Viktor.comAn AI coworker that actually does the work
Promoted
Launching today
Unplug AI
Security layer to protect LLM applications from attacks
2 followers
Security layer to protect LLM applications from attacks
2 followers
Unplug is agent runtime security for LLM applications. It tracks where text came from (user vs retrieved vs tool output), scans for prompt injection and destructive actions, and enforces tool-call policy, with span-level redaction instead of binary blocking. Most guardrails nuke the whole message. Unplug finds the exact attack span and surgically removes it, keeping the rest usable.
Hey Product Hunt š
I'm Shivanshi, the member in team of Unplug.
Unplug is agent runtime security for LLM applications. If your AI reads web pages, emails, or uploaded files before answering and you can't afford it silently getting hijacked by a hidden instruction in that content then this is for you.
āļø Here's how it works:
Drop an agent in front of untrusted content. Every web page, email, or file your agent reads is scanned before it reaches the model. Because that's the real attack surface: the injection isn't coming from the user's prompt, it's hiding in a webpage that tells your agent to ignore its instructions and exfiltrate your API keys.
Watch it find the attack, not nuke the message. Most guardrails either block the whole input (killing useful content) or let everything through (dangerous). Unplug finds the exact malicious span and surgically removes just that replaces the attack, the rest of the page survives. You get to keep the content that matters.
Check the actions, not just the text. The poisoned agent tries to send those API keys somewhere? That tool call gets flagged for review. A destructive command? Blocked outright. Unplug protects what your agent does, not just what it reads.
Run it fully offline. No cloud, no data leaving your infra, no per-request API cost. One line upgrades detection from regex to a local ML model.
Unplug the bad AI: Find the attack. Cut the attack. Keep the rest.
Thanks for checking us out. I'll be here all day.