Xagent

<div align="center">

</div>

---

What is Xagent?

Describe tasks. Not workflows.

No more flowcharts. No more rigid automation. Just tell Xagent what you want.

👏 Join Telegram | Discord

❤️ Like Xagent? Give it a star 🌟 to support the development!

---

Xagent vs OpenClaw

| Capability | Xagent | OpenClaw | | ------------------ | ------------------------------------------------------------- | ----------------------------------------------------------------------------------------- | | Core design | Enterprise agent platform | Autonomous personal agent | | Security | VM‑level sandbox for safe agent execution | Only a Docker container sandbox — and real usage often requires elevated host permissions | | Agent architecture | LLM‑driven planning + efficient multi‑agent orchestration | Mostly linear task execution | | Model capability | API-compatible + model-native intelligence* | API or local models used as simple plug-ins (no deep integration) | | Knowledge systems | Enterprise RAG / knowledge platforms | Local memory / lightweight RAG | | Deployment | Flexible deployment — local, private (on‑prem), or cloud | Mostly local runtime | | Multi-tenancy | Tenant‑aware architecture | Primarily single‑user |

* Xagent supports both API-based models and open-source models, powered by deep integration with Xinference for model-level optimization.

---

🎬 See Xagent Think

Give it a goal. Watch it plan, select tools, execute, and deliver.

---

⚡ The Problem

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

---

🎯 The Xagent Way

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

---

🚀 What You Can Build

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.

Build Anything You Can Describe:

| 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.

---

🚀 Quick Start

Get started in 3 minutes

1️⃣ Clone and configure

git clone https://github.com/xorbitsai/xagent.git
cd xagent
cp example.env .env

2️⃣ Start with Docker

docker compose up -d

To enable sandbox (requires Linux or WSL2 with KVM support):

docker compose -f docker-compose-sandbox.yml up -d

3️⃣ Open in browser

http://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.

---

✨ Core Features

🧠 Dynamic Planning Engine

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.

---

🔌 Tool & Model Orchestration

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.

---

⚡ Instant Execution Mode

For simple use cases, run tool-enabled LLM calls instantly.

• No configuration overhead
• Chat-style assistants
• Embedded AI features

Start simple. Scale when needed.

---

📊 Observability & Control

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 in Action

Watch Xagent plan, execute, and deliver results in real-time.

---

Stay Ahead

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

---

🏢 Deployment Options

Xagent supports:

• Self-hosted deployment
• Private cloud environments
• On-premise enterprise infrastructure
• Docker-based setup

You control your models, data, and infrastructure.

---

🏗 Architecture Overview

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

---

🔍 Not a Workflow Builder

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

---

🤝 Contributing

We welcome developers, product builders, and researchers.

Open issues. Submit PRs. Help shape the future of AI agents.

---

💬 Community & Contact

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

---

📄 License

This project is licensed under the Xagent Source License - see the LICENSE file for details.