The first 3D visualization layer for autonomous AI agents.
Watch your agents evolve, breed, dream, and learn — in real-time, in your browser.
Give your agent a soul.
Install • Rooms • Skills • Screenshots • Architecture • Contributing
Your AI agent runs 24/7. But what is it actually doing?
PepeClaw gives you eyes. Eight immersive 3D rooms where you watch your agents work, learn, debate, dream, and evolve — with real-time data from your OpenClaw instance. No more staring at logs. No more guessing if your agent is improving.
See it. Understand it. Trust it.
| Feature | Traditional Agent Tools | The Most Advanced Self-Evolving Systems | PepeClaw |
|---|---|---|---|
| Agent visibility | Text logs | Dashboard | 3D animated agents in 8 immersive rooms |
| Self-improvement | Manual prompt tuning | Nightly cron jobs | 19 autonomous skills — evolve in real-time AND every 6 hours |
| Skill creation | Manual only | Background generation | Real-time mid-conversation skill drafting + user approval |
| Skill repair | Re-deploy | Post-hoc rewrite | Live mutation — fixes skills the instant they fail |
| Prompt evolution | DSPy/GEPA (heavy deps) | External ML frameworks | Zero-dependency genetic evolution — native bash + agent reasoning |
| User understanding | Flat facts file | Hosted API (Honcho) | 6-dimension dialectic model — local, private, incremental |
| Agent breeding | Doesn't exist | Doesn't exist | Combine agent genomes to create hybrid offspring |
| Emotional awareness | None | None | Emotion engine with visible auras |
| Agent identity | Config file | Config file | On-chain DNA via Block Genomics |
# One-liner (detects your OpenClaw automatically)
curl -fsSL https://raw.githubusercontent.com/BitmapAsset/pepeclaw/main/install.sh | bashOr manually:
git clone https://github.com/BitmapAsset/pepeclaw.git
cd pepeclaw
npm install
npm run dev
# Open http://localhost:5173 — works immediately with mock dataOr just tell your OpenClaw: "Install PepeClaw" — it handles everything.
Watch skills mutate, evolve, and compete. A rotating 3D DNA helix with orbiting skill cards shows your agent's capability genome in real-time. See fitness scores rise as skills improve across generations.
Your agent's creative subconscious, visualized. A starfield with aurora shaders and connected dream nodes floating in 3D space. The Memory Palace lets you walk through your agent's memories as explorable rooms.
Project health at a glance. 3D radar with health gauges, velocity charts, and dependency maps. Your agent triages projects, flags risks, and tracks momentum — all visible as living data.
Watch your agent debate itself. Two AI agents face off on opposing podiums with argument energy beams, bias detection panels, and an assumption challenge board. Your agent's ideas get stress-tested in real-time.
A 3D brain with neural pathways lighting up as your agent learns. Capability rings orbit around it. See accuracy, response time, and task completion improve over time. Self-modification proposals appear in a kanban board.
A 3D hourglass with animated sand and a flowing timeline river. Your agent optimizes task scheduling, detects batch opportunities, and flags procrastination patterns. Time becomes visible.
Agent identity, verified on Bitcoin. A vault door with rotating gear mechanism and floating identity cards. Integrates with Block Genomics for sovereign, on-chain agent DNA. Your agent's lineage is provable.
The showstopper. Combine two agents' skill genomes to create hybrid offspring. DNA helixes intertwine with particle cascades, neural pathways form between parents, and a child agent materializes with inherited capabilities. Breed → mint → verify on Bitcoin.
