A locally-hosted operating system for AI agents — an open-source GitHub repository — that coordinates opencode, Hermes Agent, and Gemini CLI into a unified dashboard with persistent memory, cron scheduling, skill execution, cost analytics, and disaster recovery.
Why Agentic OS? Most agent tools work in isolation — a terminal for coding, a separate chat for research, another for memory. Agentic OS is the control plane that unifies them: one dashboard, one memory layer, one scheduler, one skill hub. Three agents, infinite capabilities.
| Category | Features |
|---|---|
| 🤖 3-Agent Engine | opencode (code/DevOps), Hermes (memory/scheduling), Gemini (research/analysis) with intelligent routing |
| 🧩 16+ Skills | Executable skill packs with eval scoring, learnings, and score history per run |
| 🧠 Persistent Memory | SQLite FTS5 + brain/ folder — shared context read by all agents at session start |
| ⏱ Cron Scheduler | APScheduler-powered jobs — heartbeat, memory consolidation, daily standup, DevOps audit |
| 💰 Cost Analytics | Track spending per provider, model, agent. Free-tier alerts prevent surprise bills |
| 📋 Audit Trail | Every action logged — chat, skill runs, settings changes, backups |
| 💾 One-Click Backup | Full tar.gz snapshot of all configs, skills, brain, agents, prompts |
| 📝 Prompt Library | 10 reusable templates — code review, system audit, project plan, brainstorm, and more |
| 📐 Standards System | Discover and inject coding conventions across your project |
| 🔌 Plugin Registry | Marketplace-style plugin management (extensible via skills) |
| 🎨 Dark/Light Theme | GitHub-style dark mode + clean light theme, toggle from sidebar |
| ⚡ Zero API Costs | Built for free tiers — Gemini Flash, OpenRouter free models, local opencode |
| 📋 Kanban Board | Visual task management — drag-and-drop columns, priority/status filtering, block/unblock, detail view |
| 🎯 Goals | Project targets with progress tracking, auto-syncs to brain/active-projects.md |
| 📓 Journal | Daily markdown entries stored as brain/journal/YYYY-MM-DD.md with full-text search |
| ❤️ Agent Health | Real-time monitoring of opencode, Hermes, and Gemini CLI availability |
| 🧭 Smart Router | Keyword-based task routing with confidence scoring — suggests best agent for any task |
| 📊 Learning Analytics | Skill evaluation scores, performance trends, and historical charts |
| 🎬 Session Replay | Browse and replay past opencode sessions from the dashboard |
| ⚠ Error Dashboard | Real-time error tracking with category filtering and circuit breaker status |
| 🔌 Circuit Breaker | Auto-trip after N failures, auto-recovery after 300s, manual reset |
| ⏱ Event-Driven Scheduler | File-watcher auto-reloads jobs on change, webhook receiver, execution history |
| 🔗 Webhook Receiver | /api/webhook — trigger skill execution from external tools |
| 🧠 SQLite FTS5 Memory | Full-text search across brain, skills, journal with entity extraction |
| 🤖 Auto-Skill Generator | POST /api/skills/generate — create SKILL.md from natural language |
| 📱 Mobile PWA | Bottom navigation bar, manifest.json, service worker, touch-friendly UI |
┌──────────────────────────────────────────────────────────────┐
│ AGENTIC OS DASHBOARD │
│ FastAPI + Tailwind SPA │
├──────────────────────────────────────────────────────────────┤
│ │
│ ┌───────────────┐ ┌────────────────┐ ┌────────────────────┐ │
│ │ opencode │ │ Hermes │ │ Gemini CLI │ │
│ │ (Code/DevOps) │ │ (Memory/Sched) │ │ (Research/Analy) │ │
│ │ File Ops) │ │ /Channels) │ │ │ │
│ └───────────────┘ └────────────────┘ └────────────────────┘ │
│ │
│ ┌──────────────────────────────────────────────────────────┐ │
│ │ 7 CORE LAYERS (Stacked) │ │
│ │ Layer 7: Identity / Persona / Constitution │ │
│ │ Layer 6: Self-Evolution + Capability Manager │ │
│ │ Layer 5: Scheduler + Awareness + Health Guardian │ │
│ │ Layer 4: Memory Graph + Memory Consolidation │ │
│ │ Layer 3: Skills Hub + Eval + Learnings Loop │ │
│ │ Layer 2: Business Brain + Context Folders │ │
│ │ Layer 1: Agent Router + Standards + Profiles │ │
│ └──────────────────────────────────────────────────────────┘ │
└──────────────────────────────────────────────────────────────┘
