textpy-llm 0.1.2

TextPy: Collaborative Agent Workflow through Programming and Prompting

TextPy ── Collaborative Agent Workflow through Programming and Prompting

🚀 Core Philosophy

P1: Workflow = Programming + Prompting
P2: Programming <-> Prompting
P3: Workflow = Function Signature with Comment

⚡ Revolutionary Features

• Mixed-Execution: Seamlessly integrates with legacy code
• AICompiler: Just write function signatures, our powerful compiler will automatically generate natural language instructions or code instructions just in time
• Self-Replication: Easier implementation of recursive workflows

🔧 Usage

Installation

$ pip install textpy-llm

In your python script, include the line

import textpy

Or use the image to run your code

$ docker pull ghcr.io/yezhengmao1/textpy:latest

🛠️ Quick Start - Simple Example

# examples/write_poem.py
from textpy import code, text

# the text/code will auto-generated by AICompiler(Just in Time)
# we use the litellm to call large language model, default model is deepseek/deepseek-chat
# if you want to use different large language model, @code(api_key="", base_url="", model="")

@text
def generate_poem(*, theme: str, style: str) -> str: ...

@code
# text should in center of the SVG, use a white background, add visually striking elements
def create_svg_from_text_with_svgwrite(*, text: str) -> str: ...

@code
def save_svg_to_file(*, svg: str, path: str): ...

# execute this code
poem = generate_poem(theme="Quantum Physics and Romance", style="haiku")
svg = create_svg_from_text_with_svgwrite(text=poem)
save_svg_to_file(svg=svg, path="poem.svg")

🤩 Quick Start - Awesome Example

Example: Download pdf from arxiv using the arxiv id, extract text from the papers, and summarize them. What makes it powerful is that it can recursively read all cited references and perform the same operations on them. Then it displays the relationships between them in an HTML file. When the script is running, visit localhost:31108 to check if the script is executing correctly, and visit localhost:31108/arxiv.html to view the results.

Highlight: All the functionalities in these workflows are generated by AICompiler in TextPy (including calling and installing other Python packages), while all you do is write function definitions (function signatures) and ... provide comments!!

NOTE: We strongly recommend running this script within a container, as our powerful AICompiler manipulates the external environment during compilation, which can affect your system. You can still run the script directly, refer to examples/arxiv_search.sh for details.

cd ./examples
DEEPSEEK_API=xxx CACHE_DIR="/workspace/.cache" ARXIV_ID=<xxxx.xxxx> ./arxiv_search.sh

Example Video

📌 Contributing

Community contributions are more than welcome, whether it be to fix bugs or to add new features at github.

📜 Citation

@misc{ye2025textpy,
  author = {Zhengmao Ye, Shuyun Yang, Zelong Huang, Jie Zuo, Mingjie Tang},
  title = {TextPy: Collaborative Agent Workflow through Programming and Prompting},
  year = {2025},
  howpublished = {\url{https://github.com/yezhengmao1/TextPy}}
}