Compiled July 25, 2026 · Vera / Hermes Agent
Julian Goldie (411K+ YouTube subscribers, CEO of Goldie Agency) released multiple videos in June–July 2026 describing his "Agent OS" — a local-first coordination layer for AI agents built on top of Hermes Agent, Claude Code, and OpenClaw. Below are the skills and patterns extracted from those videos.
| Video Title | URL | Duration |
|---|---|---|
| Claude + Hermes Agent: NEW Agent OS is INSANE! | youtube.com/watch?v=nhDYrVQxBg4 | ~22 min |
| NEW Hermes Agent OS is INSANE (FREE!) | youtube.com/watch?v=o2FnTLt9mrc | ~18 min |
| How to Build Your Own Agent OS (7-Layer Blueprint) | youtube.com/watch?v=o6uDzILqejY | ~26 min |
| How to Build Your Own Agent OS (32 min) | youtube.com/watch?v=bsYbA9ZyYtI | 32 min |
| How to Build Your Own Agent OS (FREE) | youtube.com/watch?v=HrUyGyzSZog | ~15 min |
| Rank #1 on Google with Claude + Hermes Agent OS | youtube.com/watch?v=sIHchwmbKTo | ~20 min |
Source: YT: o6uDzILqejY (7-Layer Blueprint)
Pattern: Assign clear roles to agents — one plans (CEO/Claude), one routes tasks (COO/OpenClaw), one executes (Worker/Hermes). Never have agents with overlapping responsibilities.
Solid Circle Application: Vera = CEO, vCode = COO for builds, sub-agents = workers. Already partially adopted.
Source: YT: bsYbA9ZyYtI, Blog: aiprofitboardroom.com
Pattern: Before every response, agents pull context from a shared Obsidian vault. After every task, agents write structured notes back. 1,200+ memories accumulated by Day 30.
Solid Circle Application: Adopted July 22 — "Shared Agent Memory.md" in Obsidian vault. vCode, iLegal, iKnow all write here.
Source: YT: o6uDzILqejY (Layer 7)
Pattern: Layer 7 of the blueprint — outputs don't disappear. They feed back to the memory layer. Day 100 is smarter than Day 1. Without this loop, the system resets daily.
Source: YT: HrUyGyzSZog
Pattern: Hermes Agent (MIT), Claude Code (free tier), Obsidian (free), Open Router (free tier). Run everything at $0 until the free tier genuinely can't keep up.
Source: YT: nhDYrVQxBg4
Pattern: iOS doesn't make calls or write messages — it connects apps into one coherent system. An agent OS does the same: shared memory, goal tracking, status indicators, per-agent control rooms.
Source: YT: o2FnTLt9mrc
Pattern: A single AI chat is like owning a hammer — you swing it every time. An agent OS is a construction company — crews, supervisors, plans, scheduling, process that moves without you.
Source: YT: bsYbA9ZyYtI
Pattern: Describe the dashboard to Claude. Paste API docs. Claude scaffolds Next.js + Tailwind in ~1 hour. Don't manually code — use AI to build AI OS.
Source: YT: o6uDzILqejY
Pattern: Use Open Router as a single model gateway. Swap models without rebuilding anything. Keep architecture stable, swap engines as new models release.
Source: YT: bsYbA9ZyYtI, sIHchwmbKTo
Pattern: OMI records screen + mic throughout the day. Every 30-40 min, creates a memory entry. Exports to Obsidian overnight. Agents pull personal context before every response.
Source: YT: nhDYrVQxBg4
Pattern: Left rail shows agents with live status (online/idle/working). Right rail shows goals with progress bars, daily journal, memory search. Center: active chat.
Source: YT: o6uDzILqejY
Pattern: Track sessions run, tool calls fired, tokens consumed, models used, activity patterns, peak working hours. Built into the dashboard — not a separate analytics tool.
Source: YT: sIHchwmbKTo
Pattern: Claude handles keyword research + content strategy. Hermes handles research + fact-checking. Production surface automates publishing across multiple domains.
