claif 1.0.33

A unified CLI and Python library for interacting with local and remote LLMs, compatible with the OpenAI Responses API.

Claif - Command Line Artificial Intelligence Framework

Quickstart

# Install with all providers and query Claude
pip install claif[all]
claif query "Explain quantum computing in one sentence"

# Or install specific providers
pip install claif claif_cla  # Just Claude
pip install claif claif_gem  # Just Gemini  
pip install claif claif_cod  # Just Codex/OpenAI

# Stream responses in real-time
claif stream "Tell me a story" --provider gemini

# Query all providers in parallel
claif parallel "What is the meaning of life?" --compare

What is Claif?

Claif is a unified Python framework for interacting with multiple AI language model providers through a consistent interface. It lets you seamlessly switch between Claude (Anthropic), Gemini (Google), and Codex (OpenAI) without changing your code.

Key Features:

• Single API for all AI providers - write once, use any model
• Plugin-based architecture - providers are discovered dynamically
• Rich CLI with Fire framework and beautiful terminal output
• Full async support for concurrent operations
• Type-safe with comprehensive hints throughout
• MCP server for tool integration
• Auto-install - missing provider CLIs are installed on first use

Installation

Basic Installation

# Core framework only
pip install claif

# With all providers (recommended)
pip install claif[all]

Installing Specific Providers

# Claude provider (wraps Claude Code SDK)
pip install claif claif_cla

# Gemini provider (wraps Gemini CLI)  
pip install claif claif_gem

# Codex provider (wraps OpenAI Codex CLI)
pip install claif claif_cod

Installing Provider CLIs

Claif can auto-install missing CLIs, or you can install them manually:

# Auto-install all CLIs
claif install

# Or install specific CLIs
claif install claude
claif install gemini
claif install codex

# Install via npm (if preferred)
npm install -g @anthropic-ai/claude-code
npm install -g @google/gemini-cli
npm install -g @openai/codex

Development Installation

git clone https://github.com/twardoch/claif.git
cd claif
pip install -e ".[dev,test]"

CLI Usage

Claif provides a comprehensive command-line interface built with Fire that shows clean, focused output by default. Use --verbose for detailed logging.

Basic Commands

# Query default provider (Claude)
claif query "What is Python?"

# Query specific provider
claif query "Explain recursion" --provider gemini

# Query with options
claif query "Write a haiku about coding" --temperature 0.7 --max-tokens 50

# Stream responses in real-time with live display
claif stream "Tell me a story" --provider claude

# Query a random provider
claif random "Tell me a programming joke"

# Query all providers in parallel
claif parallel "What is AI?" --compare  # Side-by-side comparison

Provider Management

# List providers with their status
claif providers list

# Check health of all providers
claif providers status

# Install provider CLIs
claif install         # All providers
claif install claude  # Specific provider

# Check installation status
claif status

Configuration

# Show current configuration
claif config show

# Set configuration values
claif config set default_provider=gemini
claif config set cache_enabled=true
claif config set output_format=markdown

# Save configuration to file
claif config save

MCP Server

# Start the MCP (Model Context Protocol) server
claif server --host localhost --port 8000 --reload

# The server provides tools:
# - claif_query: Query specific provider
# - claif_query_random: Query random provider  
# - claif_query_all: Query all providers
# - claif_list_providers: List available providers
# - claif_health_check: Check provider health

Python API Usage

Basic Usage

import asyncio
from claif import query, ClaifOptions, Provider

async def main():
    # Simple query using default provider
    async for message in query("Hello, world!"):
        print(message.content)
    
    # Query specific provider  
    options = ClaifOptions(provider=Provider.GEMINI)
    async for message in query("Explain Python decorators", options):
        print(f"[{message.role}]: {message.content}")

asyncio.run(main())

Advanced Options

from claif import query, ClaifOptions, Provider

options = ClaifOptions(
    provider=Provider.CLAUDE,
    model="claude-3-opus-20240229",
    temperature=0.7,
    max_tokens=1000,
    system_prompt="You are an expert Python developer",
    timeout=60,
    cache=True,
    verbose=False  # Clean output by default
)

async def get_code_review():
    async for message in query("Review this code: def fib(n): return fib(n-1) + fib(n-2)", options):
        print(message.content)

asyncio.run(get_code_review())