| Agent | Role | Default Model | Provider | Cost |
|---|---|---|---|---|
| opencode | Code generation, DevOps, file operations | deepseek-v4-flash-free | opencode-zen | $0 |
| Hermes Agent | Persistent memory, scheduling, messaging | Owl Alpha (1M ctx) | OpenRouter | $0 |
| Gemini CLI | Web research, multi-modal analysis | gemini-2.5-flash | Google OAuth | $0 |
- Code/DevOps task? → opencode
- Memory/Channel/Schedule? → Hermes Agent
- Research/Analysis? → Gemini CLI
- Complex multi-step? → Chain: Gemini researches → opencode implements → Hermes monitors/schedules
git clone https://github.com/modimihir07/agentic-os.git
cd agentic-os
chmod +x install.sh && ./install.sh
./start.sh
# Open http://127.0.0.1:8080| Tool | Required? | Install |
|---|---|---|
| Python 3.10+ | ✅ Required | sudo apt install python3 python3-pip |
| Node.js 18+ | ⚠ For opencode | curl -fsSL https://deb.nodesource.com/setup_20.x | sudo bash - && sudo apt install -y nodejs |
| opencode | ⚠ For code tasks | npm install -g @opencode/cli |
| Hermes Agent | ⚠ For memory/scheduling | curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash |
| Gemini CLI | ⚠ For Google AI | npm install -g @google/gemini-cli |
⚠ = Optional — the dashboard works with any subset of installed agents.
# Set your OpenRouter API key
echo 'OPENROUTER_API_KEY=sk-or-v1-your-key-here' > ~/.hermes/.envgemini auth login
# Complete OAuth in browser — tokens saved to ~/.gemini/oauth_creds.jsonEdit data/settings.json:
{
"dashboard": { "port": 8080 },
"theme": "dark",
"agents": { "opencode": true, "hermes": true, "gemini": true }
}agentic-os/
├── server.py # FastAPI backend (REST API)
├── requirements.txt # Python dependencies
├── install.sh # One-command installer
├── start.sh # Launch dashboard
├── backup.sh / restore.sh # Disaster recovery
│
├── dashboard/ # Web frontend (SPA)
│ ├── index.html # Entry point + sidebar
│ ├── app.js # SPA router + theme/sidebar toggle
│ ├── api.js # API client (all endpoints)
│ ├── styles.css # Full dark/light theme CSS
│ ├── utils.js # Shared utilities
│ └── pages/ # 21 page modules (13 original + 7 v0.2.0 + 1 v0.3.0)
│ ├── dashboard.js # Overview with stats
│ ├── skills.js # Skill grid/list/detail
│ ├── memory.js # Brain file editor
│ ├── chat.js # Multi-agent chat
│ ├── scheduler.js # Cron job manager
│ ├── audit.js # Activity trail
│ ├── cost.js # Cost analytics charts
│ ├── plugins.js # Plugin marketplace
│ ├── backups.js # Backup/restore UI
│ ├── prompts.js # Template library
│ ├── standards.js # Code conventions
│ ├── settings.js # Config editor
│ ├── setup-wizard.js # Guided setup
│ ├── kanban.js # ▸ Kanban Board (v0.2.0)
│ ├── goals.js # ▸ Goals (v0.2.0)
│ ├── journal.js # ▸ Journal (v0.2.0)
│ ├── agent-health.js # ▸ Agent Health (v0.2.0)
│ ├── smart-router.js # ▸ Smart Router (v0.2.0)
│ ├── learning-analytics.js # ▸ Learning Analytics (v0.2.0)
│ ├── session-replay.js # ▸ Session Replay (v0.2.0)
│ └── errors.js # ▸ Error Dashboard (v0.3.0)
│
├── brain/ # Shared context (all agents read)
│ ├── memory_search.py # ▸ SQLite FTS5 search (v0.3.0)
│ ├── business-brain.md # Current project context
│ ├── memory.md # Accumulated knowledge
│ ├── recent-decisions.md
│ ├── active-projects.md
│ ├── identity.md
│ ├── constitution.md
│ └── journal/ # Daily markdown entries (YYYY-MM-DD.md)
│
├── skills/ # 16 executable skills
│ ├── devops-audit/ # GCP/CloudMart infra audit
│ ├── heartbeat/ # 5-min health check
│ ├── content-draft/ # Blog/newsletter writing
│ ├── code-review/ # PR review checklist
│ ├── research-synthesis/ # Gemini research aggregator
│ ├── daily-standup/ # Morning briefing
│ ├── meeting-minutes/ # Notes processor
│ ├── project-planner/ # Step-by-step plans
│ ├── brainstorming/ # Socratic design
│ ├── systematic-debug/ # 4-phase root cause
│ ├── memory-consolidation/ # Weekly synthesis
│ ├── backup-skill/ # Snapshot creator
│ ├── cost-analytics/ # Multi-provider cost
│ ├── tdd-cycle/ # Red-green-refactor
│ ├── goal-planner/ # Goal → steps
│ └── _template/ # Starter template
│
├── agents/ # Per-agent configs
├── scheduler/jobs/ # Cron job definitions
├── registry/ # Plugin marketplace
├── standards/ # Discover/inject conventions
