plugin is evey-delegate-model, described in its manifest as "adds per-task model override to delegate_task (fixes upstream bug)." The code calls LiteLLM's OpenAI-compatible API directly, tries up to four models in a fallback chain, and retries each three times. So it is a per-task model override for delegated subtasks, not a mid-conversation model switcher. Hermes Agent's own hermes model command already covers switching the main model.

Plugin three: goal management

The frames show a goal bar filling from 0% in session 1 to 100% by session 3. The plugin is evey-goals, which stores goals in ~/.hermes/goals.md and exposes one tool, evey_goals, with list, add, complete, and remove actions. Persistence across sessions comes from the plain markdown file. The progress percentage in the video is a mockup; nothing in the plugin code computes a completion percentage.

Plugin four: agent bridge

The video draws a line between "your Hermes" and "your other agents," with a logo that looks like Anthropic's, and says you stop copying context back and forth. The plugin is evey-bridge, and its manifest is specific: "bidirectional bridge to Claude Code (Opus)." It works through a directory of inbox and outbox files plus a channel.jsonl message log. The author also publishes a companion Claude Code plugin, evey-bridge-plugin, on the other side. It bridges Hermes to Claude Code only, not to arbitrary agents.

Claims I could not verify

The video says "the same suite also ships WhatsApp bridge and Discord voice." A GitHub issue on an awesome-hermes-agent list makes the same claim, but the hermes-plugins repo tree as of today has no WhatsApp or Discord voice folder among its 36 plugin directories. Hermes Agent itself supports WhatsApp and Discord natively, so the capability exists, but I could not confirm it comes from this suite. The "free guide" is a comment-gated lead magnet with no public URL.

Key Takeaways

  • All four plugins exist in 42-evey/hermes-plugins (MIT, 459 stars, last pushed July 2026); install by copying folders into ~/.hermes/plugins/ and enabling them in config.yaml.
  • Cost control depends on Langfuse; without it the plugin has no data.
  • The "model picker" is a per-task override for delegated work with a fallback chain, not a live model switcher.
  • Goals are a markdown file with list, add, complete, and remove actions.
  • The agent bridge connects Hermes to Claude Code specifically through a file-based inbox and outbox.
  • WhatsApp and Discord voice bridges are claimed but not present in the repo tree I checked.

Resources

Published September 16, 2026. Writeup generated from a favorited TikTok.