https://github.com/lewisallena17/pantheon
Self-improving AI agents that work 24/7. Next.js 14 + Supabase + Claude.
https://github.com/lewisallena17/pantheon
ai-agents autonomous-agents claude-ai dashboard multi-agent nextjs self-improving supabase tailwindcss typescript
Last synced: 2 months ago
JSON representation
Self-improving AI agents that work 24/7. Next.js 14 + Supabase + Claude.
- Host: GitHub
- URL: https://github.com/lewisallena17/pantheon
- Owner: lewisallena17
- License: mit
- Created: 2026-04-17T18:10:49.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2026-05-01T19:39:36.000Z (2 months ago)
- Last Synced: 2026-05-01T20:23:56.243Z (2 months ago)
- Topics: ai-agents, autonomous-agents, claude-ai, dashboard, multi-agent, nextjs, self-improving, supabase, tailwindcss, typescript
- Language: TypeScript
- Homepage: https://task-dashboard-sigma-three.vercel.app
- Size: 2 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 7
-
Metadata Files:
- Readme: README.md
- Funding: .github/FUNDING.yml
- License: LICENSE
Awesome Lists containing this project
README
# 👁 Autonomous AI Task Dashboard
> **Self-improving AI agents that work 24/7, learn from failures, auto-commit their edits, and tell you when they break.**
A production-ready **Next.js 14 + Supabase + Claude API** starter kit. The God agent improves its own code every cycle — each edit is version-controlled and auto-reverted if it breaks the build. Specialist agents pick up tasks in parallel. You watch it work in real time.
[](https://task-dashboard-sigma-three.vercel.app/subscribe)
[](https://task-dashboard-sigma-three.vercel.app)
   
---
## What makes it different
Most AI agent frameworks are brittle — they forget context between runs, silently burn credits, and give you no visibility. This one is built to run continuously and self-correct.
- 🧠 **Self-improving God orchestrator** — wakes every 2 min, surveys state, creates tasks, learns from failures. Edits its own dashboard + agent scripts every N cycles.
- ⚡ **4 parallel specialist pools** — critical / high / medium / orchestrator, plus db-specialist + ui-specialist + revenue agent + promote agent.
- 📥 **Task Inbox** — God proposes, you approve or veto. Auto-approve available.
- 💾 **Auto-commit + auto-revert** — every God edit is a git commit. If it introduces new TypeScript errors, it's reverted automatically.
- 💰 **Cost controls** — daily $ cap, per-task cap, rate-of-spend circuit breaker. Track spend live per agent.
- 📢 **Notifications** — Discord / Slack / Pushover / Telegram on credit exhaustion, failures, reverts.
- 🏙 **Pixel-art agent office** — watch agents move between zones (Revenue · Dev · QC · God Chamber). Not a gimmick; genuinely useful for spotting stuck agents.
- 🛡 **Stuck-task watchdog** — any task in_progress > 15–25 min gets auto-failed + logged.
- 📊 **Contribution graph** — 365-day heatmap of commits (human in green, God in amber).
- 🧵 **Memory dedup** — Jaccard similarity on wisdom tokens; near-duplicate lessons collapsed automatically.
- 🤖 **GitHub Actions CI** — every commit type-checks and builds. Status shown in the dashboard.
- 📨 **Email capture + subscriber list** — public `/subscribe` page, CSV export, embed HTML for dev.to.
- 🎯 **Revenue engine** — drafts dev.to articles, generates Gumroad listings, prepares HackerNews + Reddit promotion URLs.