Parallel Queries

from claif import query_all, query_random

# Query all providers in parallel
async def compare_providers():
    async for results in query_all("What is machine learning?"):
        for provider, messages in results.items():
            print(f"\n{provider.value.upper()}:")
            for msg in messages:
                print(msg.content)

# Query random provider
async def get_random_response():
    async for message in query_random("Tell me a programming joke"):
        print(message.content)

asyncio.run(compare_providers())

Using the Client Class

from claif.client import ClaifClient
from claif import ClaifOptions, Provider

# Create client instance
client = ClaifClient()

# List available providers
providers = client.list_providers()
print(f"Available: {[p.value for p in providers]}")

# Query with auto-install
# If Claude CLI is missing, it will be installed automatically
options = ClaifOptions(provider=Provider.CLAUDE)

async def query_with_client():
    async for message in client.query("Explain asyncio", options):
        print(message.content)

asyncio.run(query_with_client())

How It Works

Architecture Overview

Claif uses a layered architecture that separates concerns and enables provider independence:

┌─────────────────────────────┐
│      User Application       │
├─────────────────────────────┤
│       Claif CLI           │ ← Fire-based CLI with rich output
├─────────────────────────────┤
│     Claif Client          │ ← Async client with provider routing
├─────────────────────────────┤
│    Common Types/Utils      │ ← Shared data structures and utilities
├─────────────────────────────┤
│    Provider Wrappers       │ ← Simple adapters for provider packages
├─────┬─────┬─────┬──────────┤
│claif_cla│claif_gem│claif_cod│ ← Actual provider implementations
└─────┴─────┴─────┴──────────┘

Core Components

Common Module (src/claif/common/)

types.py - Core data structures:

class Message:
    role: MessageRole  # USER, ASSISTANT, SYSTEM
    content: str | list[TextBlock | ToolUseBlock | ToolResultBlock]

class Provider(Enum):
    CLAUDE = "claude"
    GEMINI = "gemini" 
    CODEX = "codex"

class ClaifOptions:
    provider: Provider | None
    model: str | None
    temperature: float | None
    max_tokens: int | None
    system_prompt: str | None
    timeout: int | None
    cache: bool = True
    verbose: bool = False

config.py - Hierarchical configuration:

• Default values → Config files → Environment vars → CLI args
• Locations: ~/.claif/config.json, ~/.config/claif/config.json, ./claif.json
• Provider-specific settings with API key management

errors.py - Exception hierarchy:

• ClaifError → Base exception
• ProviderError → Provider-specific failures
• ConfigurationError → Configuration issues
• TransportError → Communication errors
• TimeoutError → Operation timeouts

install.py - Auto-installation support:

def install_provider(provider, package, exec_name):
    # 1. Install npm package using bun
    # 2. Bundle executable
    # 3. Install to ~/.local/bin

Providers Module (src/claif/providers/)

Simple wrapper classes that delegate to provider packages:

# claude.py
from claif_cla import query as claude_query

class ClaudeProvider:
    async def query(self, prompt: str, options: ClaifOptions) -> AsyncIterator[Message]:
        logger.debug(f"Querying Claude with prompt: {prompt[:50]}...")
        async for message in claude_query(prompt, options):
            yield message

Each provider:

• Imports the actual provider package (claif_cla, claif_gem, claif_cod)
• Implements the same query interface
• Handles provider-specific option conversion

Client Module (src/claif/client.py)

The main client with auto-install support:

class ClaifClient:
    def __init__(self):
        self.providers = {
            Provider.CLAUDE: ClaudeProvider(),
            Provider.GEMINI: GeminiProvider(),
            Provider.CODEX: CodexProvider(),
        }
    
    async def query(self, prompt: str, options: ClaifOptions) -> AsyncIterator[Message]:
        try:
            # Route to provider
            async for message in provider.query(prompt, options):
                yield message
        except Exception as e:
            if _is_cli_missing_error(e):
                # Auto-install missing CLI
                install_result = install_func()
                if install_result.get("installed"):
                    # Retry query
                    async for message in provider.query(prompt, options):
                        yield message

Key features:

• Routes queries to appropriate providers
• Auto-installs missing provider CLIs
• Implements query, query_random, and query_all
• Handles errors with graceful fallbacks

CLI Module (src/claif/cli.py)

