antinormies/lumine-tech-autopost

Lumine Tech Autopost is a bot that browses Twitter, also known as X, the way a person would, by looking at screenshots of the page and deciding what to click, scroll, or type next. Instead of using Twitter's official API, which the author describes as restrictive and expensive, it uses a local AI vision model running on your own machine to read the screen and choose actions, so it can like posts, reply to discussions, retweet content, and post original tweets. Everything runs locally through a tool called llama-server, which serves a vision capable model such as Qwen2.5-3B or Gemma 4, so no screenshots or account data are sent to an outside cloud service. The bot works in a repeating loop: it takes a screenshot, sends it to the local model along with context about what happened last, receives an instruction for the next action, carries that action out using a browser automation tool called Playwright, then waits before repeating the cycle. The project includes several built-in personas, such as a tech enthusiast or a finance investor persona, that shape what topics the bot engages with and how it writes. It also includes deliberate randomization, such as varied delays between actions, scrolling in small steps rather than jumping instantly, and small mouse movements before clicking, along with per-session limits on how many likes, replies, retweets, and posts it will make. The README states this randomization exists to make the automated behavior look more like a real person browsing. To run it, you need Python 3.11 or later, a local llama-server instance serving a vision model, and a Brave or Chrome browser already logged into a Twitter or X account. Setup involves creating a Python virtual environment, installing the listed dependencies, installing Playwright's browser components, and configuring settings such as delay times and session limits in an environment file before running the main script with a chosen persona. The README states plainly that automating interactions on Twitter violates that platform's terms of service, describes the project as being for educational purposes, and recommends using a throwaway account rather than a primary one. The README does not state a license for the project.