proteus-ai
by alishangtian
Overview
A multi-agent workflow orchestration system capable of generating and executing complex workflows, interacting with external MCP (Model Context Protocol) servers for tools and resources, and managing conversational agents and SOP memories.
Installation
docker build -t proteus:multistage -f Dockerfile.multistage . && docker run -d -p 8000:8000 --name proteus-ai --restart unless-stopped -e API_KEY=YOUR_LLM_API_KEY -e SANDBOX_HOST=http://your_sandbox_ip -e SANDBOX_PORT=8000 -e SANDBOX_API_KEY=YOUR_SANDBOX_API_KEY -e REDIS_HOST=your_redis_ip -e REDIS_PORT=6379 -e REDIS_PASSWORD=YOUR_REDIS_PASSWORD proteus:multistageEnvironment Variables
- API_KEY
- MODEL_NAME
- LONG_CONTEXT_MODEL
- BASE_URL
- SERPER_API_KEY
- CAIYUN_TOKEN
- CAIYUN_API_VERSION
- DATA_PATH
- CRYPTO_SECRET_KEY
- REDIS_HOST
- REDIS_PORT
- REDIS_DB
- REDIS_PASSWORD
- SESSION_MODEL
- SANDBOX_HOST
- SANDBOX_PORT
- SANDBOX_API_KEY
- BROSWER_EXECUTABLE_PATH
- BROSWER_USE_MODEL
- BROSWER_USE_API_KEY
- MYSQL_HOST
- MYSQL_USER
- MYSQL_PASSWORD
- MYSQL_DATABASE
- LANGFUSE_PUBLIC_KEY
- LANGFUSE_SECRET_KEY
- LANGFUSE_HOST
- MCP_CONFIG_PATH
Security Notes
CRITICAL: The `SANDBOX_API_KEY` in `src/nodes/python_execute.py` has a hardcoded default value ('El0/osJhMJnaQMCYiyOAOD4WGgJb4vbiMhQgf7g1DXgHxz10KuWodvQr'). This is a severe vulnerability as it allows anyone with access to this default key to execute arbitrary code within the sandbox environment if not changed immediately upon deployment. Additionally, `MysqlNode` generates SQL queries based on LLM input, which introduces a risk of SQL injection through carefully crafted prompts. The system extensively relies on interaction with external services (LLMs, Serper, Caiyun, MCP servers, and a separate sandbox service), expanding the attack surface. Dynamic module imports in `ToolExecutor` and `NodeConfigManager` require careful management of the Python environment to prevent arbitrary code loading. The `BrowserAgentNode` also introduces risks associated with running a full browser. While some security practices like password hashing for user authentication and environment variable usage for most secrets are present, the identified hardcoded key significantly lowers the overall security posture.
Similar Servers
MaxKB
An enterprise-grade agent platform for AI knowledge management, chatbot development, and complex AI workflow orchestration.
npcpy
A flexible framework for building state-of-the-art natural language processing pipelines and agent tooling, conducting novel research with LLMs, and integrating AI models into daily workflows for diverse tasks including multi-agent orchestration, fine-tuning, and multimedia generation.
mcpstore
MCPStore acts as an orchestration layer for managing Microservice Context Protocol (MCP) services and adapting them as tools for AI frameworks like LangChain, AutoGen, and others.
mcp_massive
An AI agent orchestration server, likely interacting with LLMs and managing multi-agent workflows.