Analysis updated 2026-05-18
Automate posting, liking, and replying on Twitter or Xiaohongshu through your own AI agent
Read a social media feed as structured JSON data for an agent to process
Run a persistent browser session on a VPS that a person can remotely control from a phone
| blueberriely/ai-social-browser | alex-nlp/denoiserl | anvarnarz/killspam-bot | |
|---|---|---|---|
| Stars | 35 | 35 | 35 |
| Language | Python | Python | Python |
| Setup difficulty | hard | hard | moderate |
| Complexity | 4/5 | 5/5 | 3/5 |
| Audience | developer | researcher | ops devops |
Figures from each repo's GitHub metadata at analysis time.
Requires a VPS, a secured tunnel, and a manually warmed up social media account before automation works safely.
This project sets up a browser that both a person and an AI agent share, so an AI can post, like, and read on Twitter and Xiaohongshu, a Chinese social platform, without getting flagged as a bot. The core idea is that automation gets caught less often when it uses a real, visible version of Chrome, not a hidden headless one, together with a login session that a human created themselves. The repo runs that visible Chrome on a server, using a virtual screen so it does not need a monitor attached. A person controls it from their phone to log in, solve captchas, and handle anything unusual, while the AI agent sends simple web requests to read the feed, like posts, reply, or post new content. The repo has two main parts. The first, called relay, keeps the shared browser running and streams its screen to a phone so a human can operate it remotely. The second, called actions, turns social media tasks into HTTP commands the AI agent can call, such as feed, read, like, reply, and post, and returns the results as structured data instead of raw web pages. The README stresses that the technique only works if the account itself is treated carefully. A brand new account that starts posting through automation right away gets flagged quickly, while an older account with a history of normal human activity is much safer to automate. It recommends letting a person use the account normally for several days before turning on automation, and even suggests using the same server's network location on a phone occasionally so it looks familiar to the platform. Security warnings are a big part of the documentation. The remote control port and the action service ports have no built in login protection, so they should never be exposed directly to the internet, only reached through a secured tunnel. Default limits on how often the agent can post or interact are kept low, since the project is meant for one person's personal AI agent rather than running many accounts at once or any kind of bulk posting. It is written in Python and licensed under MIT.
A shared browser setup that lets a person and an AI agent both control the same logged in Chrome session, so an agent can automate posting and browsing on Twitter and Xiaohongshu without tripping bot detection.
Mainly Python. The stack also includes Python, Playwright, Chrome.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.