Fire-based CLI with rich terminal output:

class ClaifCLI:
    def __init__(self, config_file=None, verbose=False):
        # Clean output by default
        logger.remove()
        if verbose:
            logger.add(sys.stderr, level="DEBUG")
        else:
            # Only errors go to stderr in non-verbose mode
            logger.add(sys.stderr, level="ERROR")

Main commands:

• query - Execute queries with options
• stream - Live streaming with rich.Live display
• random - Query random provider
• parallel - Query all providers (with --compare for side-by-side)
• session - Start provider-specific interactive session
• install - Install provider CLIs with bun bundling
• config - Manage settings
• server - Start MCP server

Server Module (src/claif/server.py)

FastMCP server implementation:

server = FastMCP("Claif MCP Server")

@server.tool()
async def claif_query(prompt: str, provider: str = None) -> list[dict]:
    """Query specific AI provider."""
    options = ClaifOptions(provider=Provider(provider) if provider else None)
    messages = []
    async for message in query(prompt, options):
        messages.append(message.to_dict())
    return messages

# Additional tools:
# - claif_query_random: Random provider selection
# - claif_query_all: Parallel queries
# - claif_list_providers: Available providers
# - claif_health_check: Provider status

Install Module (src/claif/install.py)

Handles CLI installation and bundling:

def install_claude() -> dict:
    """Install Claude Code CLI."""
    return install_provider(
        provider="claude",
        package="@anthropic-ai/claude-code", 
        exec_name="claude"
    )

# Installation process:
# 1. Install npm package globally using bun
# 2. Bundle with bun compile for fast startup
# 3. Copy to ~/.local/bin
# 4. Handle platform-specific paths

Key features:

• Uses bun for fast npm installs
• Creates bundled executables
• Platform-aware installation paths
• Automatic PATH checking

Configuration System

Hierarchical configuration loading:

1. Default values in Config dataclass
2. Config files (first found wins):
   • ~/.claif/config.json
   • ~/.config/claif/config.json
   • ./claif.json
3. Environment variables:
   • CLAIF_DEFAULT_PROVIDER
   • CLAIF_VERBOSE
   • CLAIF_CACHE_ENABLED
   • CLAIF_SESSION_DIR
4. CLI arguments (highest priority)

Example configuration:

{
    "default_provider": "claude",
    "providers": {
        "claude": {
            "enabled": true,
            "model": "claude-3-opus-20240229",
            "api_key_env": "ANTHROPIC_API_KEY"
        },
        "gemini": {
            "enabled": true,
            "model": "gemini-2.5-pro",
            "api_key_env": "GEMINI_API_KEY"
        },
        "codex": {
            "enabled": true,
            "model": "o4-mini"
        }
    },
    "cache_enabled": true,
    "cache_ttl": 3600,
    "output_format": "text"
}

Plugin System

Python's entry points for dynamic provider discovery:

# In src/__init__.py
class PluginFinder:
    """Enables imports like: from claif import claude"""
    def find_spec(cls, fullname, path, target=None):
        if fullname.startswith("claif."):
            # Look up via entry points
            eps = metadata.entry_points(group="claif.plugins")
            for ep in eps:
                if ep.name == plugin_name:
                    return ModuleSpec(fullname, Loader())

Provider packages register themselves:

# In provider's pyproject.toml
[project.entry-points."claif.plugins"]
claude = "claif_cla"
gemini = "claif_gem" 
codex = "claif_cod"

Message Flow

1. User Input → CLI command or API call
2. Options Parsing → Create ClaifOptions with configuration
3. Provider Selection → Choose provider based on options
4. Query Routing → Client routes to appropriate provider wrapper
5. CLI Check → Auto-install if CLI missing
6. Provider Execution → Provider package handles actual call
7. Response Streaming → Yield Message objects as they arrive
8. Output Formatting → Display with rich or return structured data

