Leeside

Rogue AI agents and automated clients now hit web apps faster than any person can respond: credential stuffing, scraping, scanning, and agents that misbehave at scale. Traditional firewalls either miss them or block opaquely, with no safe way to trial a rule before it breaks production. Leeside is a request-security layer that fixes this. It binds every request to a real actor (not just IP and user agent), scores it with an on-box classifier, and uses Meta’s Muse Spark Open to explain each threat in plain words and propose a blocking rule, with every signal the model cites verified against the actual requests and struck if unsupported. You clear each case in one keystroke. Leeside acts only in the mode you choose, watch, shadow, or enforce, so you can measure a rule in shadow and promote it to blocking only once you trust it. One command installs it into an existing Laravel app, with ASP.NET Core, Django, and Express adapters next.