├── prompts/ # 10 reusable templates
├── data/ # Runtime data (agent-routes.json tracked; settings/cost/chat/error/scheduler/memory/circuit/kanban gitignored)
├── audit/ # Activity log (gitignored)
└── backups/ # Snapshots (gitignored)
| Feature | Description |
|---|---|
| 🔒 Security Audit | 23 vulnerabilities fixed: 3 CRITICAL (path traversal), 8 HIGH (XSS, command injection, missing headers), 7 MEDIUM. Security headers middleware added (CSP, HSTS, X-Frame-Options) |
| ⏱ Event-Driven Scheduler | Rewritten with file-watcher auto-reload (watchdog-style), webhook receiver at /api/webhook, execution history, manual job triggers |
| ⚠ Error Dashboard | New /api/errors endpoints with category filtering (agent/skill/api/system). Dedicated dashboard page with error log + circuit breaker status cards |
| 🔌 Circuit Breaker | Auto-trip after 3 failures, half-open recovery after 300s. Per-agent state tracking with manual reset. Prevents cascading failures to offline agents |
| 🧠 Persistent Memory (SQLite FTS5) | Full-text search across brain/*.md, skills/*/*.md, brain/journal/. Entity extraction (persons, emails, URLs, acronyms, IPs). Reindex endpoint |
| 🤖 Auto-Skill Generator | POST /api/skills/generate takes natural language description → creates full SKILL.md with eval.json and context folder |
| 🔗 Webhook Receiver | Generic webhook at /api/webhook/generic plus skill-targeted webhooks. Integrates with GitHub, CI/CD, external tools |
| 📱 Mobile PWA | Bottom navigation bar, manifest.json, service worker (offline fallback), PWA meta tags, touch-friendly 16px inputs, responsive grid collapse |
- Path traversal: All file endpoints validate
..and/in user-supplied names - XSS elimination: 19 violations fixed across 7 JS files — all
onclickhandlers useencodeURIComponent, all text usesescapeHtml() - Security headers: CSP, HSTS, X-Frame-Options, X-Content-Type-Options, Referrer-Policy
- CORS restricted: Only
http://127.0.0.1:8080andhttp://localhost:8080 - API keys masked:
/api/settingsreturns keys as masked values (e.g.,sk-o****) - Input validation: Chat messages limited to 10K chars, brain/skill names reject special chars
- Session replay: Content limited to 2000 chars, path traversal prevented
- Runtime data gitignored:
error-log.json,circuit-breaker.json,scheduler-history.json,memory.db,kanban/*.json
| Feature | Description |
|---|---|
| 📋 Kanban Board | Visual task management with 6 columns (triage → todo → ready → in_progress → blocked → done), drag-and-drop, priority labels, filtering, and detail modals with complete/block/unblock actions |
| 🎯 Goals | Create and track project targets with progress bars, categories, and target dates. Auto-syncs to brain/active-projects.md for agent awareness |
| 📓 Journal | Daily markdown journal entries stored as brain/journal/YYYY-MM-DD.md. Full-text search, day streak tracking, word count |
| ❤️ Agent Health | Real-time dashboard showing online/offline status for all 3 agents (opencode, Hermes, Gemini CLI). Auto-refresh every 5 seconds |
| 🧭 Smart Router | Keyword-based routing engine — type a task description and get an AI-suggested agent with confidence score. Manual override available |
| 📊 Learning Analytics | Skill evaluation scores, performance trends, and per-skill detail breakdowns with mini bar charts |
| 🎬 Session Replay | Browse and replay past opencode sessions directly from the dashboard. View message content and timestamps |
- Glass morphism cards with subtle backdrop blur
- Glow borders and gradient accents on interactive elements
- Skeleton loaders with shimmer animation for async content
- Empty states with icons and contextual messages
- All CSS additions are zero-breaking — existing 13 pages and 28 endpoints unchanged
Select an agent from the sidebar → type your message → get response.
| Agent | Best for | Example |
|---|---|---|
| opencode | "Check system status", "Deploy to GKE" | Code + terminal automation |
| Hermes | "What did I work on recently?", "Schedule a daily backup" | Memory recall, scheduling |
| Gemini | "Research latest AI agent trends", "Analyze this image" | Web research, multi-modal |
Browse 16 skills from Skills Hub → click Run → monitor eval scores over time.
