llmschema 0.3

A structured response wrapper for LLMs using Pydantic.

llmschema

llmschema is a Python library that ensures structured and validated responses from LLMs (Large Language Models) like Ollama, OpenAI, and Gemini by enforcing user-defined Pydantic schemas. It abstracts model-specific quirks and guarantees responses in a safe, predictable, and JSON-compliant format.

🚀 Features

✅ Enforces Pydantic schema on LLM responses
✅ Works with multiple LLM providers (Ollama, OpenAI, Gemini, etc.)
✅ Handles malformed JSON responses gracefully
✅ Easy integration into existing applications
✅ Modular & scalable design

---

📦 Installation

Install llmschema via pip:

pip install llmschema

---

🛠 Usage

1️⃣ Define a Schema

from pydantic import BaseModel
from llmschema import SchemaManager, generate_response

class MyResponseSchema(BaseModel):
    text: str
    confidence: float

SchemaManager.set_schema(MyResponseSchema)

2️⃣ Generate a Response from an LLM

response = generate_response("mistral", "Summarize the latest AI news")
print(response)  # Output will follow MyResponseSchema format

3️⃣ Handling Errors

from llmschema import LLMValidationError

try:
    response = generate_response("gemini", "Give me a JSON response")
except LLMValidationError as e:
    print("Invalid response:", e)

---

⚙️ Supported LLMs

llmschema is designed to work with different LLM providers:

• ✅ Ollama (Mistral, Llama, etc.)
• ✅ OpenAI (GPT models)
• ✅ Gemini (Google's LLM)

More integrations coming soon!

---

✅ Handling Non-JSON Responses

If an LLM outputs invalid JSON, llmschema will:

1. Try to extract JSON using regex.
2. Log warnings for malformed responses.
3. Raise an error if parsing fails completely.

---

🧪 Running Tests

To test the library locally:

pytest tests/

---

📜 License

This project is licensed under the MIT License.

---

🤝 Contributing

Contributions are welcome! Feel free to submit issues and PRs on GitHub.

GitHub Repo: https://github.com/yourusername/llmschema