faskety/onchain-alpha-radar

On-Chain Alpha Radar is a local-first tool for spotting new crypto projects while they are still just raw activity on a blockchain, before they show up on listing sites or trading dashboards. It watches Ethereum, Base, and BNB Smart Chain for fresh tokens, NFT mints, new trading pools, staking or claim contracts, and other early-stage contract activity, then records anything that looks meaningful. Once a contract is flagged, the system pulls metadata from Etherscan to understand what kind of contract it is, then checks for an official Twitter or X account and website to confirm the project is a real, identifiable effort rather than noise. Only contracts with matching official-looking accounts or websites can earn a medium or high score, which helps separate genuine early projects from random chatter and scanner bot posts. Everything is scored and stored locally in SQLite, so nothing depends on an outside server. The system is built to survive interruptions. It tracks exactly which blocks on each chain have already been scanned, so if a scan is stopped or a computer restarts, it picks up from where it left off instead of losing progress or missing blocks. The work is split into separate roles: a scanner that watches for new activity, a classifier that filters out low-value contracts, and an enricher that looks up social and website evidence and updates scores. A local dashboard, run through your own browser at a fixed address, shows chain activity, how many candidates are waiting for review, worker health, and score breakdowns. Settings like API keys are managed through the dashboard and stored in a local configuration file that stays out of version control. The project can be run as a single one-time scan, continuously in the background, or across multiple chains in sequence. It is written in Python and controlled through PowerShell scripts, and is aimed at someone who wants to independently monitor early on-chain activity without relying on third-party alpha or listing services. The full README is longer than what was shown.