ingest: Alex Finn 100+ Hours Hermes Agent Lessons Learned (2026-07-09)

- raw/xpost: 9 lessons, model recommendations (Opus→GPT-5.6), Tailscale, Reverse Prompting
- wiki/tools/hermes-desktop.md: new section with model comparison table, security tips, workflow innovations
- wiki/index.md: 70. Update, Hermes Desktop entry expanded
- wiki/log.md: ingest entry + heartbeat-curation outage note

raw/xpost/2026-07-10_alex-finn-hermes-agent-100-hours.md

@@ -0,0 +1,52 @@
+---
+title: "Alex Finn: 100+ Hours with Hermes Agent — 9 Lessons Learned"
+source: https://x.com/AlexFinn/status/2075055593635844306
+date: 2026-07-09
+author: Alex Finn (CEO, Henry Intelligent Machines)
+type: xpost
+language: en
+tags: [hermes, agent-os, lessons-learned, model-comparison, tailscale, reverse-prompting, gpt-5.6, claude-opus, glm]
+---
+
+# Alex Finn: 100+ Hours with Hermes Agent — 9 Lessons Learned
+
+**Source:** X post + video by Alex Finn, July 9, 2026
+**Follow-up:** July 10, 2026 — Alex switched his main Hermes agent to GPT-5.6 for the first time
+
+## Context
+
+Alex Finn, CEO of Henry Intelligent Machines, posted a detailed video covering lessons from 100+ hours using the Hermes Agent (Nous Research). He calls Hermes "the most powerful AI agent on Earth right now" and states that nobody has used it more.
+
+## Key Lessons
+
+### Model Recommendations
+1. **Claude Opus** — Best-in-class for complex tasks. Used heavily from day one (including via API at high cost).
+2. **ChatGPT / GPT-5.6 (codex/gpt-5.6-sol)** — Now competitive or better. On July 10 Alex publicly switched his main Hermes agent to it, citing strong performance at a fraction of the previous cost. Use your existing ChatGPT subscription directly in Hermes (avoid OpenRouter version if subscribed).
+3. **GLM** — Recommended for maximum cost savings on high-volume or less critical work.
+
+**Pro tip:** Create separate agent profiles in Hermes (one per model/API) so you can easily test, compare, and switch between them.
+
+### Security & Connectivity
+4. **Tailscale** is the top recommendation for secure, zero-trust connectivity across devices. Essential for agents that need to SSH, access remote machines, or manage fleets.
+5. **Security concerns about separate accounts are "largely overblown"** — Alex argues against the common fear of giving agents their own accounts.
+6. **2+ agents as failover** — redundancy matters.
+
+### Workflow & Setup
+7. **Reverse Prompting** — 5-minute morning interview where the agent interviews YOU about your day's priorities. This flips the traditional prompt dynamic.
+8. **Separate agent accounts are a mistake** — Alex argues for using your own accounts with the agent, not creating separate ones.
+9. **Multi-device via Tailscale** — enables Hermes to act like a true "AI employee" that can move between your local machines securely.
+
+## Additional Tips from Community
+- One index file per folder dramatically speeds up agent performance
+- /learn workflows work best when isolated in a dedicated "classroom" folder with a "textbook" file of reusable paths/links
+- Build structured "vaults" with index files per major folder so the agent navigates quickly instead of wandering
+- Key interfaces to master: Dashboard, Mission Control, and Troubleshooting
+
+## Model Cost Comparison (Alex's Experience)
+- Claude Opus via API: high cost, best quality
+- GPT-5.6 via ChatGPT subscription: strong performance, fraction of cost
+- GLM: budget option for high-volume work
+
+## NotebookLM Briefing
+- Notebook: https://notebooklm.google.com/notebook/46208dab-e627-40f6-90e9-6e06a84e7340
+- German audio briefing (nb3 preset), ~23 min

