janito4 0.1.0

OpenAI CLI with Function Calling Tools

Janito4 - OpenAI CLI

A simple command-line interface to interact with OpenAI-compatible endpoints (including OpenAI or any OpenAI-compatible API) with built-in function calling capabilities.

Features

• Uses environment variables for configuration (OPENAI_BASE_URL, OPENAI_API_KEY, OPENAI_MODEL)
• Accepts prompts as command-line arguments or from stdin
• Works with any OpenAI-compatible endpoint
• Simple and lightweight
• Real-time tool progress reporting - Tools can show their progress during execution
• Extensible tool system - Easy to add new tools with automatic schema generation

Installation

1. Clone this repository or download the files

2. Install dependencies:

   pip install -r requirements.txt
   

3. The package can be used directly from the source directory, or installed in development mode:

   pip install -e .
   

Configuration

Interactive Configuration Setup

You can set up your configuration interactively using the --config flag:

python -m janito4 --config

This will prompt you for:

• Provider name (e.g., openai, anthropic, azure) - lists available providers
• API key - masked for security, shows existing key if available
• Model name - the model to use
• Context window size - max tokens (default: 65536)

Existing configuration values are used as defaults, so you can just press Enter to keep them.

Manual Configuration

Set the following environment variables:

# For standard OpenAI API (OPENAI_BASE_URL is optional)
export OPENAI_API_KEY="sk-your-openai-key"
export OPENAI_MODEL="gpt-4"

# For OpenAI-compatible endpoints (set OPENAI_BASE_URL)
export OPENAI_BASE_URL="https://api.openai.com"          # For OpenAI (explicit)

# export OPENAI_BASE_URL="http://localhost:8080/v1"      # For local servers like LM Studio, Ollama, etc.
export OPENAI_API_KEY="your-api-key-here"
export OPENAI_MODEL="gpt-4"                              # Or your preferred model

Usage

Single Prompt Mode (command-line argument):

python -m janito4 "What is the capital of France?"

Pipe Input Mode:

echo "Tell me a joke" | python -m janito4

Interactive Chat Mode:

python -m janito4
# Starts an interactive chat session where you can have multi-turn conversations
# Type 'exit' or 'quit' to end the session, 'restart' to clear conversation history, or press Ctrl+D/Ctrl+Z

Logging:

# Enable info logging
python -m janito4 --log=info "Your prompt"

# Enable debug logging for detailed output
python -m janito4 --log=debug "Your prompt"

# Enable multiple log levels
python -m janito4 --log=info,debug "Your prompt"

# Enable warning and error only
python -m janito4 --log=warning,error "Your prompt"

Examples

OpenAI API:

export OPENAI_BASE_URL="https://api.openai.com"
export OPENAI_API_KEY="sk-your-openai-key"
export OPENAI_MODEL="gpt-4"
python -m janito4 "Explain quantum computing in simple terms"

Local LLM (e.g., LM Studio, Ollama):

export OPENAI_BASE_URL="http://localhost:1234/v1"  # LM Studio default
export OPENAI_API_KEY="not-needed"                 # Often not required for local servers
export OPENAI_MODEL="local-model-name"
python -m janito4 "What is 2+2?"

PowerShell Usage:

# Set environment variables for current session
$env:OPENAI_BASE_URL = "https://api.openai.com"
$env:OPENAI_API_KEY = "your-api-key-here"
$env:OPENAI_MODEL = "gpt-4"

# Use the CLI
python -m janito4 "What is the capital of France?"
echo "Tell me a joke" | python -m janito4

Environment Variables

Variable	Description	Example
OPENAI_BASE_URL	Base URL of the OpenAI-compatible API	https://api.openai.com
OPENAI_API_KEY	API key for authentication	sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
OPENAI_MODEL	Model name/deployment name to use	gpt-4, gpt-3.5-turbo, your-local-model

Tool Progress Reporting

Tools can report their progress in real-time using the built-in progress reporting system. This provides transparency into what the AI agent is doing during tool execution.

Direct Tool Execution

Individual tools can be executed directly using Python's module syntax:

python -m janito4.tools.files.list_files . --recursive --pattern "*.py"
python -m janito4.tools.files.read_file README.md --max-lines 10

Tools are now implemented as classes that inherit from BaseTool and must implement a run() method. The @tool decorator is applied to the class, and the system automatically creates a wrapper function for AI function calling compatibility.

This allows for testing and debugging tools independently while still having access to the full tooling infrastructure including progress reporting.

• 🔄 Start messages: Indicate when a tool operation begins
• 📊 Progress messages: Show ongoing work (e.g., "Processing item 50/100")
• ✅ Result messages: Report successful completion with statistics
• ❌ Error messages: Display errors that occur during execution
• ⚠️ Warning messages: Show non-critical warnings

Progress messages are displayed via stderr so they don't interfere with the tool's actual return value that gets sent back to the AI model.

MCP Tool Progress Reporting

MCP tools also report their progress using the same system:

 🔄 MCP tool: myserver_read_file [myserver]
 ✅ returned 50 lines (2048 chars)

When calling an MCP tool, you'll see:

1. Start message with the tool name
2. Progress message showing which service is handling the call
3. Result message summarizing the output (e.g., "returned 50 lines", "returned keys: success, content")

This provides the same transparency for MCP tools as built-in tools, making it easy to track what the AI agent is doing across both local and remote tool sources.

Error Handling

The CLI will exit with appropriate error codes:

• 1: Configuration or runtime errors
• 130: User cancelled operation (Ctrl+C)

Dependencies

• Python 3.6+
• requests library

License

MIT License