AI Agent Scaffolding 'CerebrumKit' Released

The open-source framework CerebrumKit has been released to address repetitive scaffolding tasks in AI agent development. This project provides an admin panel for configuring agents, connecting business data as tools, and a client panel for end-user interaction.

Key Features and Architecture

• Database-based configuration: Skills, tools, and workflows are stored as database rows for flexible management.
• Tool execution method: Tool bodies are stored as Python code and executed with full builtins permissions without sandboxing, allowing the use of requests, SQLAlchemy, etc. Therefore, tool creation is restricted to administrators, and security precautions are necessary.
• Memory management: Facts are conveyed through tools and memory objects rather than re-injecting chat history into prompts. Memory is stored with (agent, user) keys, ensuring information persists across different chats within the same agent.

Tech Stack and Deployment

• Backend: FastAPI, SQLAlchemy, PostgreSQL (SQLite also supported)
• Frontend: Vue 3-based admin and client panels (supports 6 languages)
• Deployment: Designed to run as a single uvicorn worker in an Nginx + systemd environment
• License: Apache-2.0 (includes patent clause)

This project provides the essential infrastructure needed to build AI agents in enterprise environments and is available on GitHub.