Enterprise agent platform for building, deploying, and operating AI teammates.

Xagent
The enterprise platform for building and deploying AI teammates.
Open Xagent, describe a task in the workspace, and run it right away. For repeatable work, add a preconfigured teammate or build your own. Publish the teammate when you want other people to use it.
Try TobyBot Cloud · Self-host Xagent · Run a task · Build a teammate · Documentation
Why choose Xagent?
Once an agent starts serving other people, your organization needs to decide what it can access, where users can reach it, and how to investigate a bad result. Xagent gives you a workspace for getting work done and a way to publish and operate the teammates that take on recurring responsibilities.
- Start with a real request. Run a task directly from the workspace. In Auto mode, Xagent can answer directly, call tools, or plan and execute multiple steps based on the request.
- Reuse the same specialist. Use a published teammate directly from the workspace, or add it to a Workforce where a manager can delegate work to it. Configure the specialist's model, skills, tools, and knowledge in one place.
- Follow the request after launch. Let users reach a teammate through a link, a widget, or your own app. Find conversations by teammate and source, then inspect the transcript and any recorded tool calls for a particular request.
What is Xagent?
Xagent is a self-hostable platform for AI teammates. A teammate is a reusable agent configured for a specific job, with its own instructions, model, skills, tools, and knowledge.
Use a teammate directly in Xagent, or publish it for end users through a shareable link, an embedded widget, or an application connected via API.
What is TobyBot?
TobyBot is the commercial agent platform built on Xagent. Business teams use it to run tasks, build AI teammates for recurring work, and publish them for colleagues or customers. TobyBot Cloud is the hosted version of TobyBot.
How does Xagent work?
- Configure and test. Select the models and resources a teammate needs, connect your tools and knowledge, and test a sample request in the Builder.
- Publish and deliver. Choose where people will use the teammate. Enable a share link, embed it on an allowed domain, or call it from your application with an agent API key.
- Run and inspect. Use conversation logs and monitoring to see how teammates handle requests after you publish them.
Run a task
Enter a request in the workspace and send it. In this example, Xagent turns a few release notes into a customer update in the same conversation.

Build a teammate
Open Add a teammate to choose a preconfigured teammate or build your own. In the Builder, describe the job, choose a model, and attach the skills, tools, and knowledge the job calls for.
For a Product Support teammate, connect the product documentation and set instructions for answering setup, account, and billing questions. Try a question in Preview before publishing.

Skills package reusable instructions and resources. Knowledge bases supply reference material. MCP connectors bring external tools and systems into the teammate's workflow.
Deploy for end users
Publish the teammate, then choose an entry point that fits where your users already work.
| Entry point | How people use it |
|---|---|
| Shareable link | Visitors open a hosted chat page for the published teammate. Enable, reset, or disable the link from its deployment settings. |
| Embedded widget | Add chat to a site such as a help center and configure the allowed domains. |
| REST API | Call the teammate from your application or backend using an agent API key. |

Teammates can also run on a schedule or respond to webhook events. For jobs that require several specialists, create a Workforce with a manager agent and published teammates.
Operate Xagent
Use Conversation Logs to find a conversation by teammate or entry point. Open a run to review its transcript and execution steps, including tool calls when trace data is available. In the Product Support example, you can see the visitor's question, the reply, and the knowledge search behind it.

- Infrastructure. Self-host Xagent with Docker Compose and manage user accounts on your deployment.
- Models. Connect supported API providers or serve your own models through Xinference.
- Execution. Enable sandboxing for eligible tool calls and use monitoring to track model and tool usage.
See the deployment guide, release-specific deployment notes, and sandbox guide.
Run Xagent with TobyBot Cloud
Give TobyBot a goal. It plans the work, uses the tools you connect, and delivers the result. For work that repeats, build a teammate and run it on a schedule.
- Team workspaces. Invite colleagues, manage member roles, and share teammates, knowledge bases, and supported apps.
- Model choice. Use models hosted by Xinference or bring your own key for a supported provider or OpenAI-compatible endpoint.
- Guardrails and security auditing. Set team and teammate rules for reply review and use tool allowlists for client applications. Inspect run logs and traces. TobyBot records administrative access to task traces.
- Usage and retention. Track actions and AI credits against your plan limits. Set retention periods for conversations and execution records.
Quick start
Evaluate locally
The Python package includes the web UI. Use Python 3.11 or later.
uv tool install xagent-ai
xagent
- Open localhost:8000 and create the first administrator account at
/setup. - Connect a model provider on the Models page. To try Xagent right away, open New task and describe the work you want to run.
- Choose Add a teammate, then create or select a teammate and test it in the Builder.
- Publish the teammate, enable its share link, and open the link in a private browser window to try the visitor experience.
Moving from OpenClaw or Hermes? Use xagent migrate to import agent configuration, skills, and supported schedules. Run xagent migrate --dry-run to preview the import.
Run with Docker Compose
git clone https://github.com/xorbitsai/xagent.git
cd xagent
cp example.env .env
docker compose up -d
Open localhost and complete setup. Review the deployment settings in .env before running on a shared host. For production, serve Xagent over HTTPS and follow the deployment guide.
Forgot the administrator password? See how to reset it from the Docker host.
Try TobyBot Cloud
Sign in to TobyBot Cloud and give it your first task or add a teammate for a recurring job.
Documentation and community
Documentation · Releases · GitHub Issues · Discord · Telegram
License
Xagent is source-available under the Xagent Source License. See the license for terms covering deployment and hosted use.