ufy2024/AuC
Agents-ufy-Core
What it solves
AuC (Agents-ufy-Core) is a Python-based single-agent framework designed to build autonomous agents with high engineering discipline. It addresses the need for safe, observable, and evolvable AI agents that can interact with local files, shells, and external APIs while maintaining strict control over high-risk operations.
How it works
The framework is built on asyncio and implements a ReAct (Reasoning and Acting) loop for streaming inference and parallel tool calling. It provides a tiered tool permission system (L1 Read-only, L2 Sandbox Write, L3 Secondary Authorization) to ensure security. High-risk commands are automatically upgraded to L3, requiring manual approval via integrated platforms like Telegram, QQ, or a Web UI.
Key technical components include:
- Au-Context Slicer: Manages context window efficiency through semantic slicing.
- Au-Rules Matrix: Implements project-specific rules via
.aurulesfiles. - Checkpointing: Allows users to undo file changes or command executions.
- Evolution Memory: Uses
.auc/evolution.yamlto recall experiences and solidify "gold nuggets" of knowledge. - Sandbox: Provides isolated environments for file I/O, shell execution, and URL fetching with SSRF protection.
Who it’s for
Developers and AI engineers who want to build agents that can autonomously manage codebases or perform system tasks with a safety-first approach, as well as users who want a ready-to-use CLI or Web UI for interacting with LLMs via a controlled agentic workflow.
Highlights
- Tiered Permissions: L1/L2/L3 authorization levels to prevent accidental or malicious system changes.
- Multi-Model Support: Native integration with OpenAI, Anthropic, and DeepSeek.
- Multi-Channel Approval: Secondary authorization requests sent to Telegram, QQ, or Web UI.
- Evolvable Memory: A mechanism for the agent to learn from its own execution and refine its behavior over time.
- Comprehensive UI: Includes both a professional terminal REPL and a FastAPI-based Web UI with workspace management.
- Safety-First Design: Built-in SSRF protection, path sandboxing, and Docker isolation support.
Related
- Project
- Project
- Project
- Project
- Project