askexort/exort

Exort is a Python command-line AI agent that can decide to use tools rather than only chat. The author frames it as something built for hobbyists and small projects rather than enterprises, with the headline pitch that it costs zero dollars to start because it works out of the box with Groq's free API tier or with a fully local Ollama model. What separates an agent from a chatbot, in the README's framing, is that Exort can reason about a question, pick a tool, run it, read the result, and loop until it has an actual answer. The example given is asking for the top three Python web frameworks and their GitHub star counts: Exort runs a web search, then fetches each project's GitHub page, then writes a comparison table. The built-in tool set covers web search through DuckDuckGo, fetching any URL, reading and writing files, listing and searching directories, running shell commands, executing Python snippets, and loading images for vision models. Installation is two commands: clone the repo and run pip install. Three quick-start paths follow. Groq is the recommended starter because it is free and fast. Ollama is the offline option, picking up any local model the user has pulled. OpenAI or Anthropic keys plug in the same way for paid use. Custom OpenAI-compatible providers like Together, OpenRouter, DeepSeek, Mistral, LM Studio, and vLLM can be added by editing a YAML config file. Inside, the project is split into an engine that runs a perceive, reason, act, reflect loop, a CLI shell, a YAML config loader, per-provider backends, a tool registry the author calls the GearBox, a SQLite store for conversation history, a playbook system for markdown knowledge files, and a Telegram bot frontend. The MIT-licensed code is presented as deliberately simple and easy to extend.