xorbitsai/xagent

Build personal agents and enterprise AI workforces that plan, delegate, use tools, and deliver real work — without brittle workflows.

What it solves

Xagent is designed for knowledge work that is too messy or ambiguous for static workflow builders. Instead of requiring users to draw a flowchart of every possible branch, it allows users to describe a desired outcome and handles the dynamic planning and execution of the task at runtime.

How it works

Users provide a goal, and Xagent's planning engine decomposes the task into executable steps. It uses a "Plan $\rightarrow$ Execute $\rightarrow$ Reflect" loop to dynamically select tools, models, and knowledge bases. It can coordinate multiple specialized agents (such as Planner, Research, Data, and Writer agents) to complete complex deliverables.

Who it’s for

  • Individuals seeking a personal agent for one-off tasks like research, writing, and data exploration.
  • Teams who want to turn repeated operations into reusable, published agents with shared knowledge.
  • Enterprises requiring a private, self-hosted agent platform that integrates with internal systems, private clouds, and on-prem infrastructure.

Highlights

  • Dynamic Planning: Plans at runtime rather than following a static graph.
  • Enterprise-Ready: Supports self-hosting, multi-user control, and self-hosted models via Xinference.
  • Extensible Tooling: Integrates with APIs, MCP tools, files, and RAG-based knowledge bases.
  • Observability: Provides real-time tracking of task lifecycles, tool calls, and token usage.

Related

  • Project
  • Project
  • Project
  • Project