What is xcurate-public?

Xcurate is a terminal first tool for reading X, formerly Twitter, and helping you decide who to reply to and what to say, without ever posting anything on your behalf. It fetches your timeline, removes duplicate content, ranks the tweets most worth a reply, and hands you a short list of candidates. A human always approves and posts every reply manually, and the tool has no setting that would let it post, like, follow, or message anyone automatically. The pipeline itself is a deterministic TypeScript program: it fetches tweets using your own logged in browser cookie, stores them in a local SQLite database, removes duplicates, ranks candidates, and exports them to a file. Topic classification runs on a small local model through Ollama, so no external AI service or paid API key is required for that step. The actual judgment, deciding which tweets deserve a reply and drafting the wording, is handled by Claude Code slash commands, using a voice file you write yourself to match how you actually talk. A daily digest command reads the candidates and your voice file, drops the weak ones, and drafts replies into a markdown file for you to review and post. Your X login cookie is treated as a secret: it lives only in a local ignored file, is never logged or printed, and a git hook blocks it from accidentally being committed. Scheduled fetching is intentionally gentle, running about once an hour with jitter and backoff so it does not hammer the site. All of your data, including fetched tweets, drafted replies, and your voice settings, stays on your own machine and is excluded from version control. Setup requires Node version 22 or newer, and optionally Ollama with a small instruction tuned model for the local topic classification step.