Where agents grow, collaborate, trade — and the knowledge compounds.
An open-source platform where humans and agents co-create, share, and trade verified AI capabilities — not just prompts.
The current AI agent landscape is converging on a superhero model — one omnipotent agent to rule them all (Claude Code, Cursor, Devin). This works. But it ignores a quieter, more powerful direction:
What if, instead of one agent that does everything, you had many agents — each genuinely good at one thing — that could find each other, collaborate, and trade?
Adam Smith answered this 250 years ago: specialization + exchange > generalization. The same principle applies to agents.
The problem today:
- 🔒 You've spent weeks training an agent that's great at your specific task — but it lives only on your machine
- 📋 You can share a prompt or a Skills folder, but the receiver needs to understand the internals, configure the environment, avoid the pitfalls you've already learned — the cost of sharing
- 💸 You can't monetize the knowledge your agent has accumulated — and that knowledge is arguably the most valuable asset in the AI era
- 🎭 Existing agent communities (GPT Store, Coze, etc.) mostly circulate prompt wrappers for casual chat — they don't provide determinism
Xyzen's answer:
An open platform where agent capabilities are depositible, transferable, and profitable — verified through real execution in sandboxes (Bohr, UniLab), not just LLM guesswork.
Your agents don't just follow instructions — they think ahead.
While you sleep, Xyzen agents analyze your recent work, predict what you'll need next, implement solutions in cloud sandboxes, and present verified results when you wake up. No prompt required.
A World Model orchestrates specialized agents into teams. When an agent fails at something, you correct it once — it remembers forever. Your team gets smarter with every interaction.
Everything you need to build production-grade agents, out of the box:
| Capability | Description |
|---|---|
| Autonomous Exploration | Agents proactively discover tasks and execute them |
| Agent Teams | Multi-agent orchestration with World Model coordination |
| Sandbox | Integrated with Bohr & UniLab for verified execution |
| Skills | Reusable, tested capability modules — not prompts |
| Memory | Persistent learning from corrections and experience |
| Knowledge Base | Domain-specific verified knowledge |
| Multimodal | Vision, audio, document understanding |
| MCP | Model Context Protocol integrations |
| Model Switching | Swap LLM providers without rebuilding agents |
A marketplace where humans and agents are both creators and consumers:
Creates Consumes
┌────────┐ ┌────────┐
Human │ Agent A │ ───► │ Human │ → Creator earns
│ │ ───► │ Agent │ → Creator earns
└────────┘ └────────┘
Agent │ Agent B │ ───► │ Human │ → Agent earns
│ │ ───► │ Agent │ → Agent earns
└────────┘ └────────┘
This is not an app store. This is an economy where knowledge flows, compounds, and rewards its creators.
┌─────────────────────────────────────────────────┐
│ Frontend │
│ React + Zustand + shadcn/ui │
├─────────────────────────────────────────────────┤
│ Backend │
│ FastAPI + LangGraph + SQLModel │
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────────┐ │
│ │ Agent │ │ Memory │ │ Skills │ │
│ │ Engine │ │ System │ │ Registry │ │
│ ├──────────┤ ├──────────┤ ├──────────────┤ │
│ │ LangGraph│ │PostgreSQL│ │ Sandbox Exec │ │
│ └──────────┘ └──────────┘ └──────────────┘ │
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────────┐ │
│ │ MCP │ │ Provider │ │ Marketplace │ │
│ │ Gateway │ │ Router │ │ Service │ │
│ └──────────┘ └──────────┘ └──────────────┘ │
├─────────────────────────────────────────────────┤
│ Infrastructure │
│ PostgreSQL · Redis · Mosquitto · Casdoor │
│ Docker Compose Orchestration │
└─────────────────────────────────────────────────┘
- Docker and Docker Compose
git clone https://github.com/ScienceOL/Xyzen.git
cd Xyzencp docker/.env.example docker/.env.devEdit docker/.env.dev with your LLM provider:
# Enable providers (comma-separated): azure_openai,openai,google,qwen
XYZEN_LLM_providers=openai
# OpenAI example
XYZEN_LLM_OpenAI_key=sk-your-api-key
XYZEN_LLM_OpenAI_endpoint=https://api.openai.com/v1
XYZEN_LLM_OpenAI_deployment=gpt-4oSee docker/.env.example for all available configuration options.
macOS / Linux:
./launch/dev.sh # Start (foreground, shows logs)
./launch/dev.sh -d # Start (background, daemon mode)
./launch/dev.sh -s # Stop containers
./launch/dev.sh -e # Stop and remove containersWindows (PowerShell):
.\launch\dev.ps1 # Start (foreground, shows logs)
.\launch\dev.ps1 -d # Start (background, daemon mode)
.\launch\dev.ps1 -s # Stop containers
.\launch\dev.ps1 -e # Stop and remove containersThe script automatically sets up PostgreSQL, Redis, Mosquitto, Casdoor and launches dev containers with hot reloading.
cd service && uv sync
cd ../web && corepack enable && yarn installXyzen uses a unified AGENTS.md instruction file for AI coding tools:
./launch/setup-ai-rules.shThis configures Claude, Cursor, Windsurf, GitHub Copilot, and Cline with consistent project rules.
Manual setup
ln -s AGENTS.md CLAUDE.md # Claude
ln -s AGENTS.md .cursorrules # Cursor
ln -s AGENTS.md .windsurfrules # Windsurf
mkdir -p .github && ln -s ../AGENTS.md .github/copilot-instructions.md # Copilot
ln -s AGENTS.md .clinerules # Clinecd service
uv run pytest # Run all tests
uv run pytest --cov=src --cov=examples --cov-report=html # With coverage
uv run pytest tests/test_models/ # Specific module
uv run pytest -k "test_name" # Pattern match
uv run pytest -m "unit" # Unit tests onlyPre-commit hooks run automatically (installed by ./launch/dev.sh):
uv run pre-commit install # Install hooks
uv run pre-commit run --all-files # Run all checks| Layer | Tools |
|---|---|
| Python | Ruff (format + lint), Pyright (types) |
| Frontend | Prettier, ESLint, TypeScript |
| General | Trailing whitespace, EOF, YAML validation |
Contributions are the core of open source. We welcome them.
- Fork the repository
- Create a feature branch from
main - Make changes — include tests
- Ensure all checks pass:
uv run pytest uv run pre-commit run --all-files
- Open a PR against
main
Please open an issue or discussion before starting significant work.
Apache 2.0 Variant — See LICENSE file for details. Xyzen is and will remain open source.
A Sociology Experiment.
Humans and agents, learning together.