wiki/index.md

@@ -2,7 +2,7 @@
 
 *Auto-generated: 2026-07-07*
 
- *Letzte Aktualisierung: 2026-07-09 (69. Update — ChatGPT Live Mode (GPT-Live-1) ingestiert. Raw: `raw/youtube/2026-07-09-chatgpt-live-modus-calvin-hollywood.md`. Wiki-Update: `tools/chatgpt-live-mode.md` neu erstellt — GPT-Live-1 Echtzeit-Sprachmodus, Calvin Hollywood Review. `people/calvin-hollywood.md` neu erstellt. Cross-Ref in `tools/openai-gpt.md` aktualisiert.)*
+ *Letzte Aktualisierung: 2026-07-10 (70. Update — Alex Finn: 100+ Hours Hermes Agent Lessons Learned ingestiert. Raw: `raw/xpost/2026-07-10_alex-finn-hermes-agent-100-hours.md`. Wiki-Update: `tools/hermes-desktop.md` — neuer Abschnitt mit 9 Lessons, Modell-Empfehlungen, Reverse Prompting, Tailscale.)*
 
 ## Architecture
 
@@ -22,7 +22,7 @@
 | Seite | Beschreibung | Quellen |
 |-------|-------------|---------|
 | [OpenClaw](tools/openclaw.md) | Plattform-Referenz: v2026.6.8 stable (16.06.2026), Architektur-Übersicht, plattformrelevante Release-Changes, Update-Plan. **Mobile Apps (29.06.2026):** Native iOS (iPhone/iPad/Apple Watch) + Android Apps, Gateway-Pairing, Voice, Approvals, Device Capabilities. ClawHub-Security-Kontext (5 malicious skills, 800+ gesamt, 84.2% NL-injection) | other/2026-06-16_openclaw-releases-v2026.6.8.md + blog/2026-06-30_perplexity-openclaw-mobile-apps.md |
-| [Hermes Desktop](tools/hermes-desktop.md) | Offizielle native GUI für Hermes Agent. Electron/React/Python. Update v0.17.0 ("Reach") bringt native iMessage-Integration via Photon, asynchrone Subagents mit Watch-Windows, Image-to-Image Bearbeitung, Automation Blueprints und Telegram Rich Text Bot API 10.1. **Mixture of Agents (MoA)** (28.06.2026): Merge any N models into one virtual model (Reference + Aggregator), +8% über Opus 4.8 solo. **MoA 2.0 + Hermes Agent OS** (03.07.2026): GoldyBench-Validierung (42 Builds, top vor Opus 4.8 solo), Agent OS GUI mit Mixture/Chat/Talk/Jarvis/Oracle/Studio Tabs, "Don't chase the model, build the system". OpenClaw-Migrationstool `hermes claw migrate`. | youtube/2026-06-22_jonas-keil-hermes-desktop.md + youtube/2026-06-24_alex-finn-hermes-agent-v0170-reach.md + youtube/2026-06-24_peter-yang-hermes-full-course.md + xpost/2026-06-28-hermes-moa-vaibhavsisinty.md + youtube/2026-07-03_hermes-mixture-of-agents-2-0-agent-os.md |
+| [Hermes Desktop](tools/hermes-desktop.md) | Offizielle native GUI für Hermes Agent. Electron/React/Python. Update v0.17.0 ("Reach") bringt native iMessage-Integration via Photon, asynchrone Subagents mit Watch-Windows, Image-to-Image Bearbeitung, Automation Blueprints und Telegram Rich Text Bot API 10.1. **Mixture of Agents (MoA)** (28.06.2026): Merge any N models into one virtual model (Reference + Aggregator), +8% über Opus 4.8 solo. **MoA 2.0 + Hermes Agent OS** (03.07.2026): GoldyBench-Validierung (42 Builds, top vor Opus 4.8 solo), Agent OS GUI mit Mixture/Chat/Talk/Jarvis/Oracle/Studio Tabs, "Don't chase the model, build the system". **Alex Finn: 100+ Hours Lessons Learned** (09.07.2026): 9 praktische Lektionen, Modell-Empfehlungen (Opus → GPT-5.6 Migration), Reverse Prompting, Tailscale. OpenClaw-Migrationstool `hermes claw migrate`. | youtube/2026-06-22_jonas-keil-hermes-desktop.md + youtube/2026-06-24_alex-finn-hermes-agent-v0170-reach.md + youtube/2026-06-24_peter-yang-hermes-full-course.md + xpost/2026-06-28-hermes-moa-vaibhavsisinty.md + youtube/2026-07-03_hermes-mixture-of-agents-2-0-agent-os.md + xpost/2026-07-10_alex-finn-hermes-agent-100-hours.md |
 | [Anthropic Claude](tools/anthropic-claude.md) | Claude Opus 4.8, Mythos 5, Fable 5 (Redeployed 01.07.), Business-Kennzahlen, Pentagon-Streit, Amazon-Jailbreak, Branchen-Fallout, Cybersecurity-Classifier | 7 |
 | [OpenAI GPT Models](tools/openai-gpt.md) | GPT-5.5, GPT-5.5 Instant, GPT-Live-1 Voice Mode, Pricing | 3 |
 | [ChatGPT Live Mode (GPT-Live-1)](tools/chatgpt-live-mode.md) | OpenAIs Echtzeit-Sprachschnittstelle für ChatGPT Plus. Direkte Audio-zu-Audio-Verarbeitung, emotionale Bandbreite, natürliche Konversation | youtube/2026-07-09-chatgpt-live-modus-calvin-hollywood.md |

