PrivacyForms AI

A Python CLI tool for interacting with Large Language Models (LLMs) via Simon Willison's llm library. Supports multiple providers including OpenAI, Anthropic, Moonshot, and Ollama.

Features

• 🔧 Simple CLI - Easy-to-use command-line interface with colored output
• 💬 Interactive Chat - Multi-turn conversations with context/memory
• 🚀 Multiple Providers - Works with OpenAI, Anthropic, Moonshot, Ollama, and more
• 🧪 Well Tested - 100 % test coverage enforced via make test-cov
• ⚡ Fast - Built with modern Python tooling
• 🔍 Observable - Optional verbose logging (-v for metadata, -vv for full prompt payloads) to inspect prompt payloads

Installation

Using uv (recommended)

# Clone the repository
git clone https://github.com/zopyx/privacyforms.ai.git
cd privacyforms.ai

# Install with uv
uv sync

# Or install in development mode
uv sync --all-extras --dev

Using pip

pip install privacyforms-ai

Configuration

Set your API keys as environment variables:

# OpenAI
export OPENAI_API_KEY="your-key"

# Anthropic
export ANTHROPIC_API_KEY="your-key"

# Moonshot
export MOONSHOT_API_KEY="your-key"

For Ollama, make sure the Ollama server is running locally.

Usage

Global Options

# Show version
privacyforms-ai --version

# Show help
privacyforms-ai --help

# Enable verbose output (shows prompt logs on stderr)
privacyforms-ai -v models

# Enable debug output
privacyforms-ai -vv prompt gpt-4o-mini "Hello!"

List Available Models

privacyforms-ai models

# JSON output
privacyforms-ai models --json-output

Send a Single Prompt

# Basic prompt
privacyforms-ai prompt gpt-4o-mini "What is the capital of France?"

# With system prompt
privacyforms-ai prompt gpt-4o-mini "Explain recursion" --system "You are a computer science tutor"

# With file attachment
privacyforms-ai prompt gpt-4o-mini "Summarize this" -a document.pdf

# Start interactive chat with file attachment
privacyforms-ai chat gpt-4o-mini -a document.pdf

# Use python -m
python -m privacyforms_ai --help

Interactive Chat

Start an interactive chat session with conversation history:

# Basic chat
privacyforms-ai chat moonshot/kimi-k2.5

# With system prompt
privacyforms-ai chat gpt-4o-mini -s "You are a helpful coding assistant"

Chat Commands:

• /quit, /exit, /q - End the chat session
• /clear - Clear conversation history
• /model - Show current model

Example session:

Starting chat with model: moonshot/kimi-k2.5
Type /quit, /exit, or /q to end the session. Type /clear to reset history.
--------------------------------------------------

You: Hello!

AI: Hello! How can I help you today?

You: What can you do?

AI: I can help with a variety of tasks including...

You: /quit

Goodbye!

Custom OpenAI-compatible Endpoints

Besides the providers registered through llm, the Python API can talk to any OpenAI-compatible endpoint by passing an (api_url, api_key, model_name) triple — for example DeepSeek, Groq, Together, or a local vLLM/LiteLLM proxy:

from pathlib import Path

from privacyforms_ai import AI

model = AI.get_custom_model(
    model_name="deepseek-v4-pro",
    api_url="https://api.deepseek.com",
    api_key=Path("deepseekv4.token").read_text().strip(),
)
response = AI.send_prompt(model, "Hello!")
print(AI.extract_response_text(response))

For multi-turn conversations use AI.get_custom_conversation():

conversation = AI.get_custom_conversation(
    model_name="deepseek-v4-pro",
    api_url="https://api.deepseek.com",
    api_key=Path("deepseekv4.token").read_text().strip(),
    system="You are a helpful assistant.",
)
response = AI.send_conversation_prompt(conversation, "Hello!")
print(AI.extract_response_text(response))

Pass vision=True to get_custom_model() for endpoints whose models accept image attachments. Keep token files like deepseekv4.token out of version control — the repo's .gitignore already covers this one.

