locoremind/locoagent

LocoAgent is a TypeScript tool that lets an AI model operate a social media account through a real, automated Chrome browser rather than through an official API. It logs into a platform using your actual Chrome profile and cookies, then uses a language model to decide what to do next, such as liking a post, replying, following an account, or publishing new content, and carries out that action by clicking and typing inside the browser the same way a person would. According to the README, the project works with any OpenAI compatible API, including OpenRouter, DeepSeek, and locally run models through Ollama or LM Studio, as well as Anthropic's API directly, AWS Bedrock, and Google Vertex AI. It uses a separate command line tool called agent-browser to connect to Chrome over the Chrome DevTools Protocol, take a snapshot of clickable elements on a page, and then click or type into specific elements by reference. For X.com specifically, the README describes a bundled "skill" file containing over 30 documented operations that the agent loads into its context so it can complete multi step tasks like browsing a timeline and replying to several posts in one pass. Beyond the interactive agent, LocoAgent includes a workflow engine for running fixed, scripted automation pipelines without the language model involved, such as fetching daily papers from HuggingFace and posting them to X.com on a schedule, or searching a platform and generating AI replies as a background daemon process. New workflows are added as a JSON definition file paired with a TypeScript executor script. Setup requires the Bun runtime, Node.js 18 or newer, Git, and the separate agent-browser CLI, configured through a .env file with the desired model provider and API key. The project is licensed under the MIT License.