AIPyApp: The Complete Guide to Python-Use — The New AI Agent Paradigm (Agent 2.0)
AIPyApp introduces Python-Use, a radical new AI agent paradigm: No Agents, Code is Agent. Instead of Function Calling, Tools, MCP, or Workflows, the LLM uses Python directly as a universal interface to act on the real world. 3,860+ stars, by Knownsec.
What Is Python-Use?
Python-Use provides the entire Python execution environment to LLMs. The model sits in front of a Python interpreter — it types commands, executes them, observes results, and iterates. No plugins, no toolchains, no workflow editors.
To the user: Describe a task → AI executes it → Result returned.
The model autonomously understands, plans, writes, debugs, and executes code — and fixes bugs along the way. Code is just an internal implementation — the real deliverable is the result.
- Stars: 3,860+ ⭐
- Forks: 378
- Releases: 29
- Contributors: 10
- Author: Knownsec
- Homepage: aipy.app
The Problem: Agent 1.0 "Prosthetics"
Traditional AI agents (Agent 1.0) rely on:
- Function Calling
- Tools and plugins
- MCP Servers
- Workflows and orchestration
- Plugin-based clients
These external "prosthetics" create high entry barriers, heavy developer dependency, poor tool coordination, and code locked in cloud sandboxes.
The Solution: Agent 2.0 Philosophy
Core Principle: No Agents, Code is Agent
No Agents, No MCP, No Workflow, No Clients…
Code is Agent.
Python-Use discards all legacy layers. The model uses code to directly act on the environment:
| Capability | What It Means |
|---|---|
| Python use Data | Load, transform, analyze |
| Python use Browser | Automate the web |
| Python use Computer | Access file systems, local resources |
| Python use IoT | Control devices, embedded systems |
| Python use Anything | Code becomes universal interface |
This means:
- No MCP — code is the protocol
- No Workflow — model plans and executes on the fly
- No Tools — use existing Python ecosystem, no plugin registration
- No Agents — code replaces orchestration
Execution Mode: AI Think Do
Task → Plan → Code → Execute → Feedback → Loop
- Task — User describes intent
- Plan — Model decomposes and plans
- Code — Optimal Python strategy generated
- Execute — Direct interaction with environment
- Feedback — Output evaluated, looped back
No external agent needed. The AI completes the full loop independently.
Two Modes
Task Mode (Default)
Describe a task in natural language → AI plans, generates code, executes, returns result.
Python Mode (--python)
Interactive Python REPL powered by AI — write code with AI assistance, auto-install packages.
Single Entry Point: AiPy
One client, one interface. No plugin mess, no bloated UI wrappers. Just aipy.
Self-Evolution: Multi-Model Fusion
- Vision models for image/video understanding
- Speech models for listening and speaking
- Expert models for domain reasoning
- All fused under a unified AI control loop
Moving from chatbots to fully embodied AI agents.
AIPyApp vs Alternatives
Category: This is a Python-based AI agent execution paradigm.
| Feature | AIPyApp (Python-Use) | Open Interpreter | AutoGen | LangChain Agents |
|---|---|---|---|---|
| Focus | Code is Agent paradigm | Code interpreter | Multi-agent | Agent framework |
| Stars | 3.9K ⭐ | ~57K ⭐ | ~40K ⭐ | ~100K ⭐ |
| Philosophy | No Agents, No MCP | Open code execution | Agent orchestration | Tool chains |
| MCP Required | ❌ None | ❌ | ❌ | Optional |
| Plugins | ❌ Uses Python ecosystem | Minimal | Tools | Heavy tooling |
| Task Mode | ✅ Describe → Execute → Result | ✅ | ❌ | ❌ |
| Python REPL | ✅ Interactive | ✅ | ❌ | ❌ |
| Auto-Install Packages | ✅ | ✅ | ❌ | ❌ |
| Multi-Model Fusion | ✅ Vision + Speech + Expert | ❌ | ✅ | ✅ |
| Entry Point | Single: AiPy | Single CLI | Multi-agent setup | Code framework |
| Self-Debugging | ✅ | ✅ | ❌ | ❌ |
When to choose AIPyApp: You believe in the "Code is Agent" philosophy — no plugins, no tools, no orchestration — just Python as the universal interface.
When to choose Open Interpreter: You want a mature code interpreter with broad community support.
When to choose AutoGen: You need multi-agent orchestration for complex workflows.
When to choose LangChain: You need a comprehensive agent framework with extensive tool integrations.
Conclusion
AIPyApp's Python-Use is a paradigm shift: Agent 2.0. Instead of layering plugins, tools, and protocols on top of LLMs, it gives them Python itself — the most versatile programming language — as a direct interface to the real world. "No Agents, Code is Agent" isn't just a slogan; it's an architecture that eliminates the complexity of traditional AI agent stacks. With 3.9K stars, two execution modes, multi-model fusion, and the conviction that code is the only tool an AI needs, it's a bold vision for the future of AI agents.