Authentication note: the api_key passed to get_custom_model() is sent to the endpoint as-is. Internally the model keeps llm's needs_key flag truthy so that the explicitly passed key is used. Do not set model.needs_key = None on the returned model: llm then treats the model as key-less and substitutes the literal placeholder DUMMY_KEY as the Bearer token, which the endpoint rejects with HTTP 401 (DeepSeek reports this as Your api key: ****_KEY is invalid — the _KEY suffix is the placeholder, not your key).

A ready-made smoke test for the DeepSeek endpoint lives at scripts/deepseek_smoke.py (run with uv run python scripts/deepseek_smoke.py; requires a valid key in deepseekv4.token and network access).

Python API

All functionality is exposed through the AI class:

Example covering the full lifecycle:

from privacyforms_ai import AI

# 1. List registered models
for m in AI.get_models():
    print(m["key"], "-", m["name"], f"({m['provider']})")

# 2. Registered model, single prompt
model = AI.get_model("gpt-4o-mini")
response = AI.send_prompt(model, "Hello!", system="Be concise.")
print(AI.extract_response_text(response))

# 3. Registered model, multi-turn conversation
conversation = AI.get_conversation("gpt-4o-mini", system="You are a helpful assistant.")
print(AI.extract_response_text(AI.send_conversation_prompt(conversation, "What is 2+2?")))

# 4. File attachments
print(AI.prompt_with_attachment(model, "Summarize this", "document.pdf"))

llm response objects are lazy: the network call only happens when the response is consumed (e.g. via AI.extract_response_text(response)).

Development

Setup

# Clone and setup
git clone https://github.com/zopyx/privacyforms.ai.git
cd privacyforms.ai
uv sync --all-extras --dev
source .venv/bin/activate

Running Tests

# Run all tests
make test

# With coverage
make test-cov

# Verbose output
uv run pytest -v

Code Quality

# Format code
make format

# Check formatting
make format-check

# Lint
make lint

# Auto-fix linting issues
make fix

# Type check
make type-check

# Run the full local gate
make check

Build Package

# Build release artifacts into dist/
make dist

Upload Package

# Upload to PyPI using twine and your ~/.pypirc or TWINE_* credentials
make upload

# Upload to another configured repository, e.g. TestPyPI
make upload TWINE_REPOSITORY=testpypi

Create a Release

# 1. Update the version in pyproject.toml, src/privacyforms_ai/_version.py, README, and tests

# 2. Refresh the lockfile if needed
uv sync --all-extras --dev

# 3. Verify and build
make check
make dist

# 4. Upload
make upload

# 5. Commit and tag
git add pyproject.toml src/privacyforms_ai/_version.py src/privacyforms_ai/__init__.py tests/ uv.lock CHANGELOG.md .gitattributes LICENSE
git commit -m "Release X.Y.Z"
git tag vX.Y.Z
git push origin HEAD
git push origin vX.Y.Z

Project Structure

privacyforms.ai/
├── src/privacyforms_ai/
│   ├── __init__.py
│   ├── _version.py        # Package version
│   ├── ai.py              # AI class for LLM interactions
│   └── cli.py             # Click CLI commands
├── tests/
│   ├── conftest.py        # Pytest fixtures
│   ├── test_ai.py         # AI class tests
│   └── test_cli.py        # CLI tests
├── pyproject.toml         # Project configuration
├── uv.lock               # Locked dependencies
├── CHANGELOG.md          # Release notes
├── LICENSE               # MIT license
├── .gitattributes        # Line-ending configuration
└── README.md

CI/CD

GitHub Actions workflow runs on:

• Python 3.12, 3.13, 3.14, 3.14t (free-threaded)
• Ubuntu Linux

Jobs:

• test - Run pytest with coverage
• lint - ruff (formatting, linting) and ty (type checking)
• build - Build package artifacts and validate with twine

License

MIT License - see LICENSE file for details.

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Make your changes with tests
4. Ensure all checks pass (make check)
5. Submit a pull request

Acknowledgments

• Built on top of Simon Willison's llm library
• Uses Astral's uv for fast Python package management