AG - AI Command Line Assistant

Overview

AG is a powerful command-line interface (CLI) tool that allows you to interact with OpenAI-compatible AI models directly from your terminal. It supports both single-query mode and interactive chat sessions, with advanced features like file/URL content integration, command execution, and grep-like filtering.

Key Features

• Multi-API Compatibility: Works with any OpenAI-compatible API (including Ollama, LiteLLM, etc.)
• Dual Model Support: Optimized for both OpenAI-style and Gemini-style model behaviors
• Smart Content Integration:
  • @file(path) - Directly include file contents
  • @url(URL) - Fetch and process webpage content
  • @search(query) - Web search results integration
• Built-in Command Execution:
  • Execute shell commands with !command syntax
  • View output directly in your session
• Intelligent Text Filtering:
  • AI-powered grep with natural language patterns (-g flag)
  • Pipe-aware input handling
• Enhanced Terminal Experience:
  • Real-time streaming responses
  • Optional reasoning visibility
  • Full terminal control with proper TTY management

Installation

pip install openai requests beautifulsoup4 wcwidth

Usage

Basic Query

ag "Your question here"

Interactive Mode

ag

With Piped Input

cat file.txt | ag "Your question about the file content"

AI-Powered Grep Filter

cat logfile.txt | ag -g "error messages related to database connections"

Include File/URL Content / Search from web

@file(/path/to/file.txt)
@url(https://example.com)
@search(your search content)

Command Syntax

Directives

• @file(path): Include content from a file
• @url(URL): Include content from a webpage

Shell Commands (interactive mode only)

• !command: Execute a shell command
• !clear: Clear the conversation history

Configuration Options

Option	Description	Default
--api-url	API endpoint URL	http://localhost:11434/v1
--model	Model name to use	deepseek-reasoner
--model-type	Model API type (openai or gemini)	openai
--api-key	API key	sk-... or OPENAI_API_KEY env var
--max-tokens	Maximum tokens for response	4096
--frequency-penalty	Frequency penalty (0.0-2.0, -1 to disable)	0.5
--presence-penalty	Presence penalty (0.0-2.0, -1 to disable)	0.5
-g/--grep-enabled	Enable AI-powered grep filter mode	False
-r/--hide-reasoning	Hide model reasoning content	False

Examples

1. Query about local files:

ag "@file(~/projects/notes.txt) summarize the key points"

2. Research a webpage:

ag "@url(https://en.wikipedia.org/wiki/Machine_learning) what are the three main types of machine learning?"

3. Interactive research session:

ag
> @url(https://news.ycombinator.com/)
> what are today's top AI stories?
> !curl https://api.example.com/data | jq .stats
> analyze these statistics

4. Web search integration:

ag "@search(2024 AI trends) summarize top 3 trends"

Keybindings (Interactive Mode)

• Ctrl+D: Submit current input
• Ctrl+C: Cancel current input or exit if input is empty
• Tab: Auto-completion (when readline is available)