Dispatched July 22, 2026 — 2 research sub-agents. Results saved to ~/julian-goldie-agent-os-research.md and ~/agent-os-research.md.
| Layer | Name | Component | Role |
|---|---|---|---|
| 1 | Foundation | Hardware (local machine) | Everything runs locally — no cloud dependency |
| 2 | Memory | Obsidian Vault + OMI | Markdown vault (100+ CLI commands), screen/mic recording → vault |
| 3 | Brain | Open Router | Model-agnostic routing layer. Swap engines without rebuilding |
| 4 | Agents | Claude / OpenClaw / Hermes | CEO (planning), COO (execution), Worker (research/automation) |
| 5 | Command Center | Next.js Dashboard | Single sidebar with all agents, mission control, analytics |
| 6 | Production Surfaces | Goals, SEO, Studio, Notebook | Output surfaces for different workflow types |
| 7 | Compounding Loop | Output → Memory → Input | Every output flows back to memory. Day 100 > Day 1 |
| Layer | Name | Tool | Role |
|---|---|---|---|
| 1 | Intelligence | Claude / Claude Code | CEO — planning, reasoning, decision-making |
| 2 | Execution | OpenClaw | COO — task routing, session management, multi-agent coordination |
| 3 | Research | Hermes Agent | Workhorse — tool calls, Kanban, skills, plugins, scheduled workflows |
| 4 | Self | Obsidian Vault + OMI | Memory/identity — goals, journals, notes, personal context |
| Component | Location | Function |
|---|---|---|
| Agent List | Left Rail | Live status indicators (online/idle/working) |
| Active Chat | Center Panel | Real-time streaming, auto-saved history |
| Goals + Progress | Right Rail | Weekly/monthly/quarterly with progress bars |
| Daily Journal | Right Rail | Voice or text input, feeds memory layer |
| Memory Search | Right Rail | Full-text search across thousands of notes |
| API Keys | Per-Agent Room | Provider configuration, model routing |
| Session History | Per-Agent Room | Full chat history with search |
| Skills/Plugins | Per-Agent Room | Management interface for agent capabilities |
| Kanban Board | Per-Agent Room | Task tracking, status columns |
| Analytics Panel | Per-Agent Room | Sessions, tool calls, tokens, models, patterns |
| Goals Mode | Production Surface | Project and goal tracking |
| SEO Section | Production Surface | Content production workflows |
| Studio | Production Surface | Image/video generation |
| Notebook | Production Surface | Research and notes |
| Workspace | Production Surface | All created artifacts, organized by type |
These 30+ skills were built through real work — TON contracts, Android debugging, backend pricing, agent design. Not downloaded. Crafted session by session.
| Skill | Category | What It Does | Source |
|---|---|---|---|
solid-circle | productivity | Project umbrella — token specs, ecosystem architecture, builder workflows | Us |
ime-android | ime | Android app dev — Compose, voice, theming, common pitfalls | Us |
ime-backend | productivity | 3-tier FastAPI server, SQLite ledger, DeepSeek API, A100 GPUs | Us |
ime-frontend | productivity | Mini App, Android Compose, cross-platform UI consistency | Us |
ime-soul | autonomous | Growth template for every iMe agent instance | Us |
ime-mindset-switcher | autonomous | Detects user intent, switches iMe personality mode | Us |
ime-content-crafter | autonomous | Crafts responses in selected mindset | Us |
| Skill | Category | What It Does |
|---|---|---|
agent-builder | autonomous | Spawn dedicated builder agents with isolated profiles + workspaces |
project-builders | autonomous | Isolated autonomous builders for multi-session projects |
multi-profile-builder | autonomous | Spin up dedicated Hermes profiles for isolated builds |
hermes-builder-profiles | autonomous | Create autonomous build worker profiles |
vcode | autonomous | SC internal builder — TON contracts, Android, infrastructure |
vcode-ui-ux | autonomous | vCode UI/UX design training — principles for beautiful interfaces |
circular-build | autonomous | Structured dev methodology — Discovery→Architecture→Build→Review |
agent-safety-protocol | devops | Constitutional safety rule — never delete user files |
progress-report | productivity | Structured periodic progress reports across all projects |
| Skill | Role |
|---|---|
itrade | Day trading agent — stock & options, teach users |
itrade-chart-analyst | Sub-agent: raw chart analysis (S/R, MAs, RSI, MACD) |
itrade-options-strategist | Sub-agent: options strategy (covered calls, CSPs, spreads) |
icode | Full-stack developer for small businesses |
icode-backend-builder | Sub-agent: FastAPI/Python or Node.js APIs |