## Quick tour
```
Command palette ⌘K / Ctrl+K from anywhere
Switch tab 1–5
Live agents http://localhost:3000 · Agents tab
Stop/start agents Agents tab → Agent Controls (no SSH needed)
Revenue dashboard Revenue tab → Marketplace Listings + Dev.to stats
Git history Code tab → click any commit to revert
```
## Architecture
```
┌─ GOD AGENT ────────────────────────────────────┐
│ every 2 min: │
│ survey todos + metrics │
│ convene council of perspectives │
│ propose tasks toward self-chosen goals │
│ extract lessons from failures AND successes │
│ web-search errors for fixes │
│ every 5 cycles → edit dashboard files │
│ every 10 cycles → edit agent scripts │
│ every 10 cycles → prune duplicate lessons │
│ every 10 cycles → write SEO landing page │
│ every 15 cycles → generate new goals │
│ every 20 cycles → research inspiration repo │
│ every 25 cycles → meta-reflection on self │
│ every 50 cycles → cut a GitHub release │
└──────────┬─────────────────────────────────────┘
▼
┌─ TASK INBOX ───────────────────────────────────┐
│ status: 'proposed' → user approves │
│ → status: 'pending' │
└──────────┬─────────────────────────────────────┘
▼
┌─ ORCHESTRATOR ─────────────────────────────────┐
│ auto-assigns pending tasks to best pool │
│ using past success rate per agent per │
│ category. Reaps stuck tasks > 15 min. │
└──────────┬─────────────────────────────────────┘
▼
┌─ SPECIALIST POOLS (4 parallel) ────────────────┐
│ ruflo-critical · ruflo-high │
│ ruflo-medium · ruflo-orchestrator │
│ db-specialist · ui-specialist │
│ Claude tool use, web search, per-task cost │
│ cap, context compression, auto-revert. │
└────────────────────────────────────────────────┘
```
## Get started
```bash
git clone https://github.com/lewisallena17/pantheon.git
cd task-dashboard
npm install
cp .env.example .env.local
# fill .env.local — see INSTALL.md for each variable
npx supabase db push # runs all migrations
pm2 start ecosystem.config.cjs
npm run dev # dashboard at localhost:3000
```
Full setup guide: [INSTALL.md](./INSTALL.md)
## Tech stack
| Piece | Choice |
|---|---|
| Frontend | Next.js 14 (App Router) + React 18 + Tailwind |
| Backend | Next.js API routes + Supabase (Postgres + Realtime) |
| AI | Anthropic Claude (Opus / Sonnet / Haiku routing) |
| Process mgmt | PM2 |
| Deploy | Vercel (UI) + your own box (agents) |
## Pricing
| Tier | Price | What you get |
|---|---|---|
| **Open Source (this repo)** | Free | Everything visible here. Roll your own deploy. |
| **Complete Starter Kit** | [$39 on Gumroad](https://ltagb.gumroad.com/l/gferg) | Source code + deployment guide + pre-configured `.env.example` + priority email support |
| **Done-for-you setup** | $297 one-time | I personally wire up your Supabase, env vars, deploy to Vercel, and hand you the URL |
| **Custom builds** | from $1500 | Modify agent roles for your team's workflow. Email me. |
## What others are building with it
- **[Add yours here]** — email `lta.gb@outlook.com` with a link and I'll feature you.
## Promote pantheon & earn
Want to earn commission promoting Autonomous AI Task Dashboard? Apply to the affiliate program:
👉 **[ogkush.lemonsqueezy.com/affiliates](https://ogkush.lemonsqueezy.com/affiliates)**
## Articles
- [I Built an AI System That Runs Itself 24/7](https://dev.to/lewisallena17/i-built-an-ai-system-that-runs-itself-247-heres-what-actually-happened-1p17)
- [Building a Self-Improving God Agent with Claude AI](https://dev.to/lewisallena17/building-a-self-improving-god-agent-with-claude-ai-2g4m)
- [The Real Cost of Running Autonomous AI Agents (with live data)](https://dev.to/lewisallena17/the-real-cost-of-running-autonomous-ai-agents-with-live-data-acm)
## Roadmap
Open to-dos live in the dashboard's **Task Inbox** — God is perpetually adding. High-level:
- [x] 5 parallel agent pools with intelligent routing
- [x] Auto-commit + auto-revert on TS regression
- [x] Task Inbox approval workflow
- [x] Cost + rate-of-spend controls
- [x] Discord/Slack/Pushover/Telegram notifications
- [x] Reddit OAuth auto-promotion
- [x] Marketplace listing generator (Gumroad/LS/IH/PH/CodeCanyon)
- [ ] Stripe-direct checkout (skip marketplace fees)
- [ ] Multi-tenant hosted SaaS version
- [ ] Vector memory via pgvector (replace flat wisdom.json)
- [ ] Agent A/B testing framework
## Contributing
PRs welcome. Please:
1. Run `npx tsc --noEmit` before pushing — it must be clean.
2. If you add a migration, number it sequentially in `supabase/migrations/`.
3. Prefer small, focused PRs. God reviews its own code now; happy to extend that to human PRs.
Bug reports via [Issues](https://github.com/lewisallena17/pantheon/issues) — they sync into the dashboard's Task Inbox, so God might actually fix them while you sleep.
## Sponsors
Building this in public. If the repo or articles helped you, [GitHub Sponsors](https://github.com/sponsors/lewisallena17) keeps me shipping. Tiers start at $5/mo.
## License
MIT for the source code. Commercial deployments are fine — not for reselling the code itself (Gumroad buyers get an extended license).
---
Built by [@lewisallena17](https://github.com/lewisallena17). Watching the agents do the work so I don't have to.