A private ChatGPT workbench with CrewAI and 260+ data hooks

What it does

AgentCloud is a self-hosted platform for building private LLM chat apps that can read your company data. It provides a web interface for creating conversational assistants—essentially internal ChatGPT clones—and multi-agent process automations powered by CrewAI, wrapped in team permissions and access controls. A RAG pipeline pulls context from the project’s claimed 260-plus native data connectors so answers stay grounded in live databases and documents rather than generic training data.

The interesting bit

The architecture is split into three distinct tiers: a Python backend running CrewAI over socket.io, a Next.js frontend with an Express server, and a Rust vector proxy handling communication with Qdrant. That Rust middle layer is an unusual choice in the LLM-app space, where vector work usually gets folded directly into the Python backend.

Key highlights

• Self-hosted alternative to OpenAI’s GPTs with team permissions and user management
• RAG pipeline with native connectors for over 260 data sources, per the project description
• Multi-agent process automation via CrewAI, not just simple chat completions
• Three-tier stack: Python agent backend, Next.js/Tailwind UI, and Rust vector-db proxy for Qdrant
• Dockerized deployment with an install.sh script for Mac and Linux

Caveats

• Windows support is listed as “coming soon,” so local deployment is currently limited to Mac and Linux.
• The README is sparse on technical detail; most architecture specifics are condensed into a single introduction paragraph.
• Licensed under AGPL v3, which carries copyleft obligations for network use.

Verdict

Teams looking for a private, self-hosted ChatGPT workbench with RAG and multi-agent automation should take a look. Solo developers wanting a lightweight drop-in LLM UI will likely find the stack heavier than necessary.