Describe tasks. Not workflows.
No more flowcharts. No more rigid automation. Just tell Xagent what you want.
❤️ Like Xagent? Give it a star 🌟 to support the development!
Give it a goal. Watch it plan, select tools, execute, and deliver.
Workflow builders are rigid. They break when requirements change.
- You map every decision branch manually
- You orchestrate tools by hand
- You maintain fragile flow diagrams
- You re-engineer when logic changes
With Xagent, you describe the outcome — not the steps.
- Plans the task dynamically
- Decomposes into executable steps
- Selects the right tools automatically
- Executes, evaluates, and iterates
With Xagent, you don’t design workflows. You describe the task. That’s it.
Just define what you want done — and Xagent plans, decomposes, selects tools, and executes.
| Domain | Examples |
|---|---|
| Content Creation | Automated PPT generation, marketing posters, creative design assets |
| Research & Analysis | Deep analysis reports, research summaries, data synthesis |
| Enterprise Automation | AI copilots for internal teams, knowledge assistants, task automation |
| Business Intelligence | Data reporting automation, SaaS features, multi-step reasoning systems |
| Knowledge Work | Document processing, information extraction, structured outputs |
If you can clearly describe the goal, Xagent can turn it into an executable system.
All powered by one unified runtime.
Get started in 3 minutes
git clone https://github.com/xorbitsai/xagent.git
cd xagent
cp example.env .envdocker compose up -dhttp://localhost:80
On first startup, Xagent redirects to /setup.
Create the first administrator account there to complete initialization.
If the admin password is forgotten, reset it via CLI:
python -m xagent.web.reset_admin_password --username <admin_username>That's it. Xagent is now running.
Unlike traditional workflow tools, Xagent plans tasks dynamically at runtime.
- Automatic task decomposition
- Plan → Execute → Reflect loops
- Conditional branching
- Multi-step reasoning
No static flows. No brittle chains.
Xagent connects to your entire stack:
- OpenAI, Anthropic and other LLM providers
- Self-hosted models via Xinference
- External APIs
- Knowledge bases (RAG)
- Internal enterprise systems
It selects and orchestrates tools automatically during execution.
For simple use cases, run tool-enabled LLM calls instantly.
- No configuration overhead
- Chat-style assistants
- Embedded AI features
Start simple. Scale when needed.
Built for real production use:
- Task lifecycle tracking
- Token usage monitoring
- Execution state management
- Multi-user support
Operate agents like real systems — not demos.
Xagent is actively developed and rapidly evolving.
Follow our progress:
- ⭐ Star us on GitHub to stay updated
- 🐛 Report issues and request features
- 💬 Join our community discussions
Xagent supports:
- Self-hosted deployment
- Private cloud environments
- On-premise enterprise infrastructure
- Docker-based setup
You control your models, data, and infrastructure.
Xagent separates core responsibilities:
| Layer | Responsibility |
|---|---|
| Agent Definition | Intent & constraints |
| Planning Engine | Dynamic decomposition |
| Execution Runtime | Orchestration layer |
| Tool Layer | Integrations & actions |
| Model Layer | LLM & inference backend |
This architecture enables:
- Stability under complex reasoning
- Safe iteration
- Horizontal scalability
- Long-term maintainability
Xagent is not:
- A drag-and-drop flow editor
- A static template engine
- A chatbot wrapper
Xagent is:
- A dynamic task execution engine
- An autonomous planning system
- A foundation for building real AI agents
We welcome developers, product builders, and researchers.
Open issues. Submit PRs. Help shape the future of AI agents.
Documentation - Full documentation and guides
GitHub Issues - Report bugs or propose features
Discord - Share your tasks or agents and connect with the community
Telegram - Join our Telegram group for discussions
X (Twitter) - Follow for updates and share your work
This project is licensed under the Xagent Source License - see the LICENSE file for details.


