# AetherAgent

**Repository Path**: yeshengmogu/aether-agent

## Basic Information

- **Project Name**: AetherAgent
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-05-01
- **Last Updated**: 2026-05-01

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# AetherAgent

AetherAgent is a lightweight, self-evolving autonomous agent framework for local computer control.

It combines a compact agent loop, layered memory, browser and system tooling, and reusable task crystallization. Instead of hardcoding every capability up front, AetherAgent learns by doing: when it finishes a new class of task, it can turn that execution path into a reusable skill or SOP for future runs.

## Highlights

- Lightweight architecture with a small Python core.
- Autonomous execution loop for planning, acting, checking, and iterating.
- Layered memory for reusable context, skills, and long-term task knowledge.
- Tool-based interaction with files, code, web pages, browsers, and local workflows.
- Extensible frontend architecture for desktop and chat-style interfaces.
- Compatible with multiple LLM backends through configurable session adapters.

## Core Design

AetherAgent is built around four ideas:

1. Minimal agent core: keep the runtime small and understandable.
2. Tool-driven execution: use atomic tools to interact with the environment.
3. Memory layering: separate stable rules, insights, reusable skills, and archived sessions.
4. Self-evolution: convert solved tasks into reusable operational knowledge.

## Project Structure

- `agentmain.py`: main runtime and session orchestration.
- `agent_loop.py`: compact autonomous loop.
- `ga.py`: handler layer and tool wiring.
- `llmcore.py`: model sessions and backend integration.
- `frontends/`: Streamlit, Qt, bot, and desktop frontend entry points.
- `memory/`: memory files, SOPs, and skill-related utilities.
- `reflect/`: scheduled and reflective automation scripts.
- `assets/`: prompts, tool schemas, images, and helper resources.

## Quick Start

### Standard install

```bash
git clone https://gitee.com/yeshengmogu/aether-agent.git
cd aether-agent
pip install requests streamlit pywebview
cp mykey_template.py mykey.py
python launch.pyw
```

### Using `uv`

```bash
git clone https://gitee.com/yeshengmogu/aether-agent.git
cd aether-agent
uv pip install -e ".[ui]"
cp mykey_template.py mykey.py
python launch.pyw
```

## Frontends

Available frontend entry points include:

- `python launch.pyw`: default desktop launcher.
- `streamlit run frontends/stapp_clean.py`: clean Streamlit chat UI.
- `python frontends/qtapp.py`: Qt desktop frontend.
- Additional bot frontends in `frontends/` for Telegram, QQ, Feishu, WeCom, and DingTalk.

## How It Works

At runtime, AetherAgent follows an agent loop:

`observe -> reason -> call tools -> verify -> write memory -> continue`

Its layered memory system helps keep context useful without bloating the working window:

- `L0`: meta rules and operating constraints
- `L1`: compact insight index
- `L2`: global memory and durable facts
- `L3`: task skills and SOPs
- `L4`: archived session summaries for long-horizon recall

## Status

This repository contains the local framework, frontends, memory assets, and integration code for running AetherAgent on your machine.

## License

MIT. See [LICENSE](LICENSE).