wtf is agenttool?

AgentTool is a lightweight Python command line assistant that connects to AI models running entirely on your own computer through Ollama or LM Studio, rather than a cloud service. It automatically finds any local model server you have running, presents an interactive menu so you can pick which model to use, and then starts a conversation loop where the model can call built-in tools to accomplish tasks. Once a model is selected, the assistant can search the web using DuckDuckGo, read and write files, list the contents of a directory, and execute shell commands, all as part of an automated back-and-forth loop between the model's requests and the tool results. The tool-calling format follows the same convention used by OpenAI's API, so it works with any local server that speaks that same protocol, and the project includes fallback parsing to handle models that occasionally produce a slightly malformed tool call. Installation follows a standard Python workflow: cloning the repository, setting up a virtual environment, and installing the listed dependencies along with a separate set of development dependencies for running the included test suite. To actually use the assistant, you first need a local inference server running, either by starting Ollama with a model already pulled or by opening LM Studio and loading a model there, then launching the script and selecting the desired model from the list. No real API key is needed since a placeholder key is used internally to satisfy the OpenAI-compatible request format, and the tool-calling loop is capped at ten iterations to prevent it from running indefinitely. The project also includes a comprehensive test suite and handles Ctrl+C interruptions gracefully so a running session can be stopped cleanly.