Dynamic Pydantic is the easiest way for AI agents to create and validate tools and databases in runtime.
It is a provides a powerful approach to creating pydantic models dynamically for any given task.
The project leverages Instructor and Pydantic to generate models dynamically based on provided prompt and context.
uv pip install dynamic-pydanticStart contributing by cloning the repository:
git clone https://github.com/lukafilipxvic/dynamic-pydantic.gitThis package simplifies the interaction between language models and pydantic's create_model() function.
Below is a basic example showcasing the generation of a Pydantic schema with Cerebras' Llama 3.3 70b inference.
Dynamic Pydantic works with any OpenAI-compatible endpoint via LiteLLM.
from dynamic_pydantic import dynamic_model
genModel = dynamic_model(prompt='User = Name, Age')
print(f'{genModel.schema_json()}')
# {"properties": {"Name": {"default": null, "description": "The user's name", "title": "Name", "type": "string"}, "Age": {"default": null, "description": "The user's age", "title": "Age", "type": "integer"}}, "title": "User", "type": "object"}
This project is licensed under the terms of the MIT license.
For more details, refer to the LICENSE file in the repository.