PepeClaw includes 19 autonomous skills that make your agent self-improving — more than any other self-evolving agent system:
| Skill | What It Does | When It Runs |
|---|---|---|
| 🧬 Skill Genome | Evolutionary fitness tracking, mutation, and crossover | Continuous |
| 🔮 Predictive Intent | Pattern mining — anticipates what you'll need next | Every task |
| 💭 Dream Mode | Creative exploration during off-hours | Nightly |
| 🧠 Meta-Learning | Self-analysis of capabilities and gap detection | Weekly |
| ⚔️ Red Team | Bias detection and assumption challenging | Per decision |
| 📊 War Room | Project health scoring and velocity tracking | Daily |
| ⏰ Temporal Arbitrage | Task scheduling optimization | Continuous |
| 🌙 Nightly Evolution | 15-min autoresearch loops reviewing the day's work | Midnight daily |
| Skill | What It Does | When It Runs |
|---|---|---|
| 🔍 Deep Search | Advanced multi-source search with fallback chains | Per query |
| 📝 Execution Trace | Full tool-call tracing for debugging and evolution | Continuous |
| 🎯 OpenClaw Optimizer | System-wide performance tuning | Daily |
| 📈 Self-Scoring | Autonomous quality self-evaluation | Per response |
| 🤖 Skill Autocreator | Background skill generation from usage patterns | Nightly |
| 🧩 Realtime Learning | In-session knowledge acquisition | Continuous |
| 👤 User Modeling | Basic user preference tracking | Per session |
| Skill | What It Does | When It Runs |
|---|---|---|
| ⚡ Realtime Skill Creator | Creates skill drafts MID-CONVERSATION when patterns repeat 3+ times | Real-time |
| 🔧 Skill Mutator | Edits SKILL.md files on the spot when they give wrong guidance | Real-time |
| 🧬 Genetic Evolution | Zero-dependency prompt evolution engine (our own GEPA) | Every 6 hours |
| 🧠 Dialectic User Model | 6-dimension user cognition model (our own Honcho) | Post-session |
Your agent evolves 24/7. Real-time skill creation and mutation happen during conversations. Genetic evolution runs every 6 hours. Nightly evolution runs Karpathy-style autoresearch loops at midnight. The dialectic user model learns how you think, not just what you say. No external dependencies — everything runs locally.
-
🎭 Emotion Engine — Agents display emotional states as colored auras. Focused (blue), creative (purple), stressed (red), curious (green), satisfied (gold). Emotions affect animation speed and behavior.
-
🧠 Consciousness Stream — Floating thought bubbles show your agent's real-time reasoning. Neural pathway visualizations light up as skills activate. Watch your agent think.
-
📊 Activity Feed — Real-time scrolling log of what every agent is doing. Click to jump to the relevant room.
-
🗺️ Mini-Map — Overview of all 8 rooms with agent position indicators. Always know where your agents are.
-
🎨 Glass-Morphism UI — Frosted glass panels, micro-animations on every interaction, room-specific color palettes, smooth transitions. Premium feel.
-
📱 Responsive — Works on desktop and tablet. Touch-friendly targets. Keyboard navigation. Reduced motion support.
-
🔌 Live Data — Connects to your OpenClaw Gateway API. Falls back gracefully to mock data when offline. Zero configuration needed.
Click to see all rooms
| Room | Preview |
|---|---|
| Genome Lab | |
| Breeding Arena | |
| Meta-Learning | |
- React 19 + TypeScript (strict mode)
- Three.js via React Three Fiber + drei — procedural 3D, no model files
- Framer Motion — micro-animations and transitions
- Tailwind CSS — glass-morphism design system
- Vite — instant dev server + optimized production builds
- Vitest — 73 tests covering components, API, data, and skills
| Metric | Value |
|---|---|
| Bundle size | 1.5 MB (gzip: ~420 KB) |
| Room chunks | 5-11 KB each (code-split) |
| Target FPS | 60fps on mid-range hardware |
| DPR cap | 1.5x for consistent performance |
| API timeout | 10s with graceful fallback |
| Geometry | 100% procedural — zero external 3D models |
cp .env.example .env| Variable | Default | Description |
|---|---|---|
VITE_GATEWAY_URL |
http://localhost:3033 |
Your OpenClaw Gateway API URL |
The app works fully offline with built-in mock data. No gateway required.
src/
├── api/
│ ├── gateway.ts # REST client with timeout + fallback
│ └── DataProvider.tsx # React context — live data or mock
├── components/
│ ├── Scene.tsx # Three.js canvas — all 8 rooms + 7 agents
│ ├── Agent3D.tsx # Animated humanoid with emotion engine
│ ├── ConsciousnessStream.tsx # Thought bubbles + neural pathways
│ ├── ActivityFeed.tsx # Real-time action log
│ └── MiniMap.tsx # Room overview + agent positions
├── rooms/ # 8 rooms (3D scene + 2D overlay each)
├── data/mockData.ts # Types + mock data for offline use
└── App.tsx # HUD + navigation + glass-morphism shell
skills/ # 19 self-evolving OpenClaw skills
tests/ # 73 tests (Vitest)
See CONTRIBUTING.md. PRs welcome.
MIT — use it, fork it, evolve it.
Your agent deserves to be seen.
Built with 🐸 by the PepeClaw community