2026-06-12

Hermes Roundup: June 12, 2026

hermesroundup

Patrick Suzuki shipped Token Lens, an open-source Hermes Agent plugin that shows where every token goes and surfaces the fix. Built with gstack and Fable over two days during vacation. 150K input tokens, 600K output tokens. The plugin visualizes token consumption across the agent pipeline, letting users identify waste and optimize prompts.

Token Lens: an open-source Hermes Agent plugin that shows where every token goes then hands you the fix: https://t.co/2BJtXnWfXm

Built using gstack and fable high. 150k input tokens. 600k output tokens. Over two days during vacation. pic.twitter.com/yEVUFnUwgH
— Patrick Suzuki (@patrickluvsoj) June 12, 2026

Hermes Desktop Ships

The Hermes Desktop app launched today. No terminal, no configuration commands - an app that auto-delegates tasks to specialized AI agents. One curl command installs it in under three minutes, and the Control Room profile routes work out of the box. EK Cheung posted a walkthrough:

The "I'll automate this later" excuse just died. 💀

Hermes shipped a desktop app. No terminal. No cryptic commands. Just an app that auto-delegates your tasks to specialized AI agents.

a. Install: one curl command, zero code → under 3 mins
b. Control room profile routes work… pic.twitter.com/WbPn0ptGU8
— EK (@ekcheungAI) June 12, 2026

The desktop app addresses the CLI barrier. New users can install and run without learning terminal conventions, and the visual routing panel replaces manual profile configuration.

Community Comparison: CREAO AI vs Hermes

PODARITY published a comparison between CREAO AI's multi-agent system and Hermes Agent for secure agent workflows. The thread breaks down two architectural approaches: CREAO's bundled multi-agent orchestration versus Hermes's modular plugin-and-profile model.

CREAO AI vs Hermes Agent in 2026: Which One I Actually Use for Secure Agents 🛡️

Two very different approaches dominating the agent space right now.

After building and auditing secure agents, here's my direct comparison.

1/
CREAO AI – The multi-agent system (RUN IT ALL)

You… https://t.co/2MYfqiwjdF pic.twitter.com/bhlgVFmhCp
— PODARITY🫳🏽❤️ (@fineboyPOD) June 12, 2026

Self-Evolution: Codex vs Hermes

Maki from Sunwood AI Labs published a Japanese-language analysis comparing Codex's AGENTS.md self-evolution with Hermes Agent's memory and skill system. The finding: the two frameworks approach self-improvement differently. Hermes persists lessons into memory and skill layers, with the curator cleaning stale or duplicate skills between sessions. Codex's AGENTS.md self-evolution operates in a different domain.

CodexのAGENTS. md自己進化が「効いてないように見える」問題を、Hermes Agentの設計と比べて調査した。

結論はシンプルで、自己進化の主戦場が違う。

✅ Hermesは、教訓をmemory / skillへ残す
✅ curatorが古いskillや重複skillを整理する
✅ 次回セッションのsystem promptやskill… https://t.co/y4zcxywtAX pic.twitter.com/iSVm8qha8P
— Maki@Sunwood AI Labs. (@hAru_mAki_ch) June 12, 2026

Mobile Access via Tailscale

Sugumaran Balasubramaniyan submitted his second PR to Hermes Agent this month. The first was a model-task-router skill that went through 13 review rounds before approval. The new contribution is a mobile access guide: use a phone as a Hermes client over Tailscale, extending agent access beyond the desktop.

Second PR to @NousResearch's Hermes Agent this month 🎉

The first was a model-task-router skill (13 review rounds, approved).

This one: a mobile access guide - use your phone as a Hermes client over Tailscale.

Two contributions, two angles. https://t.co/dCmjHPM4Od @Teknium
— Sugumaran Balasubramaniyan (@sugumaran___) June 12, 2026

Other Signals

@zknark is building an observability system for Hermes, OpenClaw, and NanoClaw agents, offering insights into skill changes and agent actions.
@JulianGoldieSEO shared a Hermes Agent workflow that turns ideas into apps using multi-model brainstorming (Claude, Hermes, Gemini).
@FarAICoder posted about agent autonomy guardrails: a memory layer spotted a six-week trend after four hours of initial setup.
@thevoyagevibe surfaced a full architecture breakdown of Hermes Agent by DWF Ventures.
@CsWiz1 shared a five-day test of Hermes with Obsidian, GitHub, Telegram, and Grok - day one was strong, the following four days hit session resumption and OAuth friction.

The day's signals point to three themes: tooling maturing beyond the CLI (desktop app, mobile access), deeper introspection (token visualization, observability), and comparative analysis between agent frameworks. The desktop app launch and Token Lens plugin represent concrete artifacts from the community - shipping code, not just commentary.