THE AI
CTRLbot.AI // v0.7.0
most ai talks. this one asks.
CTRLbot.AI is a diagnostic AI that asks before it answers. You describe what is broken, it fires back fast YES/NO questions to narrow the problem, then delivers one targeted answer with real sources — and forgets you the moment you close it.
A CHATBOT SHOULD ASK QUESTIONS
instead of making assumptionsInference is powerful, and it is also how AI confidently walks you down the wrong road. When something is broken, the useful machine is not the one that guesses fastest — it is the one that interrogates. CTRLbot.AI asks the highest-value question it can, hands you the answer options so you barely have to type, and only commits to a solution once it has narrowed the field.
Every mainstream assistant answers your first sentence. When the inference is wrong, you have to notice it is wrong — which is the one thing you cannot do while troubleshooting something you do not understand.
Search has the opposite failure. It returns everything and diagnoses nothing.
ASK. NARROW. CONFIRM. ANSWER.
roughly thirty to sixty secondsAnswers you can click.
Questions arrive with their answers attached — YES, NO, and on the final question a PASS, each one explaining what that choice would mean for the diagnosis. You supply judgment, not prose.
There is no countdown on the final question. The session waits.
Within about a second it states what it thinks is going on. It exists so you see the machine engaging — and so a wrong turn is visible and correctable before anything is committed to.
YES/NO questions stream in roughly two seconds apart, personalized with the brand, device, symptom and error code pulled from your prompt. Answer as many as you like inside a 15-second window.
A longer assumption built from everything you answered, one real resource link, and a single sharper question whose buttons each explain what that choice would mean.
A collapsible playback of every question and answer collected, then three resource links with source, title and date, and the answer itself. Direct, actionable, in voice.
Then it ends. One answer, three sources, done — not an open-ended chat you have to manage. Press ✕ and the session is gone.
IT SKIPS THE QUESTIONS YOU ALREADY ANSWERED
the entry gateNot every prompt deserves the same treatment. At submit time a local, synchronous gate reads the prompt for specificity — brand, model or part number, concrete symptom, conditional clause, steps already tried, verbatim error evidence — and picks where the session enters. Length alone never earns a skip: a long ramble with no specifics stays at Stage 1, and a five-word error code goes straight to Stage 3.
| Entry | What you get | When |
|---|---|---|
| Stage 1 | The full path: assumption → YES/NO questions → assumption and targeted question → answer | The default. Vague, short, or general prompts. Garbage input is pinned here regardless. |
| Stage 2 | Skips the YES/NO questions. Straight to the assumption, one resource link, and the targeted question | The prompt already names the thing and the condition. |
| Stage 3 | Straight to three links and the answer | The prompt contains a pasted error, code, or log — a hard artifact, answerable with no clarification. |
Stage 3 is the only path with no confirmation step, so it gets an escape hatch: "Nope, that's not it" drops the session back to Stage 2, told not to repeat that angle and not to reuse the same links.
IT STARTED WITH ELIZA
mit, 1966Joseph Weizenbaum's DOCTOR script imitated a psychotherapist by reflecting what people typed back at them as a question. It had almost no intelligence, and people talked to it anyway — because being asked a question feels like being listened to. CTRLbot.AI takes that mechanism and points it somewhere useful. Instead of reflecting feelings back at you, it reflects the problem back at you, one narrowing question at a time, the way a technician works a diagnosis. The therapeutic voice was replaced by a blunt one, but the bet is the same one ELIZA proved sixty years ago: the right question does more work than a confident answer.
The voice is an IT admin who got called in on a Saturday. Blunt, dry, and never flattering — that tone is a trust mechanism. Nobody suspects this thing of telling them what they want to hear.
INSIDE CTRLbot IoT IT ANSWERS AS AV
the one integration pointCTRLbot IoT opens the assistant in a secured in-app view and tags the request, so the same engine behaves differently.
It opens in about a second instead of waiting 15, because in the app the FAQ is the feature — drivers, connections, live testing, presets, switching, audio, the EUI, settings and security.
"Switcher", "driver", "matrix" and "endpoint" resolve toward professional AV instead of general IT, and anything unclassifiable defaults to AV.
App questions are answered from the CTRLbot IoT documentation set, so the answers use the app's real screen names and workflow.
It does not discover devices, configure connections, or carry any device-control traffic. That stays on the local network between the Android device and the equipment.
WHAT IT KEEPS
short listThere is no login, no email, no sign-up, no sidebar of past chats, and no way to retrieve a previous session. The only thing stored in your browser is the colour theme, and that is local storage rather than a cookie.
No login, no email, no sign-up. Open the page and type.
Nothing to retrieve. Copy the answer before you press ✕.
No analytics and no advertising. The three sources are never sponsored.
Held in server memory while the session is live, cleared on ✕, and expiring on its own within the hour.
Prompt text is processed by the model provider through OpenRouter. Question matching and the FAQ banks run on the CTRLbot server with no external call.
Using the report control sends that question, that answer, your optional note and a random session ID to a moderation channel. Nothing else.
THE TRADE
what it deliberately is notCTRLbot.AI ends a session with an answer. The corrections available are rejecting the assumption and, on an immediate answer, "Nope, that's not it" — there is no open-ended follow-up chat. It needs an internet connection, the interface and content are English, and because nothing is saved, nothing can be recovered.
Answers are generated by a hosted large language model. The resource links exist so you can verify.
It answers and points at real sources, and it says plainly that it is not a professional advisor.
It answers anything, but the diagnostic machinery has nothing to bite on when the input has no subject.
No saved sessions, no conversation history, no paid tier, no API product, no separate mobile app.
Free to use. Thirty seconds is typical for a full session, not a guarantee — model latency varies.