AuC is a Python core framework for single agents within the ufy system, built upon asyncio. It offers a pluggable inference loop (defaulting to ReAct), LLM adaptation, tiered tool permissions (L1/L2/L3), and observable event streams. Designed with Claude Code-inspired engineering discipline, it provides context slicing, project rules (.aurules), and secondary authorization for high-risk operations, enabling developers to build sophisticated and secure agentic applications.

How It Works

AuC leverages asyncio for asynchronous operations, featuring a pluggable inference loop that defaults to the ReAct pattern. It supports multiple LLM providers (OpenAI, Anthropic, DeepSeek) and implements a granular tool permission system (L1 read-only, L2 sandboxed write, L3 requiring secondary authorization), automatically escalating dangerous commands. The framework incorporates advanced features like context compression, checkpoint rollback, and an "evolutionary memory" mechanism for experience recall and solidification, promoting robust and adaptable agent behavior.

Quick Start & Requirements

Installation is straightforward via pip:

• Core framework: pip install ufy-auc
• Recommended for terminal chat: pip install "ufy-auc[chat]"
• For Web UI: pip install "ufy-auc[web]"
• All components: pip install "ufy-auc[all]"

Configuration requires initializing LLM settings (auc config init) and setting environment variables for API keys (e.g., export OPENAI_API_KEY=sk-...). Python 3.11/3.12 is used for development and CI. Detailed documentation is available within the repository's docs/ directory.

Highlighted Details

• ReAct Loop: Supports streaming inference, parallel tool calls, context compression, and checkpoint rollback.
• Multi-Model Support: Integrates with OpenAI Chat Completions, Anthropic Messages, and DeepSeek-compatible gateways.
• Tool Tiering: L1 (read-only), L2 (sandboxed write), L3 (secondary authorization) with automatic escalation for dangerous commands.
• UIs: Offers both a rich Terminal UI (prompt-toolkit, slash commands, Plan/Autonomy modes) and a Web UI (Code+Chat dual modes, workspace, project preview/proxy, SSE chat).
• IM Authorization: Integrates Telegram and QQ (OneBot 11) for L3 approval workflows.
• Evolutionary Memory: Utilizes .auc/evolution.yaml for experience recall and "gold nugget" solidification.
• Sandboxing: Provides secure execution environments for file R/W, Shell commands, grep/glob, and URL fetching (with SSRF protection).

Maintenance & Community

The project roadmap indicates ongoing development for features like ConversationStore sharing and AuM productionization. CI is handled via GitHub Actions with Python 3.11/3.12 and a coverage threshold of >=75%. Extensive design and technical documentation are provided within the repository.

Licensing & Compatibility

AuC is released under the MIT license, which is permissive for commercial use and integration into closed-source projects.

Limitations & Caveats

The roadmap suggests that certain components, such as ConversationStore sharing and the productionization of the AuM repository, are still under active development or planned, indicating potential areas for future enhancement or stabilization.