mwchapel/tars-ai-agent

TARS Agent is a local voice chatbot styled after the robot TARS from the movie Interstellar, with a retro green on black terminal look and an optional desktop app built with Electron. It runs entirely against a local language model through a tool called LM Studio, so no cloud API keys or internet connection are needed just to chat. The bot has a configurable personality, with humor and honesty settings baked into its system prompt so its responses stay in character. It can optionally address the user by name, either a single name or one picked at random from a list each turn. If left alone for five to ten random minutes it can proactively break in with a trivia question or a hypothetical question, and it actually remembers having asked if the user replies later. Replies appear typed out character by character, with speech starting at the same time rather than after the text finishes. Voice input is handled fully locally using the Whisper speech recognition model, run in a separate process so it does not interfere with the Electron app. Voice output uses the Mac's built in text-to-speech by default, with an option to swap in a higher quality local voice server for more natural-sounding speech, including voice cloning or a choice of nine built in named voices. The assistant can also search the web through DuckDuckGo or an optional Brave Search API key when it decides a question needs current information like news, prices, or scores. There are two ways to use it: a terminal interface and a matching Electron desktop app, both sharing the same keyboard shortcuts for recording, clearing the conversation, toggling speech, and quitting. Both show a live analytics panel with token counts, response time, and turn count, and the Electron app adds a cosmetic oracle panel showing tarot cards, quotes, and similar flavor content generated by the same local model, plus a small animated bar graph that reacts to the spoken audio. It requires Node.js 18 or newer, a running LM Studio server, and the SoX audio tool for recording voice input, and the TTS and Electron build are built with macOS in mind while text chat and speech recognition should work on other platforms too. The README does not state a license for the project.