freeflow-llm 0.1.3

Chain multiple free-tier LLM APIs with automatic rate limit fallback

FreeFlow LLM

Chain multiple free-tier LLM APIs with automatic rate limit fallback.

FreeFlow LLM is a lightweight Python package that lets you use powerful LLMs completely free by intelligently chaining multiple free-tier providers (Groq, Google Gemini, GitHub Models). When one provider hits a rate limit, it automatically switches to the next one, giving you effectively unlimited free usage! +`

✨ Features

• 100% Free-Tier Only — No paid tiers, no credit card required
• Automatic Fallback — Detects rate limits (HTTP 429) and switches providers instantly
• Smart Prioritization — Starts with fastest providers (Groq), falls back to others
• Clean & Unified API — Simple and consistent client.chat() interface across all providers
• Lightweight & Simple — Minimal dependencies, easy to use
• 16,000+ Requests/Day — Aggregate free usage across all providers

Quick Start

Installation

pip install freeflow-llm

Set Up API Keys

Get free API keys from these providers (you only need at least one):

1. Groq (Recommended): https://console.groq.com/keys
2. Google Gemini: https://makersuite.google.com/app/apikey
3. GitHub Models: https://github.com/settings/tokens

Set them as environment variables:

export GROQ_API_KEY="your_groq_key"
export GEMINI_API_KEY="your_gemini_key"
export GITHUB_TOKEN="your_github_token"

Or create a .env file:

GROQ_API_KEY=your_groq_key
GEMINI_API_KEY=your_gemini_key
GITHUB_TOKEN=your_github_token

Multiple API Keys per Provider (New!)

You can now configure multiple API keys for each provider. When rate limits are hit, FreeFlow will automatically rotate through all available keys before moving to the next provider:

Format 1: JSON Array (recommended for complex keys):

export GEMINI_API_KEY='["key1", "key2", "key3"]'
export GROQ_API_KEY='["groq_key1", "groq_key2"]'

Format 2: Comma-Separated (simpler):

export GEMINI_API_KEY="key1,key2,key3"
export GROQ_API_KEY="groq_key1,groq_key2"

Format 3: Single Key (traditional):

export GEMINI_API_KEY="single_key"

With multiple keys, your effective rate limit multiplies! For example, 3 Gemini keys = 4,500 requests/day instead of 1,500.

Basic Usage

from freeflow_llm import FreeFlowClient

# Use context manager for automatic resource cleanup (recommended)
with FreeFlowClient() as client:
    response = client.chat(
        messages=[
            {"role": "system", "content": "You are a helpful assistant."},
            {"role": "user", "content": "What is the capital of Ethiopia?"}
        ],
        temperature=0.7,
        max_tokens=100
    )

    print(response.content)
    # Output: "The capital of Ethiopia is Addis Ababa."

    print(f"Provider used: {response.provider}")
    # Output: "Provider used: groq" (or whichever provider responded)
# Resources (HTTP connections) are automatically closed here

That's it! FreeFlow will automatically try providers in order and handle rate limits transparently.

Error Handling

from freeflow_llm import FreeFlowClient, NoProvidersAvailableError

with FreeFlowClient() as client:
    try:
        response = client.chat(
            messages=[{"role": "user", "content": "Hello!"}]
        )
        print(response.content)
    except NoProvidersAvailableError as e:
        print(f"All providers exhausted: {e}")
        # Handle gracefully or retry later

Manual Resource Management

If you need more control, you can manually manage resources:

client = FreeFlowClient()
try:
    response = client.chat(messages=[{"role": "user", "content": "Hello!"}])
    print(response.content)
finally:
    client.close()  # Clean up resources

Why use context managers?

• Prevents memory leaks in long-running applications
• Ensures proper cleanup of HTTP connections
• Prevents resource exhaustion warnings
• Best practice for production code

🔧 Advanced Configuration

Default Models

Each provider has a default model configured. When you don't specify a model, FreeFlow uses the provider's default:

from freeflow_llm import FreeFlowClient, config

# Use default models (no model parameter needed)
client = FreeFlowClient()
response = client.chat(messages=[{"role": "user", "content": "Hello!"}])

# Check default models
print(config.DEFAULT_MODELS)
# Output: {
#   'groq': 'llama-3.3-70b-versatile',
#   'gemini': 'gemini-1.5-flash',
#   'mistral': 'mistral-small-latest',
#   ...
# }

# Override with custom model
response = client.chat(
    messages=[{"role": "user", "content": "Hello!"}],
    model="llama-3.1-70b-versatile"  # Override default
)

List Available Providers

client = FreeFlowClient()
print(client.list_providers())
# Output: ['groq', 'gemini']

# Check how many keys are loaded per provider
for provider in client.providers:
    num_keys = len(provider.api_keys)
    print(f"{provider.name}: {num_keys} API key(s)")
# Output:
# groq: 2 API key(s)
# gemini: 3 API key(s)

Custom Provider Configuration

from freeflow_llm import FreeFlowClient
from freeflow_llm.providers import GeminiProvider, GroqProvider

# Option 1: Pass multiple keys directly to providers
custom_providers = [
    GroqProvider(api_key=["key1", "key2"]),
    GeminiProvider(api_key=["key1", "key2", "key3"]),
]
client = FreeFlowClient(providers=custom_providers)

# Option 2: Single key per provider
custom_providers = [
    GroqProvider(api_key="single_key"),
    GeminiProvider(api_key="single_key"),
]
client = FreeFlowClient(providers=custom_providers)

📊 Provider Details

Provider	Free Tier Limit	Speed	Models
Groq	~14,400 req/day	⚡ Very Fast	Llama 3.3 70B, Mixtral
Gemini	1,500 req/day	⚡ Fast	Gemini 1.5 Flash
GitHub	Rate limited	⚡ Fast	GPT-4o, GPT-4o-mini

Total Effective Limit: 16,000+ requests/day by chaining all providers! 🎉

🛠️ Development

Setup

# Clone the repository
git clone https://github.com/thesecondchance/freeflow-llm.git
cd freeflow-llm

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install in development mode
pip install -e ".[dev]"

# Install pre-commit hooks (recommended)
pre-commit install

Code Quality Standards

We maintain strict code quality standards. All PRs must pass:

✅ Black - Code formatting
✅ Ruff - Linting & code quality
✅ MyPy - Type checking
✅ Circular Import Detection
✅ Import Safety Checks
✅ All Tests

Check your code:

# Run all checks
python scripts/check_code_quality.py

# Auto-fix what you can
bash scripts/fix_code_quality.sh  # Linux/macOS
powershell scripts/fix_code_quality.ps1  # Windows

Or use pre-commit hooks (runs automatically on commit):

pre-commit run --all-files

Running Tests

pytest tests/ -v
pytest tests/ --cov=src/freeflow_llm  # With coverage

🤝 Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.

Ideas for Contributions

• Add more free-tier providers
• Implement streaming support
• Add async support
• Improve error handling
• Write more tests

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

⚠️ Disclaimer

This package is designed for free-tier usage only. Please respect each provider's rate limits and terms of service. FreeFlow LLM is not affiliated with any of the LLM providers.

🙏 Acknowledgments

• Groq for their blazingly fast free tier
• Google for Gemini API
• GitHub for free model access

🔗 Links

• Documentation
• PyPI Package
• Issue Tracker
• Changelog

---

Made with ❤️ for the developer community

If this project helps you, please consider giving it a ⭐ on GitHub!