GenericAgent

A minimal, self-evolving autonomous agent framework with ~3K lines of core code. It provides 9 atomic tools + ~100 line Agent Loop, granting any LLM system-level control over a local computer.

Overview

GenericAgent’s core philosophy: don’t preload skills — evolve them.

• Self-Evolving: Automatically crystallizes each task into a skill. Capabilities grow with every use, forming a personal skill tree
• Minimal Architecture: ~3K lines of core code, Agent Loop is ~100 lines
• Strong Execution: Injects into real browser (preserving login sessions)
• High Compatibility: Supports Claude / Gemini / Kimi / MiniMax
• Token Efficient: <30K context window (6x less than other agents)

Core Components

1. Layered Memory System

Layer	Purpose
L0 — Meta Rules	Core behavioral rules and system constraints
L1 — Insight Index	Minimal memory index for fast routing and recall
L2 — Global Facts	Stable knowledge accumulated over long-term operation
L3 — Task Skills / SOPs	Reusable workflows for specific task types
L4 — Session Archive	Archived task records for long-horizon recall

2. 9 Atomic Tools

Tool	Function
code_run	Execute arbitrary code
file_read	Read files
file_write	Write files
file_patch	Modify files
web_scan	Perceive web content
web_execute_js	Control browser behavior
ask_user	Human-in-the-loop confirmation
update_working_checkpoint	Persist context across sessions
start_long_term_update	Accumulate experience

3. Autonomous Execution Loop

[New Task] → [Autonomous Exploration] → [Crystallize into Skill] → [Write to Memory] → [Direct Recall on Next]

Key Features

• Browser Control: Real browser injection with session preservation
• OS Control: Mouse/keyboard, screen vision, ADB for mobile
• Self-Bootstrap: Can autonomously install deps, write scripts, debug & verify
• Skill Tree: Grows organically with usage — unique to each user

Comparison

Feature	GenericAgent	OpenClaw	Claude Code
Codebase	~3K lines	~530K lines	Large
Deployment	pip + API Key	Multi-service	CLI
Browser	Real (preserved session)	Sandbox	MCP
OS Control	Mouse/kbd, vision, ADB	Multi-agent	File + terminal
Self-Evolution	Autonomous skill growth	Plugin ecosystem	Stateless

Latest Updates

• 2026-04-21: Technical Report on arXiv (GenericAgent: Token-Efficient Self-Evolving LLM Agent)
• 2026-04-11: L4 session archive memory + cron scheduler
• 2026-03-23: WeChat bot frontend support
• 2026-03-10: Million-scale Skill Library

Quick Start

git clone https://github.com/lsdefine/GenericAgent.git
cd GenericAgent
pip install requests streamlit pywebview
cp mykey_template.py mykey.py
# Edit mykey.py with your LLM API Key
python launch.pyw

Bot Interfaces

• Streamlit Web UI (default)
• Telegram Bot
• WeChat Bot
• QQ Bot
• Feishu (Lark)
• Enterprise WeChat
• DingTalk
• Qt Desktop App

Related Links

• arXiv Technical Report
• Tutorial (Datawhale)
• Community Groups

See Also

• openclaw — Multi-service orchestration agent (~530K lines)
• moc-claude-code — CLI-focused AI coding agent
• 2026-05-11-agentmemory — agentmemory — #1 Persistent memory for AI coding agents
• 2026-03-22-supermemory — Supermemory — AI 메모리 & 컨텍스트 엔진 (LongMemEval 3관왕)
• moc-ai-agents-memory — AI 에이전트 메모리 MOC
• ai-agents — AI agent frameworks comparison