glagos-flowai 0.5.96

A Python CLI tool for streamlining development workflows with LLMs - commit messages, PR descriptions, code reviews and more

FlowAI

FlowAI is a Python-based CLI tool that helps developers streamline their development workflow by automating common tasks using LLMs (Language Learning Models).

Features

• Generate detailed commit messages from git diffs
• Create comprehensive pull request descriptions
• Perform automated code reviews
• Interactive chat mode with streaming support
• Web search capability with Gemini models
• Thinking mode for Gemini 2.5 models
• Support for multiple LLM providers:
  • OpenAI
  • Anthropic
  • Groq
  • Gemini
  • Ollama
• Cross-platform compatibility (Windows, Mac, Linux)
• Markdown rendering in terminal
• Streaming responses for real-time feedback
• Configurable output formatting per command

Command System

FlowAI uses a powerful command system to automate common tasks. Commands are defined in ~/flowai-prompts/prompt-index.txt and can be customized to your needs.

Command Features

• Pre-configured context gathering
• Template-based prompts
• Interactive user input
• Platform-specific variants (Windows/Unix)
• Configurable output formatting:
  • markdown - Rich formatted output (default)
  • raw - Plain text output (ideal for commit messages, PR descriptions)

Example Commands

# Generate a commit message for staged changes (raw output)
flowai --command staged-commit-message

# Review code changes (markdown formatted)
flowai --command staged-code-review

# Create PR description (raw output)
flowai --command pull-request

Chat Mode Features

FlowAI's chat mode is a powerful way to interact with the AI assistant. You can:

1. Start a direct chat session:

flowai --chat

2. Turn any command into a chat session by adding --chat:

# Start with a code review and continue chatting about it
flowai --command staged-code-review --chat

# Generate a commit message and discuss it
flowai --command staged-commit-message --chat

# Create a PR description and refine it through chat
flowai --command pull-request --chat

When using --chat with a command, FlowAI will:

1. Execute the command normally first
2. Use the command's output as context for a new chat session
3. Allow you to discuss, refine, or ask questions about the output
4. Keep the original context (e.g., git diff, code changes) available for reference

Chat Features

• Stream mode toggle (/stream, /stream on, /stream off)
• Web search capability with Gemini models
• Thinking mode for Gemini 2.5 models
• Token usage tracking
• Real-time response streaming
• Command system for common operations
• Chat history persistence
• Markdown rendering
• Loading indicators with timing information

Chat Commands

• /help - Show available commands
• /quit - Exit chat mode
• /clear - Clear chat history
• /stream - Toggle stream mode
• /stream on - Enable stream mode
• /stream off - Disable stream mode

Known Issues

We are actively working on fixing several issues in the chat mode:

• Ctrl+C handling may not work correctly in some scenarios
• Status display (tokens and stream mode) may not show correctly in some terminals
• Double "Generating response..." message may appear
• Some formatting issues with streamed responses
• Terminal compatibility issues with certain commands

Please check our TODO.md file for a complete list of issues being tracked.

Installation

pip install flowai

Configuration

Run the initial setup:

flowai --init

This will guide you through:

1. Setting up API keys
2. Choosing your default model
3. Configuring stream mode preferences

Usage

Basic Commands

# Start chat mode
flowai --chat

# pipe output into flowai as context
git diff | flowai "summarise these changes in 1 paragraph"

# ask any question
flowai "how do i do a git rebase? Is it dangerous? Be concise"

# Generate commit message for staged changes (raw output)
flowai --command staged-commit-message

# Review staged changes (markdown formatted)
flowai --command staged-code-review

# Get help (markdown formatted)
flowai --command help

# Get specific help on any flowai feature
flowai --command help "how do i create a custom command that will work in windows and unix style platforms?"

# Use web search (only works with Gemini models)
flowai --web-search "What are the latest developments in quantum computing?"

# Enable thinking mode for Gemini 2.5 models
flowai --model gemini/gemini-2.5-pro --thinking-budget 2048 "Analyze this code for security issues"

Advanced Features

Web Search

FlowAI supports web search capabilities with Google's Gemini models, allowing you to access up-to-date information from the internet.

# Enable web search (only works with Gemini models)
flowai --web-search "What are the latest developments in quantum computing?"

# Use in chat mode
flowai --chat --web-search

# Combine with commands
flowai --command help --web-search "What are the latest features in FlowAI?"

When web search is enabled:

• The current date and time are included in the prompt
• The AI will cite its sources in a dedicated "Sources" section
• Citations include webpage titles and URLs
• Only works with Gemini models (automatically disabled for other models)

Thinking Mode (Gemini 2.5 Models)

Gemini 2.5 models support a "thinking mode" that allows the AI to perform more thorough reasoning before responding.

# Enable thinking mode with a specific budget (1024+ recommended)
flowai --model gemini/gemini-2.5-pro --thinking-budget 2048 "Analyze this code for security issues"

# Use in chat mode
flowai --chat --thinking-budget 2048

# Combine with commands
flowai --command code-review --thinking-budget 2048

Thinking mode:

• Allows the AI to perform more thorough reasoning
• Higher budgets (1024+) enable more complex thinking
• Setting to 0 disables thinking mode
• Status is displayed in the terminal output
• Only works with Gemini 2.5 models

Output Formatting

Commands can be configured to output in either markdown or raw format:

• Markdown format: Rich text with formatting, ideal for reviews and documentation
• Raw format: Plain text, perfect for commit messages and PR descriptions

You can:

1. Set format per command in prompt-index.txt
2. Override with --no-markdown flag
3. Default to markdown if not specified

Chat Commands

• /help - Show available commands
• /quit - Exit chat mode
• /clear - Clear chat history
• /stream - Toggle stream mode
• /stream on - Enable stream mode
• /stream off - Disable stream mode

Contributing

Please see our CONTRIBUTING.md for guidelines on how to contribute to this project.

License

MIT License - see LICENSE for details