What is local-ai-agent?

This project is an AI assistant that lives entirely in your terminal and runs against your own local Ollama installation, so nothing you type ever leaves your computer unless you explicitly ask it to search the web. There is no account to create and no API key to buy. The assistant remembers you across conversations. Chats are saved and can be resumed later, and it quietly pulls out lasting facts from what you tell it, then recalls them later based on meaning rather than exact keywords. When you add /web to the end of a message, it goes and searches the internet using DuckDuckGo, visits the relevant pages in real time, and answers using what it found, showing you its sources. It can also run commands on your computer and read or write files: safe actions run automatically, riskier ones ask for your permission first, and genuinely dangerous actions are refused outright, with everything it does written to a log you can review. It also pauses its background work if it notices you are playing a game, so it does not compete for your computer's resources. Getting started involves running an install script for Linux, or an experimental one for Windows, which checks your system, sets up the needed tools, and offers to install Ollama and download the two AI models required, roughly six gigabytes total. After that you can chat through a simple command menu or a direct chat command. The chat model must specifically support something called tool calling, meaning it can trigger actions like running commands, and the readme warns that some models will pretend to complete a task rather than admit they cannot, if this feature is missing. Everything the assistant knows about you is stored in a single file on your machine, combining the conversation history, extracted facts, and a search index built from them. The project is released under the MIT license, and the author says it is intentionally kept small and easy to read and modify.