wiki/log.md

@@ -2,6 +2,17 @@
 
 *Append-only changelog. Start: 2026-06-05*
 
+## [2026-07-10] Ingest | Alex Finn: 100+ Hours Hermes Agent — 9 Lessons Learned
+**Type:** ingest | **Scope:** raw/xpost, wiki/tools (update), wiki/index, wiki/log
+**Source:** X post + video by Alex Finn (CEO Henry Intelligent Machines) — https://x.com/AlexFinn/status/2075055593635844306 (2026-07-09)
+**Trigger:** Kai (@PWeber) posted in OME Topic 13 "News & Infos"
+**Actions:**
+- raw: `raw/xpost/2026-07-10_alex-finn-hermes-agent-100-hours.md` (created — 3.1 KB; Frontmatter [type: xpost, source_url, date: 2026-07-09, author: Alex Finn, tags: hermes, agent-os, lessons-learned, model-comparison, tailscale, reverse-prompting, gpt-5.6, claude-opus, glm]. Content: 9 lessons, model recommendations [Opus→GPT-5.6 migration], Tailscale, Reverse Prompting, NotebookLM briefing link)
+- wiki (UPDATE): `tools/hermes-desktop.md` — new section "Alex Finn: 100+ Hours Lessons Learned" with model comparison table, security/connectivity tips, workflow innovations (Reverse Prompting, Classroom folders)
+- wiki: `index.md` (updated — Header auf "70. Update", Hermes Desktop entry expanded with Alex Finn lessons)
+- log: this entry
+**Note:** Heartbeat-Kuratierung war vom 08.07.-10.07. deaktiviert (Heartbeat-Intervall auf 2d gesetzt). Fix: zurück auf 30m, Gateway restart.
+
 ## [2026-07-07] Ingest | heise+ — Prompt-Caching: Token sparen durch KV-Cache-Wiederverwendung
 **Type:** ingest | **Scope:** raw/blog, wiki/concepts (new), wiki/tools (update), wiki/index, wiki/log
 **Source:** heise+ Ratgeber — https://www.heise.de/ratgeber/KI-Kosten-reduzieren-Wie-man-mit-Prompt-Caching-messbar-Token-sparen-kann-11335003.html (heise online, 2026-07-07, paywall)

