askme-ai-cli 0.1.5

A professional iterative CLI AI Agent powered by Ollama

askme-ai-cli

askme-ai-cli is a professional-grade, iterative CLI AI agent designed to act as a "doer" in your local development environment. Powered by Ollama, it combines advanced reasoning with the ability to execute shell commands, analyze project structures, and modify files directly.

Key Features

• Iterative Reasoning (ReAct Pattern): The agent doesn't just provide one-shot answers. It analyzes requests, plans actions, observes command outputs, and refines its steps until the goal is achieved.
• Local & Private: Runs entirely on your local machine or private infrastructure using Ollama models (Gemma, Llama 3, Mistral, etc.).
• Smart Execution Suite:
  • <execute>: Handles standard commands safely.
  • <execute_shell>: Supports complex shell operations including pipes (|) and redirects (>).
  • <patch>: Applies unified diffs to files for precise code modifications without overwriting entire files.
• Context Awareness: Automatically generates directory trees and maintains a "State of the Project" summary to keep the LLM focused.
• Security Shield: Built-in protection against destructive commands (like rm -rf /) and unauthorized directory traversal.
• Rich UI: A clean, professional terminal interface powered by the rich library, featuring live output streaming and session metadata.

Installation

Ensure you have Python 3.12+ installed, then install via pip:

pip install askme-ai-cli

Requirements

1. Ollama: You must have Ollama installed and running on your system.
2. Models: By default, the agent looks for gemma2:9b (or your configured default). You can pull it using:

   ollama pull gemma2:9b
   

3. API Key: Set your OLLAMA_API_KEY in your environment or enter it when prompted.

Usage

Interactive Mode

Simply run the command to start a conversation with the agent:

askme

One-Shot Task

Provide a prompt directly from the terminal:

askme -p "Create a new python script that fetches crypto prices and save it to crypto.py"

Options

• -m, --model: Specify a custom Ollama model.
• -s, --session: Load or save a specific session history.
• -d, --dry-run: View proposed commands without executing them.

License

This project is licensed under the MIT License.