GenericAgent
A small, self-improving AI agent that controls your computer and saves successful steps as reusable skills.
- Stars
- 14.4k
- Forks
- 1.7k
- Licence
- MIT
- Language
- Python
- Last update
- 4 days ago
- Type
- AI model / agent
What it is
GenericAgent is an open-source Python framework that lets a large language model control a local computer through a small set of basic tools, including a browser, terminal, filesystem, keyboard and mouse, and screen reading. Each time it solves a new task, it stores the steps as a reusable skill, so its library of skills grows with use. The project says its core is about 3K lines of code and that it uses fewer tokens than many other agents, though the README's efficiency claims are not independently verified here.
- Automating browser tasks such as filling forms or navigating websites, as shown in the README demos
- Running terminal and filesystem tasks through natural-language instructions
- Sending batch messages or reading app screens through ADB on mobile devices
- Using the bundled terminal or Streamlit interfaces to chat with the agent
Great for
- Developers who want a small, readable agent codebase they can modify
- People experimenting with LLM-driven browser or desktop automation
- Users who want an agent to remember and reuse task workflows over time
- Tinkerers comfortable running Python and supplying their own LLM API key
Not ideal for
- Non-technical users expecting a polished one-click app, since setup needs Python and an API key
- Anyone needing strong guarantees about what an agent does on their machine, since it has broad system control
- Users on Python 3.14, which the README says is incompatible with some dependencies
How to get it
- 1
Clone and build from source
You'll need: Python 3.11 or 3.12 (not 3.14) · uv (used in the README install command) · Git · An LLM API key
Clone the repository and enter it
git clone https://github.com/lsdefine/GenericAgent.git && cd GenericAgentCreate a virtual environment and install with the UI extras
uv venv && uv pip install -e ".[ui]"Copy the config template, then add your LLM API key to it
cp mykey_template_en.py mykey.py - 2
Run it
- Fill in your LLM API key in mykey.py
- Run python frontends/tui_v3.py to open the terminal UI, which the README recommends
- Or run python launch.py for the Streamlit web UI
- Give the agent a simple task and watch how it works
- 3
Get the most out of it
- Use the terminal UI (tui_v3.py), which the README lists as the recommended interface
- Connect bot frontends such as Telegram, Discord or Lark using the scripts listed in the README
- Let the skill library build up over repeated tasks, since the project's main idea is that skills accumulate with use
- Read docs/installation.md for setup details beyond the quick start
Downloads
Strengths
- Core design is small (about 3K lines of seed code and a roughly 100-line agent loop), which makes it easier to read
- Supports several LLM providers named in the README, including Claude, Gemini, Kimi and MiniMax
- Dependencies are described as deliberately light, with no Playwright or LangChain required
Trade-offs
- Broad system control means it can take real actions on your computer, so it needs careful supervision
- The README's performance and token-savings figures come from the authors and should be checked for your own use
- Windows terminal rendering can be unreliable, and the README recommends Git Bash
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| GenericAgent (this) | 14.4k | AI model / agent | intermediate | 4 days ago | Developers who want a small, readable agent codebase they can modify |
| OpenHands | 92.5k | App | intermediate | 4 days ago | Developers who want coding agents running on a server while their laptop is off |
| browser-use | 120k | Library | intermediate | 4 days ago | Python developers who want to prototype an AI agent that operates websites |
| ego-lite | 21.3k | App | beginner | 4 days ago | Developers who use Claude Code, Codex, or similar agent CLIs and want them to act in a logged-in browser |
| camofox-browser | 12.1k | App | intermediate | 4 days ago | Developers building AI agents that need to read or interact with websites |
| hermes-agent | 260k | AI model / agent | intermediate | 4 days ago | Developers who want a personal agent they can host on their own server or machine |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.