lingyuanli/MultiGen

What it solves

MultiGen provides a general-purpose AI agent system designed for fully private, on-premise deployment. It allows users to run powerful agents that can browse the web, execute code, and generate multimodal content without their data leaving their own infrastructure.

How it works

The system uses a two-stage agent architecture: a Planner agent decomposes a user's goal into a sequence of JSON sub-steps, which a ReAct agent then executes iteratively by reasoning and calling tools. To ensure security, all actions (shell commands, browser interactions, file operations) are executed within an isolated Docker sandbox (Ubuntu + Chrome + VNC).

Who it’s for

It is intended for users and organizations that need a self-hosted, private alternative to commercial agent platforms (like Claude Agent or GPT Agent) and want full control over their data, models, and execution environment.

Highlights

• Sandboxed Execution: Every action runs in a disposable Ubuntu container, preventing the model from accessing the host machine.
• MCP & A2A Native: Supports the Model Context Protocol (MCP) for external tools and Agent-to-Agent (A2A) protocols for delegating tasks to peer agents.
• Multimodal Generation: Built-in tools for creating images, videos, 3D models, and TTS audio (including multi-speaker podcasts).
• One-Command Deployment: Full stack (UI, API, Sandbox, Postgres, Redis) can be deployed via Docker Compose.
• LLM Agnostic: Compatible with any OpenAI-compatible LLM provider, including DeepSeek, vLLM, and Ollama.