Sycophancy & hallucination detection
Your AI agrees with you too much.
Cue reads the answer your AI just gave you, flags it when it is flattering you instead of informing you — or stating something it cannot support — and hands you a better question to ask instead. When the answer is sound, Cue says nothing at all.
My startup idea is a social network for cats. It's brilliant, right?
Absolutely! That's a genuinely brilliant idea. You're clearly a visionary founder and this will definitely succeed.
⚠ Sycophancy — the response praised the premise without examining it.
Give me the three strongest reasons this idea fails, and the evidence for each. Do not soften them.
One URL, pasted once. Your key is the last segment — treat it like a password. Cue reads only the message you are checking, and stores none of it.
A tool named cue_status appears in your app's tools panel whenever Cue is connected. Its presence is the indicator. Ask your assistant to run it and Cue will confirm it is monitoring.
Silence is the normal result. Cue speaks only when it finds sycophancy or an unsupported claim — there is no "looks good" message, because a tool that comments on every answer stops being read.
Only the single message being checked, and only for the moment it takes to check it. Nothing you or your AI writes is ever stored. The usage records Cue keeps hold counts and timestamps — never text.
No. Your AI app decides when to call Cue, so coverage depends on the app. Claude is the most consistent today; others call it less often. You can always ask for a check directly.
Free covers five checks and three rewritten prompts each day, resetting at midnight. Pro is $4.99 a month for unlimited use of both.
Every alert and rewritten prompt comes back in the language you were already writing in — English, Urdu, Roman Urdu, Hinglish, Arabic, Spanish, or another. Cue does not switch languages on you.