A Self-Hosted Boss for Claude, Dify, and Agno

What it does

Wegent is a self-hostable web workspace that consolidates chat, coding, knowledge management, and task automation into a single interface. It acts as a control plane for multiple agent backends—Claude Code, Agno, Dify, and its own chat engine—dispatching work to sandboxes, cloud devices, or your local machine via WebSocket. You can also plug it into DingTalk or Telegram so the agents live where your team already talks.

The interesting bit

Rather than building yet another LLM wrapper, Wegent positions itself as an “AI-native operating system” with Kubernetes-style YAML CRDs to define bots, teams, and skills. It ships with out-of-the-box multi-agent collaboration patterns—Sequential, Parallel, Router, and Loop—which suggests the authors think in workflows, not just prompts.

Key highlights

• Supports a wide model zoo: Claude, OpenAI, Gemini, DeepSeek, and GLM, with multi-turn memory via mem0.
• Code mode wraps Claude Code in a cloud execution engine with GitHub/GitLab/Gitea/Gerrit integration and concurrent task runners.
• Feed mode adds cron-style and event-driven triggers, plus filtering logic like “only notify me if it will rain tomorrow.”
• Knowledge base handles uploads (txt, pdf, ppt, xls) and web imports, with a NotebookLM-style Q&A view.
• OpenAI-compatible API and MCP tool support for external integrations.

Caveats

• The “AI-native operating system” branding is ambitious for what appears to be a TypeScript/Next.js application orchestrating existing tools.
• The README is heavy on feature lists and quickstart commands, light on architecture depth or performance characteristics.
• Some features, like the local AI Device executor, require a separate install and authenticated WebSocket connection, adding deployment complexity.

Verdict

Teams that want a self-hosted alternative to scattered AI SaaS tools—and need both chat and code agents in one place—should look here. If you just want a lightweight CLI coding assistant, this is overkill.