🦞 Agentic OS

A beautiful local command centre for your AI agents. Built by Julian Goldie for AIPB members.

A single dashboard for Claude Code, OpenClaw, Hermes, and any other CLI agent.

Chat. Voice input. Goals. Journal. Every interaction auto-logged to your Obsidian vault.

All running on your laptop. None of your data leaves.

---

✨ What's inside

• 💬 Chat with multiple AI agents from one beautiful dashboard
• 🎤 Voice input in every chat box (Chrome/Safari)
• 🧠 Auto-saved to Obsidian — every chat becomes a markdown note
• 🎯 Goals page that writes a real task list to your vault
• 📓 Journal page — daily entries, one file per day
• 📖 Built-in build guide — teach others how you made yours
• ✨ Mission-control aesthetic — aurora gradients, glass panels, voice-pulse animations
• 🦞 Real CLI bridge — calls your local Claude / OpenClaw / Hermes binaries

---

🟢 Requirements

• Node 22+ (node -v to check, brew install node if missing)
• macOS or Linux (Windows works with WSL2)
• At least one AI agent CLI installed locally:
  • Claude Code
  • OpenClaw
  • Hermes Agent
• Optional but recommended: an Obsidian vault

If you only have one agent installed, that's fine — missing ones just don't show in the dashboard.

---

🚀 Quick start (5 minutes)

# 1. Clone the repo (you need to be added as a collaborator — see below)
git clone https://github.com/JulianGoldie/agentic-os.git
cd agentic-os

# 2. Install dependencies
npm install

# 3. Configure your paths
cp agentic-os.config.example.json ~/.agentic-os/config.json
# Edit ~/.agentic-os/config.json with your vault path and agent binary paths

# 4. Run it
npm run dev

Open http://localhost:3000 in your browser.

---

⚙️ Configuration

Agentic OS reads config from (in priority order):

1. Environment variables
2. ~/.agentic-os/config.json
3. Auto-detection (which claude, common Obsidian paths)
4. Sensible defaults

Minimal config.json

{
  "claude": "/Users/you/.local/bin/claude",
  "openclaw": "/Users/you/local/node/bin/openclaw",
  "hermes": "/Users/you/.local/bin/hermes",
  "vaultRoot": "/Users/you/Documents/Obsidian Vault",
  "goalCategories": ["Health", "Work", "Personal"]
}

Find the right paths with:

which claude     # → paste into "claude"
which openclaw   # → paste into "openclaw"
which hermes     # → paste into "hermes"

For your Obsidian vault, just point it at the folder you open in Obsidian.

Environment variables (alternative)

If you'd rather not edit a JSON file, use a .env.local:

cp .env.example .env.local
# Edit .env.local with your paths

---

🧪 First-run check

Once running, hit each route to confirm everything's wired:

• http://localhost:3000 — Mission Control overview
• http://localhost:3000/claude — Claude chat (needs Claude Code installed)
• http://localhost:3000/openclaw — OpenClaw chat + control room
• http://localhost:3000/hermes — Hermes chat + control room
• http://localhost:3000/memory — Search your Obsidian vault
• http://localhost:3000/goals — Goals (writes to vault)
• http://localhost:3000/journal — Daily journal
• http://localhost:3000/guide — How-to-build-your-own guide

If an agent tile says "not installed", check which <agent> returns a path. If it does, paste that path into your config.json.

---

🎨 Customising

Six files for the most common changes:

Want to...	Edit
Add a new agent	src/lib/runner.ts + src/lib/config.ts
Change vault location	your config.json or .env.local
Change colours	src/app/globals.css (CSS variables at the top)
Change goal categories	your config.json (goalCategories)
Add a new sidebar page	src/components/Sidebar.tsx + src/app/<page>/page.tsx
Tweak the build guide	BUILD-YOUR-OWN.md

---

🔒 Privacy & data

• Everything runs on localhost. No accounts, no telemetry, no servers.
• Your chats are written only to your Obsidian vault as plain markdown.
• API routes shell out to your local CLIs via child_process.spawn — no shell interpolation, so prompt content can't run commands.
• /api/run enforces a per-agent regex allowlist for any tool-style commands.
• Path traversal blocked on vault-read endpoints.

Audit it yourself — the whole thing is about 2,500 lines.

---

🛠 Troubleshooting

"agent is not installed" Your CLI isn't on PATH or auto-detection missed it. Edit ~/.agentic-os/config.json and paste the full path from which <agent>.

"no output" from chat Run the agent directly in your terminal first (claude -p "hi"). If that works, restart npm run dev. If it doesn't, the agent's broken, not the dashboard.

Voice button is grey Voice needs Chrome or Safari. Firefox doesn't support the Web Speech API.

Slow agents (>30s) Normal for some local models (ollama/deepseek). The "thinking… 18s" counter shows it's still working. If too slow, point that agent at a faster cloud model.

Routes return 404 Make sure you ran npm install and you're on Node 22+.

---

🤖 Building your own from scratch

The full guide for using Claude Code to build this same system is at:

• BUILD-YOUR-OWN.md in this repo
• http://localhost:3000/guide once it's running

It's 8 copy-paste prompts that mirror exactly how Julian built his.

---

📜 Licence

For AIPB members only. Not for redistribution. You can fork it for personal use. Just don't resell.

— Julian Goldie · AIPB