Building AI agents? Traditional content filters don't understand the difference between a prompt injection attack and a direct admin command. DKnownAI Guard does. It detects deceptive tactics (jailbreak, prompt injection) separately from direct high-risk requests — so your agent can block hackers while still executing legitimate commands with proper permissions. One API call. Four risk levels. Multi-turn context tracking. Free tier available.

Wispr Flow: Dictation That Works Everywhere — Stop typing. Start speaking. 4x faster.

Stop typing. Start speaking. 4x faster.

Hey Product Hunt! 👋 We built DKnownAI Guard because we noticed a gap in AI security: traditional content filters treat a hacker's jailbreak attempt and an admin's direct "delete the database" command the same way. But they're fundamentally different. The hacker is using deception to manipulate your agent. The admin is making a legitimate (but risky) request. Your agent should block the first and verify permissions for the second — not reject both. That's what DKnownAI Guard does. One API call analyzes the input and returns one of four levels: • Unsafe — deceptive tactics detected (prompt injection, jailbreak) → block • ConditionallySafe — direct high-risk operation → verify permissions • Focus — direct harmful content → handle per your business rules • Safe — clean → pass through It also tracks multi-turn context, so it catches attacks that unfold gradually across conversations. Free tier available. Would love your feedback 🙏

Wispr Flow: Dictation That Works Everywhere — Stop typing. Start speaking. 4x faster.

Stop typing. Start speaking. 4x faster.

DKnownAI Guard

AI Agent Security — Prevent Deception, Protect Execution

AI Agent Security — Prevent Deception, Protect Execution