Code Structure

claif/
├── src/
│   ├── __init__.py              # Plugin system initialization
│   └── claif/
│       ├── __init__.py          # Public API exports
│       ├── common/              # Shared components
│       │   ├── __init__.py      
│       │   ├── types.py         # Core data structures
│       │   ├── config.py        # Configuration management
│       │   ├── errors.py        # Exception hierarchy
│       │   ├── install.py       # CLI installation utilities
│       │   └── utils.py         # Formatting and helpers
│       ├── providers/           # Provider wrappers
│       │   ├── __init__.py      
│       │   ├── claude.py        # Claude wrapper
│       │   ├── gemini.py        # Gemini wrapper
│       │   └── codex.py         # Codex wrapper
│       ├── client.py            # Main client implementation
│       ├── cli.py               # Fire CLI interface
│       ├── server.py            # FastMCP server
│       └── install.py           # Provider CLI installation
├── tests/                       # Test suite
├── pyproject.toml               # Package configuration
└── README.md                    # This file

Key Implementation Details

Async Everywhere: All I/O operations use async/await for efficiency

Message Types: Flexible content that can be string or structured blocks:

class Message:
    role: MessageRole
    content: str | list[TextBlock | ToolUseBlock | ToolResultBlock]

Provider Adapters: Thin wrappers that maintain provider-specific features:

# Each provider wrapper is ~20 lines
# Imports provider package
# Converts options if needed  
# Yields messages unchanged

Auto-Install Logic: Missing CLIs trigger automatic installation:

if _is_cli_missing_error(e):
    install_func = _get_provider_install_function(provider)
    if install_func():
        # Retry query with fresh provider instance

Clean Output: Logging configured to minimize noise:

# Verbose OFF: Only errors to stderr
# Verbose ON: Full debug logging
# AI responses always go to stdout

Installation Details: Bun and Node

Claif uses a hybrid approach for installing provider CLIs:

Why Bun?

Bun is used for:

1. Fast npm installs - 10x faster than npm
2. Binary bundling - Creates standalone executables
3. Cross-platform - Works on Windows, macOS, Linux
4. Minimal dependencies - Single binary, no Node.js required

Installation Process

# 1. Bun installs the npm package globally
bun add -g @anthropic-ai/claude-code

# 2. Bundle script creates optimized binary
bun build claude-wrapper.ts --compile --outfile dist/claude

# 3. Binary is copied to ~/.local/bin
cp dist/claude ~/.local/bin/

Manual Installation Options

If you prefer Node.js:

# Using npm
npm install -g @anthropic-ai/claude-code
npm install -g @google/gemini-cli  
npm install -g @openai/codex

# Using yarn
yarn global add @anthropic-ai/claude-code

# Using pnpm
pnpm add -g @anthropic-ai/claude-code

Bundle Benefits

1. Faster startup - No Node.js initialization
2. Smaller size - Single file vs node_modules
3. No conflicts - Isolated from system Node.js
4. Portable - Copy binary anywhere

Troubleshooting

# Check if CLIs are found
claif status

# Manually set CLI paths
export CLAUDE_CLI_PATH=/usr/local/bin/claude
export GEMINI_CLI_PATH=/usr/local/bin/gemini
export CODEX_CLI_PATH=/usr/local/bin/codex

# Skip auto-install and use existing CLIs
claif config set auto_install=false

Why Use Claif?

1. Provider Independence

• Write once, use any AI model
• Switch providers with a single parameter
• Compare responses side-by-side
• No vendor lock-in

2. Zero Friction

• Auto-installs missing CLIs on first use
• Clean output - only show what matters
• Smart defaults that just work
• Async for speed

3. Developer Experience

• Full type hints and IDE support
• Rich CLI with progress indicators
• Comprehensive error messages
• Fast MCP server for tool integration

4. Production Ready

• Battle-tested error handling
• Timeout protection
• Response caching (coming soon)
• Configurable everything

5. Extensible

• Plugin architecture for new providers
• Clean API for custom integrations
• Well-documented codebase
• Active development

Contributing

Based on the development guide in CLAUDE.md:

Development Setup

# Clone the repository
git clone https://github.com/twardoch/claif.git
cd claif

# Install with dev dependencies
pip install -e ".[dev,test]"

# Install pre-commit hooks
pre-commit install

Running Tests

The Claif framework includes comprehensive tests to ensure robust functionality:

# Install with test dependencies
pip install -e ".[test]"

# Run all tests
uvx hatch test

# Run specific test modules
uvx hatch test -- tests/test_functional_simple.py -v
uvx hatch test -- tests/test_client.py -v

