united-llm 0.1.6

A streamlined, lightweight LLM client library for OpenRouter and Ollama providers with structured output generation

United LLM

A streamlined, lightweight LLM client library that provides unified access to OpenRouter and Ollama providers with structured output generation, multi-image vision processing, and schema prompt enhancement.

✨ Key Features

• 🤖 Three Generation Methods: gen_text(), gen_dict(), gen_pydantic()
• 🔌 Dual Provider Support: OpenRouter (cloud) + Ollama (local)
• 👁️ Multi-Image Vision: Process multiple images simultaneously
• 🎯 Schema Enhancement: Optional prompt enhancement for better compliance
• 🔗 String-Schema Integration: Automatic validation and conversion
• 📝 Comprehensive Logging: Daily organized logs with detailed tracking
• ⚙️ Simple Configuration: JSON-based configuration
• 🔍 Provider Auto-Detection: Automatic provider selection from model prefixes

📦 Installation

pip install united-llm

Requirements

• Python: 3.8+
• Dependencies: instructor>=1.3.0, pydantic>=2.0.0, string-schema>=0.1.0, openai>=1.0.0

🚀 Quick Start

1. Configuration Setup

Copy the example configuration and add your API keys:

cp united_llm.json.example united_llm.json

Edit united_llm.json:

{
  "api_keys": {
    "openrouter": "your-openrouter-api-key-here",
    "ollama": null
  },
  "base_urls": {
    "openrouter": "https://openrouter.ai/api/v1",
    "ollama": "http://localhost:11434/v1"
  },
  "log_dir": "logs/llm_calls"
}

2. Basic Usage

from united_llm import LLMClient
from pydantic import BaseModel
from typing import List

# Initialize client (auto-loads united_llm.json)
client = LLMClient()

# 1. Text Generation
text = client.gen_text(
    model="ollama:qwen3:8b",
    prompt="Explain quantum computing in simple terms."
)

# 2. Dictionary Generation with Schema Enhancement
data = client.gen_dict(
    model="ollama:qwen3:8b",
    prompt="Extract: John Smith, 30, Engineer in Boston",
    schema="{name: str, age: int, job: str, city: str}",
    add_schema_to_prompt=True  # 🎯 Enhanced compliance!
)
# Output: {"name": "John Smith", "age": 30, "job": "Engineer", "city": "Boston"}

# 3. Pydantic Model Generation
class Person(BaseModel):
    name: str
    age: int
    occupation: str
    city: str

person = client.gen_pydantic(
    model="ollama:qwen3:8b",
    prompt="Extract: Alice Johnson, 25, Designer in Seattle",
    response_model=Person,
    add_schema_to_prompt=True  # 🎯 Enhanced compliance!
)
# Output: Person(name="Alice Johnson", age=25, occupation="Designer", city="Seattle")

3. Multi-Image Vision Processing

from united_llm.utils.image_input import ImageInput

# Load multiple images
images = [
    ImageInput("image1.jpg", name="first"),
    ImageInput("image2.png", name="second")
]

# Analyze multiple images with structured output
class ImageAnalysis(BaseModel):
    description: str
    text_content: str
    objects_detected: List[str]
    image_type: str

result = client.gen_pydantic(
    model="openrouter:google/gemini-2.5-flash-lite",
    prompt="Analyze these images and extract key information",
    response_model=ImageAnalysis,
    images=images,
    add_schema_to_prompt=True
)

🎯 Core Methods

gen_text(model, prompt, images=None)

Generate plain text responses.

gen_dict(model, prompt, schema, images=None, add_schema_to_prompt=False)

Generate structured dictionaries with schema validation.

gen_pydantic(model, prompt, response_model, images=None, add_schema_to_prompt=False)

Generate validated Pydantic model instances.

📋 Schema Syntax

Uses string-schema format:

# Basic types
"{name: str, age: int, active: bool, score: float}"

# Arrays
"{tags: [str], scores: [float]}"

# Nested objects
"{user: {name: str, email: str}, posts: [{title: str, content: str}]}"

🔌 Supported Providers

OpenRouter (Cloud)

• Prefix: openrouter:
• Example: openrouter:google/gemini-2.5-flash-lite
• Requirements: API key required
• Vision Support: ✅ (with compatible models)
• Models: All OpenRouter supported models

Ollama (Local)

• Prefix: ollama:
• Example: ollama:qwen3:8b
• Requirements: Local Ollama server running
• Vision Support: ❌ (text only)
• Models: Any model available in your Ollama installation

⚙️ Configuration

Required Keys

• api_keys: Provider API keys
• base_urls: Provider endpoints

Optional Keys

• log_dir: Log directory (default: "logs/llm_calls")

Provider Setup

OpenRouter Setup

1. Sign up at OpenRouter
2. Get your API key
3. Add to united_llm.json

Ollama Setup

# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

# Pull a model
ollama pull qwen3:8b

# Start server (runs on localhost:11434)
ollama serve

📝 Logging

Automatic logging to daily files:

logs/llm_calls/2024-12-27.log

Log format:

2024-12-27 10:30:15 | ollama:qwen3:8b | gen_dict | Extract: John... | {"name": "John", "age": 30} | 1.25s

🧪 Testing

Run the comprehensive test suite:

# Basic unit tests
python -m pytest tests/test_basic.py -v

# Integration tests (requires API keys and running services)
python -m pytest tests/ -v

📚 Examples

Comprehensive examples available in the examples/ directory:

• basic_usage.py - Perfect starting point for new users
• advanced_features.py - Complex use cases and schema enhancement
• vision_examples.py - Multi-image processing examples

cd examples
python basic_usage.py

🔧 Advanced Features

Schema Prompt Enhancement

The add_schema_to_prompt=True parameter significantly improves structured output compliance:

# Without enhancement
result = client.gen_dict(
    model="ollama:qwen3:8b",
    prompt="Complex extraction task...",
    schema="{complex: {nested: {schema: str}}}"
)

# With enhancement - better compliance!
result = client.gen_dict(
    model="ollama:qwen3:8b",
    prompt="Complex extraction task...",
    schema="{complex: {nested: {schema: str}}}",
    add_schema_to_prompt=True  # 🎯 Adds schema to prompt
)

Vision Capabilities

Process multiple images with structured output:

# Single image
image = ImageInput("document.jpg")
text = client.gen_text(
    model="openrouter:google/gemini-2.5-flash-lite",
    prompt="Extract all text from this document",
    images=[image]
)

# Multiple images with comparison
images = [ImageInput("before.jpg"), ImageInput("after.jpg")]
comparison = client.gen_dict(
    model="openrouter:google/gemini-2.5-flash-lite",
    prompt="Compare these two images",
    schema="{similarities: [str], differences: [str], summary: str}",
    images=images,
    add_schema_to_prompt=True
)

🤝 Contributing

Contributions welcome! Please feel free to submit a Pull Request.

📄 License

MIT License - see LICENSE file for details.

🔗 Links

• PyPI Package: https://pypi.org/project/united-llm/
• GitHub Repository: https://github.com/xychenmsn/united-llm
• String-Schema Library: https://github.com/unaidedelf87/string-schema