icode-frontend-builder | Sub-agent: React/Next.js frontends |
iknow | Research agent — market, grants, competitive, legal |
iknow-grant-searcher | Sub-agent: government + private grant databases |
iknow-market-researcher | Sub-agent: market demand, competitive landscape |
ilegal | Georgia criminal defense lawyer |
ilegal-brief-writer-jr | Sub-agent: appellate briefs + trial motions |
ilegal-daniels-jr | Sub-agent: GA trial procedure expert |
ilegal-motion-drafter | Sub-agent: motions to suppress, habeas petitions |
ilegal-ocr-jr | Sub-agent: photos of documents → searchable text |
ilegal-transcript-analyzer | Sub-agent: 500+ page transcript analysis |
icorporate | Business entity formation + corporate governance |
| Skill | What It Does |
|---|---|
ton-development | FunC smart contracts — compile, audit, deploy (testnet + mainnet) |
ton-connect-wallet-urls | TON Connect 2.0 wallet URL formats for native Android schemes |
crypto-trader | Bitunix futures analysis, live position monitoring |
crypto-trading-analysis | Exchange API integration, trade data pipelines, behavioral analysis |
crypto-trading-tools | Exchange API patterns, local trade-history databases |
vtrade | SC internal crypto trading — chart analysis, pattern recognition |
| Skill | What It Does |
|---|---|
local-gpu-setup | Windows CUDA model installation for RTX 3070 Ti 8GB |
relaunch-a100 | Relaunch Vast.ai A100, Ollama, Qwen 3.6, reconnect backend |
windows-security-hardening | Firewall rules, registry hardening, service disablement |
windows-system-diagnostics | Hardware, OS, storage specs on Windows hosts |
hermes-update-pipeline | Update Hermes core, restart all profiles |
hermes-windows-recovery | Recover from 500 errors, sniffio, jiter corruption |
hermes-gateway-ops | Restart, diagnose, recover Telegram/Discord connections on Windows |
vercel-deployment | Deploy, configure, debug Vercel projects |
obsidian-bridge | Read/write to Solid Circle Obsidian vault for cross-agent sharing |
vehicle-sitrep | Military-style hardware diagnostic for the physical host PC |
120+ skills installed from GitHub taps. Listed by category. These are third-party utilities — not Solid Circle IP.
| Skill | Source |
|---|---|
claude-design | One-off HTML artifacts (landing pages, decks, prototypes) |
popular-web-designs | 54 real design systems (Stripe, Linear, Vercel) as HTML/CSS |
excalidraw | Hand-drawn diagram JSON |
pixel-art | Pixel art with era palettes (NES, Game Boy, PICO-8) |
ascii-art | pyfiglet, cowsay, boxes, image-to-ascii |
ascii-video | Convert video/audio to colored ASCII MP4/GIF |
architecture-diagram | Dark-themed SVG architecture/cloud diagrams |
baoyu-article-illustrator | Article illustrations (宝玉/JimLiu) |
baoyu-comic | Knowledge comics (宝玉/JimLiu) |
baoyu-infographic | Infographics: 21 layouts × 21 styles (宝玉/JimLiu) |
p5js | p5.js generative art sketches |
manim-video | 3Blue1Brown math/algo animations |
songwriting-and-ai-music | Songwriting craft + Suno AI music prompts |
sketch | Throwaway HTML mockups, 2-3 design variants |
design-md | Google's DESIGN.md token spec files |
| Skill | Source |
|---|---|
doubt-driven-development | addyosmani/agent-skills — adversarial fresh-context review |
code-review-and-quality | Multi-axis review with quality gates |
test-driven-development | RED-GREEN-REFACTOR cycle |
source-driven-development | Ground decisions in official documentation |
systematic-debugging | 4-phase root cause debugging |
incremental-implementation | Thin vertical slices — implement, test, verify, commit |
subagent-driven-development | Execute via delegate_task subagents |
simplify-code | Parallel 3-agent cleanup of recent changes |
codebase-improvement-audit | Polish/quality audit of working codebases |
codebase-inspection | pygount: LOC, languages, ratios |
performance-optimization | Measure-first optimization across stack |
security-and-hardening | OWASP Top 10 vulnerabilities |
plan | Write actionable markdown plan, no execution |
spike | Throwaway experiments to validate ideas |
hermes-agent-skill-authoring | Author SKILL.md with frontmatter + validator |
| Skill | Source |
|---|---|
llama-cpp | Local GGUF inference, HuggingFace model discovery |
huggingface-hub | hf CLI: search/download/upload models |
dspy | Declarative LM programs, auto-optimize prompts |
segment-anything-model | Zero-shot image segmentation |
comfyui | Generate images/video/audio with ComfyUI |
weights-and-biases | Log ML experiments, sweeps, dashboard |
computer-use | Drive user desktop (Cocoon AI) |
