AA

aetherheart-ai/aeloon

Developer tools
132 stars 품질 40 트렌드 40

Aeloon is a lightweight, high-performance, and modular AI agent framework featuring graph-driven parallel acceleration and flexible hot-swappable plugins for many usages (Deep research, Wiki,...

개요

Aeloon is a lightweight, high-performance, and modular AI agent framework featuring graph-driven parallel acceleration and flexible hot-swappable plugins for many usages (Deep research, Wiki,...

README

English | 中文


What is Aeloon?

Aeloon is a lightweight, high-performance, and modular AI agent framework featuring graph-driven parallel acceleration and flexible hot-swappable plugins for many usages (Deep research, Wiki, SkillGraph, ACP Bridge and StatusPannel).

Core philosophy: one agent, all platforms, self-hosted data, privacy first.


✨ Key Features

Feature Description
🔄 Message Bus Architecture Channel → MessageBus → Dispatcher → AgentLoop → Tool execution → Routed reply — decoupled and extensible
📡 Multi-Channel Telegram, Feishu, DingTalk, WeCom, Slack, Discord, QQ, Email, WhatsApp, Matrix, WeChat
🧠 Smart Context Automatic context building, memory consolidation, skill injection, sub-agent delegation
🔧 Built-in Tools FileSystem, Web search/fetch, Shell exec, Cron, MCP protocol, Message sending
🤖 Multi-Provider OpenAI, Anthropic, DeepSeek, Gemini, Zhipu, Moonshot, MiniMax, Groq, Ollama, vLLM, Azure OpenAI, OpenRouter, custom endpoints
📝 Skill System Pure Markdown-driven — extend agent capabilities with zero code
🔌 MCP Protocol Built-in MCP client to connect external tool servers
🔧 Plugin SDK Complete plugin development framework: commands, tools, services, middleware, status bar, lifecycle management
🌉 ACP Bridge Built-in ACP Bridge plugin connecting to external agent ecosystems via ACP protocol
📊 Task Graph Tool calls compiled into a DAG — read-only ops run concurrently, writes serialized conservatively
🔒 Security First Network safety policies, execution sandbox, API key isolation
🐳 Docker Ready One-command docker compose up for gateway mode

🚀 Quick Start

One-Click Install

curl -fsSL https://raw.githubusercontent.com/AetherHeartAI/Aeloon/main/install.sh | bash

The installer walks you through provider and channel setup interactively. Offline and source installs are also supported:

# Install from local source
bash install.sh --from-source

# Install a specific version
bash install.sh --version v1.0.0

# Offline mode (requires local wheel)
bash install.sh --offline

Manual Install

# Install from source
git clone https://github.com/AetherHeartAI/Aeloon.git
cd Aeloon
pip install -e .

# First-time config
aeloon onboard

Build a wheel

# Output to dist/ by default
bash scripts/build_wheel.sh

# Use a custom directory and clear old artifacts
bash scripts/build_wheel.sh ./artifacts --clear

First Conversation

# One-shot CLI chat
aeloon agent -m "Hello Aeloon."

# Start gateway mode (connects all channels)
aeloon gateway

# Check channel status
aeloon channels status

🔌 Channel Support

WeChat and Feishu support QR scan login out of the box (/wechat login, Feishu in-app auth).

Other channels: Telegram, DingTalk, WeCom, Slack, Discord, QQ, Email, WhatsApp, Matrix.

All channels inherit BaseChannel and are auto-discovered — drop a file in aeloon/channels/ and it works, no registration needed.


🤖 Provider Support

Major LLMs with one-click setup: OpenAI, Anthropic, DeepSeek, Gemini, Zhipu, Moonshot, MiniMax, Groq, OpenRouter (free tier included), Azure OpenAI, Ollama, vLLM, plus OAuth-based OpenAI Codex / GitHub Copilot.

Selection priority: explicit name → model keyword match → API key prefix → API base match.


🔧 Tool System

Tool Module Function
FileSystem filesystem.py Read/write files, list directories, search
Web web.py Search (DuckDuckGo), fetch pages, extract content
Shell shell.py Execute commands (with safety policy)
Cron cron.py Create/list/remove scheduled tasks
MCP mcp.py Connect to external MCP tool servers
Message message.py Cross-channel message sending
Policy policy.py Tool execution policy control
Spawn spawn.py Sub-agent delegation

📝 Skill System

Pure Markdown — extend agent capabilities with zero code. Drop a SKILL.md into .aeloon/skills/ and it’s injected into the system prompt. Built-in skills cover scheduling, doc conversion, GitHub, memory, summarization, terminal, and more.


🔌 Plugin System

Plugins are modular components that extend Aeloon’s core: custom slash commands, agent tools, background services, message middleware, status bar contributions, and isolated config/storage namespaces.

Plugin types:

  • Task Plugin (commands + middleware) — request-response workflows, e.g. /sr, /se
  • Hybrid Plugin (commands + tools + services) — long-running agents, e.g. market monitoring
  • Service / Status Plugin (services or status bar contributions) — service-style extensions or CLI status bar contributions, e.g. StatusPannel

Built-in plugins:

Plugin Type Description
ScienceResearch Task AI4S research workflow: literature search, ArXiv, risk assessment, orchestrator, structured output
SkillGraph Task Skill dependency graph compilation and visualization
Wiki Hybrid Local knowledge base: content ingestion, smart digest, conversational knowledge enhancement, knowledge graph
ACP Bridge Hybrid External ACP agent bridging — connect third-party Agent protocols
StatusPannel Service CLI status bar: model name, context token usage

Lifecycle: Discover → Validate → Register → Commit → Activate → Running → Deactivate

Five-step dev paradigm: Extend Plugin → Implement register(api) → Implement activate() → Write handlers → Create aeloon.plugin.json manifest.

Full development guide: aeloon/plugins/README.md.


🐳 Docker Deployment

# Gateway mode
docker compose up

# CLI interactive mode
docker compose run --rm aeloon-cli

Default docker-compose.yml configuration:

  • Gateway port: 18790
  • Config mount: ~/.aeloon:/root/.aeloon
  • Resource limits: 1 CPU / 1GB memory

⚙️ Configuration

Config lives at ~/.aeloon/config.json and supports camelCase aliases. Loading precedence:

--config flag  >  AELOON_CONFIG env var  >  ~/.aeloon/config.json

Minimal example:

{
  "agents": {
    "defaults": {
      "provider": "openrouter",
      "model": "deepseek/deepseek-r1:free"
    }
  },
  "providers": {
    "openrouter": {
      "apiKey": "sk-or-v1-..."
    }
  }
}

🧪 Development

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Run specific tests
pytest tests/test_kernel.py
pytest -k "telegram"

# Lint + auto-fix
ruff check --fix .

# Format
ruff format .

Requirements: ruff==0.15.6 (CI enforced), Python ≥ 3.11, type hints on all public functions, async/await preferred for I/O.

Commit Convention

Uses Conventional Commits:

(): 

Types: feat, fix, chore, docs, test, refactor, perf, ci


🔒 Security

  • Network policy: _network_safety.py restricts reachable URLs to prevent SSRF
  • Execution sandbox: Shell tool uses policy controls for allowed commands
  • API key isolation: Keys stored in config, never logged
  • Channel auth: Each channel supports allowFrom whitelisting
  • MCP security: MCP connections can configure allowed tool scopes

📜 License

MIT License © 2026 Aether Heart contributors


View this README on GitHub

추천 도구

다른 키워드를 입력하거나 필터를 제거해 보세요.

설치

npx skillfish add aetherheart-ai/aeloon