Create cron jobs: heartbeat (5 min), memory consolidation (weekly), daily standup, DevOps audit.
Track spending per provider/model/agent. Free-tier alerts warn when nearing limits.
Drag tasks across columns, filter by priority/category, click to view details. Block tasks when blocked, mark complete when done.
Create goals with categories and target dates. Progress tracked via +25% increments. Completed goals auto-sync to brain context.
Write daily entries with markdown support. Auto-saves after 2 seconds. Search across all entries from the dashboard.
Describe a task in plain English — the router analyzes keywords and suggests the best agent. Route manually or let AI decide.
Monitor online status of all 3 agents in real time with 5-second auto-refresh. Health checks are filesystem-based (no subprocess calls).
View evaluation scores for all 16 skills. Trends chart shows score progression over time. Top skills ranked by performance.
Browse opencode session logs by date and size. Click "Replay" to view all messages in a chat-like interface.
View system errors grouped by category (agent, skill, API, system). Filter by category, clear all errors. Circuit breaker cards show per-agent failure state — reset from the dashboard.
Jobs auto-reload when JSON files in scheduler/jobs/ change. Trigger skills via webhooks at POST /api/webhook with {"skill": "skill-name", "payload": {...}}. Manual trigger at /api/scheduler/trigger/{job_id}.
Full-text search across all brain files, skills, and journal entries via SQLite FTS5. Query at GET /api/memory/search?q=... with highlighted snippets. Reindex at POST /api/memory/reindex. Entities (persons, emails, URLs) auto-extracted.
Create new skills from natural language: POST /api/skills/generate with {"name": "my-skill", "description": "Does X by doing Y"}. Generates SKILL.md, eval.json, learnings.md, and context folder.
POST /api/webhook — trigger any skill by name. POST /api/webhook/generic — catch-all for external tool integration (GitHub, CI/CD, etc.).
Open Agentic OS on your phone — bottom nav bar replaces sidebar, touch targets are 44px+, service worker caches assets. Add to home screen for app-like experience.
| Feature | Claude Agent OS (Video) | Agentic OS (This Project) |
|---|---|---|
| Core Agents | Claude + OpenClaw + Hermes | opencode + Hermes + Gemini CLI |
| Cost | $20/mo (Claude subscription) | $0 — all free tiers |
| Stack | Next.js + Tailwind | FastAPI + vanilla JS SPA |
| Architecture | 4 layers | 7 layers |
| Skills System | Plugin marketplace (2,000+ from Hermes) | 16 curated skills + eval scoring + learnings |
| Memory | Obsidian vault (external) | Built-in brain/ + SQLite FTS5 |
| Scheduler | Not shown | APScheduler cron jobs |
| Cost Tracking | Not shown | Built-in per-provider analytics |
| Backup/Restore | Not shown | One-click tar.gz snapshots |
| Audit Trail | Not shown | Full activity log |
| Standards System | Not shown | Discover/inject conventions |
| Client Timeout | Not shown | 200s AbortController |
| Kanban Board | Yes | Yes — built-in with drag-and-drop, priority, block/unblock, filters |
| Open Source | No (tutorial only) | MIT License |
- OS: Linux (Ubuntu 22.04+), macOS
- Python: 3.10, 3.11, 3.12
- Browsers: Chrome, Firefox, Edge
- Agents: opencode v0.8+, Hermes Agent v1.0+, Gemini CLI v1.0+
- Fork the repo
- Create a feature branch (
git checkout -b feature/amazing-feature) - Commit your changes (
git commit -m 'Add amazing feature') - Push (
git push origin feature/amazing-feature) - Open a Pull Request
See data/agent-routes.json for routing rules and skills/_template/ for creating new skills.
MIT License — free to use, modify, and distribute.
Agentic OS — Built with ❤️ by Mihir Modi
📖 Read the dev.to article · Report Bug · Discussions · GitHub