hermes-agent | Configure, extend, contribute to Hermes Agent (Teknium) |
| Skill | What It Does |
|---|---|
github-auth | HTTPS tokens, SSH keys, gh CLI login |
github-pr-workflow | Branch → commit → open PR → CI → merge |
github-code-review | Review PRs with gh or REST API |
github-repo-management | Clone/create/fork repos, remotes, releases |
github-issues | Create, triage, label, assign issues |
claude-code | Delegate coding to Claude Code CLI |
codex | Delegate coding to OpenAI Codex CLI |
opencode | Delegate coding to OpenCode CLI |
kanban-orchestrator | Decomposition playbook for Kanban workers |
webhook-subscriptions | Event-driven agent runs |
| Skill | What It Does |
|---|---|
obsidian | Read, search, create, edit Obsidian notes |
notion | Notion API + ntn CLI |
airtable | Airtable REST API via curl |
google-workspace | Gmail, Calendar, Drive, Docs, Sheets |
linear | Manage issues, projects, teams via GraphQL |
arxiv | Search arXiv papers |
polymarket | Query prediction markets |
blogwatcher | Monitor blogs and RSS feeds |
powerpoint | Create/read/edit .pptx decks |
nano-pdf | Edit PDF text via NLP prompts |
| Skill | What It Does |
|---|---|
youtube-content | Transcripts to summaries, threads, blogs |
spotify | Play, search, queue, manage playlists |
heartmula | Suno-like song generation from lyrics |
songsee | Audio spectrograms and features |
gif-search | Search/download GIFs from Tenor |
maps | Geocode, POIs, routes via OpenStreetMap |
pokemon-player | Play Pokemon via headless emulator |
petdex | Animated petdex mascots for Hermes |
| Category | Count | Examples |
|---|---|---|
| 1 | himalaya — CLI IMAP/SMTP | |
| Smart Home | 1 | openhue — Philips Hue lights |
| Red Teaming | 1 | godmode — LLM jailbreak techniques |
| MCP | 1 | native-mcp — MCP client (stdio/HTTP) |
| Data Science | 1 | jupyter-live-kernel |
| AI Agents | 5 | hermes-s6-container-supervision, kanban-codex-lane, profile-build-agents, create-hermes, agent-platform-scaling |
| Ops | 3 | fleet-command-center, new-user-interview, debugging-hermes-tui-commands |
| OCR | 1 | ocr-and-documents |
| Other | ~25 | dogfood, teams-meeting-pipeline, humanizer, ideation, llm-wiki, macos-computer-use, etc. |
| Dimension | Julian Goldie's Agent OS | Solid Circle |
|---|---|---|
| Concept | Dashboard coordination layer for AI agents | Named agent characters with domain-specific roles |
| Architecture | 7-layer blueprint (hardware→memory→models→agents→command→production→loop) | Agent profiles/skills within Hermes framework |
| Core Agents | 3 (Claude/OpenClaw/Hermes) | 18+ (Vera + 5 iAgents + 10 subs + 2 vAgents) |
| Agent Design | General-purpose with role assignment | Highly specialized domain agents |
| Memory | Obsidian vault + OMI screen/mic recording | Hermes native memory + session DB + Obsidian bridge |
| Dashboard | Custom Next.js + Tailwind web app | Telegram Mini App + Android app |
| Monetization | $59/mo AI Profit Boardroom | Free → tiered (iMe staking, $0.01 SOLID) |
| Models | OpenRouter (any model, free-first) | Mix (DeepSeek, Claude, Gemini, local GPU) |
| Skills Count | 27+ pre-built "launch kits" | 150+ skills (30 trained, 120 GitHub) |
| Self-Improvement | Compounding memory vault | vAgent self-documenting SKILL.md + cross-agent vault |
| Strengths | Shared memory, compounding knowledge, local-first | Deep domain specialization, 18-agent roster, real blockchain integration |
| Pattern | When Adopted | Status |
|---|---|---|
| Shared Memory Vault (Obsidian) | July 22, 2026 | ✅ Active — Solid Circle/Shared Agent Memory.md |
| Compounding Loop | July 22, 2026 | ✅ Active — agents write learnings per session |
| Free-First Stack | Inherent | ✅ Active — proved workflows before paying |
| 3-Agent Team (CEO/COO/Worker) | Partial | ⚠️ Informal — Vera CEO, vCode COO for builds |
| Live Agent Dashboard | Not yet | ❌ Not built — Mini App has agent cards, not live status |
| Per-Agent Analytics | Not yet | ❌ Not built — tokens/sessions not tracked per agent |
| 27 Launch Kits | Not yet | ❌ Not organized — 150 skills exist but aren't bundled |
~/julian-goldie-agent-os-research.md — 19KB comprehensive report~/agent-os-research.md — Technical implementation details~/julian-goldie-agent-os-architecture.html — Interactive SVG architecture diagram~/OneDrive/Documents/Obsidian Vault/Solid Circle/Shared Agent Memory.md — Cross-agent vault
Solid Circle · iMe Protocol · July 25, 2026 · Compiled by Vera / Hermes Agent
150+ skills · 30+ trained in-house · 12 extracted from Agent OS research · 120 from GitHub