wiki/tools/hermes-desktop.md

@@ -1,8 +1,8 @@
 ---
 created: 2026-06-22
-updated: 2026-07-03
+updated: 2026-07-10
-sources: [youtube/2026-06-22_jonas-keil-hermes-desktop.md, youtube/2026-06-24_alex-finn-hermes-agent-v0170-reach.md, youtube/2026-06-24_peter-yang-hermes-full-course.md, xpost/2026-06-28-hermes-moa-vaibhavsisinty.md, youtube/2026-06-28_hermes-setup-2min-matthew-berman.md, youtube/2026-07-03_hermes-mixture-of-agents-2-0-agent-os.md]
+sources: [youtube/2026-06-22_jonas-keil-hermes-desktop.md, youtube/2026-06-24_alex-finn-hermes-agent-v0170-reach.md, youtube/2026-06-24_peter-yang-hermes-full-course.md, xpost/2026-06-28-hermes-moa-vaibhavsisinty.md, youtube/2026-06-28_hermes-setup-2min-matthew-berman.md, youtube/2026-07-03_hermes-mixture-of-agents-2-0-agent-os.md, xpost/2026-07-10_alex-finn-hermes-agent-100-hours.md]
-tags: [tools, hermes-agent, hermes-desktop, nous-research, desktop-app, electron, react, gui, openclaw-migration, self-improving, v0.17.0, reach, mixture-of-agents, moa, moa-2, model-fusion, ensemble, quickstart, hostinger, vps-hosting, matthew-berman, agent-os, goldybench, system-vs-model, business-automation]
+tags: [tools, hermes-agent, hermes-desktop, nous-research, desktop-app, electron, react, gui, openclaw-migration, self-improving, v0.17.0, reach, mixture-of-agents, moa, moa-2, model-fusion, ensemble, quickstart, hostinger, vps-hosting, matthew-berman, agent-os, goldybench, system-vs-model, business-automation, lessons-learned, reverse-prompting, tailscale, gpt-5.6, model-comparison]
 ---
 
 # Hermes Desktop — Native GUI für Hermes Agent
@@ -356,6 +356,40 @@ hermes desktop --ignore-existing  # CLI auf PATH ignorieren
 
 ---
 
+## Alex Finn: 100+ Hours Lessons Learned (09. Juli 2026)
+
+> **Quelle:** X-Post + Video von [Alex Finn](https://x.com/AlexFinn/status/2075055593635844306) (CEO Henry Intelligent Machines), 09.07.2026 → [Raw-Datei](../../raw/xpost/2026-07-10_alex-finn-hermes-agent-100-hours.md)
+> **Follow-up (10.07.):** Alex switched main Hermes agent to GPT-5.6 for the first time
+
+Nach 100+ Stunden mit Hermes Agent teilt Alex Finn 9 praktische Lektionen. Er nennt Hermes "the most powerful AI agent on Earth right now."
+
+### Modell-Empfehlungen
+
+| Modell | Rolle | Kosten |
+|--------|-------|--------|
+| **Claude Opus** | Best-in-class für komplexe Tasks | Hoch (API) |
+| **GPT-5.6 (codex/gpt-5.6-sol)** | Jetzt competitive/besser, Alex switched am 10.07. | Bruchteil von Opus (ChatGPT-Subscription) |
+| **GLM** | Budget-Option für High-Volume/Low-Criticality | Minimal |
+
+**Pro-Tipp:** Separate Agent-Profile pro Modell/API anlegen zum einfachen Testen und Vergleichen.
+
+### Security & Connectivity
+- **Tailscale** als Top-Empfehlung für Zero-Trust-Konnektivität über Geräte hinweg
+- Security-Bedenken zu separaten Accounts: "largely overblown"
+- **2+ Agents als Failover**
+- Separate Agent-Accounts sind ein Fehler — eigene Accounts mit dem Agent nutzen
+
+### Workflow-Innovationen
+- **Reverse Prompting:** 5-Minuten-Morgeninterview — der Agent interviewt DICH zu den Tagesprioritäten
+- **Multi-Device via Tailscale:** Agent als "AI Employee" der zwischen lokalen Maschinen wechselt
+- **Strukturierte Vaults:** Ein Index-File pro Ordner beschleunigt Agent-Navigation drastisch
+- **Classroom-Ordner:** /learn Workflows isolieren mit "textbook"-File für wiederverwendbare Pfade/Links
+
+### Einordnung
+Alex Finns Modell-Empfehlung (Opus → GPT-5.6 Migration) bestätigt den Trend: Frontier-Modelle werden commoditized, die System-Architektur (Hermes als Agent-OS) ist der bleibende Wert. Das Reverse-Prompting-Pattern ist eine interessante Umkehrung des üblichen Prompt-Workflows und könnte für Hectors Morning-Briefing adaptiert werden.
+
+---
+
 ## Relevanz für RamaDama / OpenClaw
 
 - **Konkurrenz-Analyse:** Hermes v0.17.0 ("Reach") weitet seine Präsenz drastisch aus. Vor allem die iMessage-Integration über Photon nimmt viel Reibung aus dem iMessage-Setup.