wecode-ai/Wegent

What it solves

Wegent provides a self-hostable platform for teams to build, manage, and deploy AI agent teams. It bridges the gap between high-level agent orchestration (web-based management) and local execution (coding and system access), allowing agents to perform real-world work across chat, coding, and automation while maintaining control over where the code and data reside.

How it works

The system is split into two primary interfaces: Wegent Web for creating and sharing agents, managing knowledge bases, and handling permissions, and Wework, a desktop workbench that allows agents to interact directly with local files, terminals, and development environments.

Technically, it uses a modular agent model where a "Bot" is composed of a Ghost (prompts, skills, MCP), a Shell (runtime like ClaudeCode or Dify), and a Model. Multiple Bots can be organized into a "Team" to collaborate on complex tasks. The architecture includes a FastAPI backend, a Next.js frontend, and a dedicated executor manager to handle tasks across cloud, containers, or local devices.

Who it’s for

• Development Teams looking for AI coding assistants that can actually commit code and run tests in local or isolated environments.
• Enterprises needing private, self-hosted AI assistants with grounded knowledge from internal documents.
• Automation Engineers wanting to build scheduled agents that track information and analyze webpages.

Highlights

• Hybrid Execution: Run tasks in the cloud, in containers, or directly on local devices/private networks.
• Collaborative Agents: Organize multiple bots to divide research, analysis, and review work.
• Local Integration: Through Wework, agents can use local CLIs, browsers, and development environments.
• Extensible Integration: Connects to existing systems via APIs, MCP, and IM bots.