OpenHosta 4.3.1

A lightweight library integrating LLM natively into Python

OpenHosta

The semantic layer for Python.
Write what you mean. Python does the rest.

---

OpenHosta integrates Large Language Models directly into Python as native functions. Define a function with type hints and a docstring — OpenHosta uses AI to implement it. No DSL, no wrappers, just Python.

from OpenHosta import emulate

def translate(text: str, language: str) -> str:
    """Translates the text into the specified language."""
    return emulate()

print(translate("Hello World!", "French"))
# 'Bonjour le monde !'

OpenHosta also enables semantic testing — evaluate conditions that require cultural knowledge or fuzzy logic, something traditional assert statements can never do:

from OpenHosta import test

sentence = "You are an nice person."

if test(f"this contains an insult: {sentence}"):
    print("The sentence is considered an insult.")
else:
    print("The sentence is not considered an insult.")
# The sentence is not considered an insult.

Why OpenHosta?

• Zero DSL — Pure Python syntax. Your functions stay readable, testable, and IDE-friendly.
• Type-safe — Guarded types validate LLM output against your annotations (int, dict, Enum, Pydantic, Callable…).
• Model-agnostic — Works with OpenAI, Ollama, Azure, vLLM — any OpenAI-compatible endpoint.
• Runs offline — Full local execution with Ollama. Your data stays private.
• Production-ready — Uncertainty tracking, cost tracking, audit mode, and async support built-in.

Installation

pip install OpenHosta

We recommend using a virtual environment (python -m venv .venv). See the full installation guide for local model setup, optional dependencies, and troubleshooting.

Quick Start

Option A: Local Execution (Ollama)

Ensure you have Ollama installed and run ollama run qwen3.5:4b in your terminal.

from OpenHosta import emulate, OpenAICompatibleModel, config

# 1. Point OpenHosta to your local Ollama instance
local_model = OpenAICompatibleModel(
    model_name="qwen3.5:4b",
    base_url="http://localhost:11434/v1",
    api_key="none"  # Ollama does not require a key
)
config.DefaultModel = local_model

# 2. Define and call your function
def translate(text: str, language: str) -> str:
    """Translates the text into the specified language."""
    return emulate()

print(translate("Hello World!", "French"))
# 'Bonjour le monde !'

Option B: Remote API (OpenAI)

Create a .env file in your project directory:

OPENHOSTA_DEFAULT_MODEL_NAME="gpt-4.1"
OPENHOSTA_DEFAULT_MODEL_API_KEY="your-api-key-here"

from OpenHosta import emulate

def translate(text: str, language: str) -> str:
    """Translates the text into the specified language."""
    return emulate()

print(translate("Hello World!", "French"))
# 'Bonjour le monde !'

What Can You Do?

Feature	Description
emulate	AI-implemented functions from docstrings
emulate_async	Non-blocking async variant for concurrency
emulate_variants	Streaming results via lazy generators
closure	Semantic lambda functions
test	Fuzzy logic / semantic boolean tests
Types & Pydantic	int, dict, Enum, dataclass, Pydantic, Callable…
Safe Context	Uncertainty tracking & error handling
Image input	Pass PIL.Image directly to functions

📖 Full Documentation · 📝 Changelog · 🧪 Examples

Contributing

We warmly welcome contributions! Please refer to our Contribution Guide and Code of Conduct.

Browse existing issues to find contribution ideas.

License

MIT License — see LICENSE for details.

Authors

• Emmanuel Batt — Manager and Coordinator, Founder of Hand-e
• William Jolivet — DevOps, SysAdmin
• Léandre Ramos — AI Developer
• Merlin Devillard — UX Designer, Product Owner

GitHub: https://github.com/hand-e-fr/OpenHosta

---

The future of development is human. — The OpenHosta Team