# Run with coverage
uvx hatch test -- --cov=src/claif --cov-report=html

Test Structure

tests/
├── test_functional_simple.py  # Simple functional tests (no provider dependencies)
├── test_client.py             # Core client functionality tests
├── test_providers.py          # Provider discovery and loading tests
├── test_config.py             # Configuration management tests
└── conftest.py                # Test fixtures and configuration

Example Test Usage

The functional tests demonstrate the unified client interface:

# Test provider auto-detection
def test_provider_auto_detection():
    # Auto-detects Claude if ANTHROPIC_API_KEY is set
    with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "test-key"}):
        client = ClaifClient()
        assert client.provider == "claude"

# Test unified API across providers  
def test_unified_chat_api():
    client = ClaifClient(provider="claude")
    
    response = client.chat.completions.create(
        model="claude-3-5-sonnet-20241022",
        messages=[{"role": "user", "content": "Hello"}]
    )
    
    assert isinstance(response, ChatCompletion)
    assert response.choices[0].message.role == "assistant"

# Test provider delegation
def test_provider_delegation():
    with patch("claif_cla.client.ClaudeClient") as mock_claude:
        mock_client = MagicMock()
        mock_claude.return_value = mock_client
        
        client = ClaifClient(provider="claude")
        client.chat.completions.create(
            model="claude-3-5-sonnet-20241022",
            messages=[{"role": "user", "content": "Test"}]
        )
        
        # Verify delegation to provider client
        mock_client.chat.completions.create.assert_called_once()

# Test streaming support
def test_streaming():
    client = ClaifClient(provider="claude")
    
    stream = client.chat.completions.create(
        model="claude-3-5-sonnet-20241022",
        messages=[{"role": "user", "content": "Count to 3"}],
        stream=True
    )
    
    chunks = list(stream)
    assert len(chunks) > 0
    assert all(isinstance(chunk, ChatCompletionChunk) for chunk in chunks)

Testing Provider Packages

Each provider package (claif_cla, claif_gem, claif_cod) includes its own comprehensive test suite. The main claif package includes simple functional tests that don't require provider dependencies for CI/CD compatibility.

For full integration testing with providers:

# Install all providers
pip install claif[all]

# Run provider-specific tests
cd claif_cla && uvx hatch test
cd claif_gem && uvx hatch test  
cd claif_cod && uvx hatch test

# Run integration tests
uvx hatch test -- tests/integration/ -v

Code Quality

# Format code
ruff format src tests

# Lint code  
ruff check src tests --fix

# Type checking
mypy src/claif

# Run all checks (as per CLAUDE.md)
fd -e py -x ruff format {}
fd -e py -x ruff check --fix --unsafe-fixes {}
python -m pytest

Making Changes

1. Check existing utilities in claif.common before implementing
2. Maintain API compatibility - this is a plugin framework
3. Add tests for new functionality
4. Update documentation in docstrings and README
5. Follow style - 120 char lines, descriptive names

Release Process

# Update version (uses hatch-vcs)
git tag v1.0.7

# Build distribution
python -m build

# Upload to PyPI
twine upload dist/*

License

MIT License - see LICENSE file for details.

Copyright (c) 2025 Adam Twardoch

Links

Claif Ecosystem

Core Framework:

• GitHub: twardoch/claif - This repository
• PyPI: claif - Core package
• Documentation - Full docs
• Issues - Bug reports & features

Provider Packages:

• claif_cla - Claude provider (wraps claude-code-sdk)
• claif_gem - Gemini provider (wraps Gemini CLI)
• claif_cod - Codex provider (wraps OpenAI Codex CLI)

Upstream Projects

Provider CLIs:

• Claude Code - Anthropic's official CLI
• claude-code-sdk - Python SDK for Claude
• Gemini CLI - Google's Gemini CLI
• OpenAI Codex CLI - OpenAI's code generation CLI

Related Tools:

• Fire - Python CLI framework
• Rich - Terminal formatting
• FastMCP - MCP server framework
• Bun - Fast JavaScript runtime used for bundling

Resources

Documentation:

• Anthropic Docs - Claude documentation
• Google AI Studio - Gemini documentation
• OpenAI Platform - OpenAI documentation

Community:

• Discussions - Q&A and ideas
• Twitter: @adamtwar - Author updates

Articles & Tutorials:

• Building a Unified AI CLI - Design decisions
• Plugin Architecture in Python - Technical deep dive
• Async Patterns for AI APIs - Performance guide