Upsonic
Upsonic is a Python framework for building autonomous AI agents, from simple task-based agents to complex autonomous systems like OpenClaw. It provides a unified interface for creating agents, adding custom tools, and integrating OCR capabilities, solving the problem of rapid agent development. It is for developers who want to build AI agents in Python.
🎬 Introducing Autonomous Agents | OpenClaw-Like Agents in Python · Upsonic
✨ Key features
- Create autonomous agents with workspace restrictions and safety.
- Prebuilt autonomous agents ready to run from the community.
- Traditional agents with custom tool support via decorators.
- Unified OCR interface with multiple engine support.
- Integrate with MCP tools for external data sources.
- IDE integration with docs for Cursor, VSCode, Windsurf.
🎯 Use cases
- Analyze server logs and detect anomaly patterns.
- Build a stock analyst agent for market trends.
- Create a calculator agent with custom arithmetic tools.
- Extract text from invoices or documents using OCR.
- Develop autonomous agents for cloud execution with E2B.
📦 Installation
🧰 Requirements: Python environment with pip or uv; model API keys (e.g., Anthropic) for agent execution; optional OCR dependencies.
uv pip install upsonic
# pip install upsonic
For OCR support:
uv pip install "upsonic[ocr]"
🚀 Usage
from upsonic import AutonomousAgent, Task
agent = AutonomousAgent(
model="anthropic/claude-sonnet-4-5",
workspace="/path/to/logs"
)
task = Task("Analyze server logs and detect anomaly patterns")
agent.print_do(task)
⚠️ Good to know
All file and shell operations are restricted to the workspace; path traversal and dangerous commands are blocked.
❓ FAQ
How do I install Upsonic?
Use uv pip install upsonic or pip install upsonic. For OCR support, install with uv pip install "upsonic[ocr]".
What models are supported?
The examples use models like anthropic/claude-sonnet-4-5, but the README does not specify a full list; it likely supports various models via providers.
Can I add custom tools to an agent?
Yes, you can define custom tools using the @tool decorator and pass them to a Task's tools parameter.
What OCR engines are supported?
Supported engines include EasyOCR, RapidOCR, Tesseract, PaddleOCR, DeepSeek OCR, and DeepSeek via Ollama.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.