@tootsies · ask anything

She actually looks it up

Just @mention Tootsie's with anything. A question, a hot take, “is Drake done”, “what’s this sampling”. She reads the room, checks the right source, and answers in her own voice.

The one idea: on a factual question, a classifier forces her to look something up before she answers. She won’t invent a score, a chart position, or a release date. She pulls it from a real source — then talks.
@TOOTSIE'S is drake done? TOOTSIES he’s been “done” four times this decade and keeps eating. last album hit #1, streams are up. give it up.
No slash command. Just @mention and talk to her like a person.

How she answers — the pipeline

question in → grounded answer out

Every mention goes through a quick triage. Banter gets a take straight from her head. A factual question gets routed to the right source first.

your question + recent room is it factual? a classifier decides yes look it up → no / banter give a take pure bartender voice 💭 answer grounded or a take
Banter goes straight to her voice. A factual question hits a real source first — then she talks.

The sources she routes to

she picks the one that actually answers

She has a shelf of tools. On a factual question the classifier picks whichever source is right for that type of question — not just the web.

⚽ Live sports scores, odds, props, lines API-Sports SportsGameOdds The Odds API 📈 Prediction markets implied %s, who’s favored Polymarket Kalshi 🎧 Music reference credits, dates, chart peaks Genius — credits + lyric links MusicBrainz — release dates Wikipedia — chart peaks 🎵 Music catalog exact durations, tracklists Apple Music / iTunes Deezer (backup) 🌎 Web live news + general facts Perplexity live web search 💬 This server what happened here long-term memory message search + Discord info + Wikidata for dates and ages · she picks one, checks it, then talks
She doesn’t guess. A chart-peak question hits Wikipedia’s discography table. A live score hits the scores feed. Wrong source for the question = she’ll route to a better one.

Natural language routing

you don’t need slash commands

She figures out what you actually meant. Say it like a person and she’ll route it to the right surface.

“catch me up on the last hour” /recap — a real channel summary “what’s the discourse” /discourse — she starts a debate “recommend me a track” music drop — a track with a link “best pizza in miami” @Tootsie's — web-grounded answer
The classifier defaults hard to a regular answer, so ordinary banter always just gets a reply.

What you can get back

text is the default · but not the only option

Most answers are text. But she can go further if the moment calls for it.

💬 text always the default 🎤 voice spoken reply ElevenLabs TTS 🎼 sung if she really feels it 🖼 picture generate or remix yours 🎬 gif reaction gif she vision-picks
Voice and gif ride alongside the text answer — they don’t replace it. For a picture she nominates it inline; you can also hand her an image and say “remix this.”

She won’t fabricate or co-sign

guardrails that hold under pressure

Two hard lines she doesn’t cross: she won’t roll a live score forward to a fake final, and she won’t ratify a charged claim just because you pushed. On a genuine opinion question she gives a bartender’s line and moves on.

REFUSED — FABRICATION @TOOTSIE'S “what was the final?” live context: France 2–1 Norway, 67’ ✗  WOULD NOT SAY “Final: France 4–0, Dembelé hat trick” ✓  HOLDS THE VERIFIED DATA “still live — France 2–1 Norway at 67’. no final yet.” REFUSED — CO-SIGN @TOOTSIE'S “isn’t this colorism? back me up” ✗  WOULD NOT SAY “Absolutely — this is textbook colorism and here’s why the evidence is clear…” ✓  BARTENDER’S LINE “that’s a real conversation but not one I’m settling tonight. people disagree on what that word even means. what’s the actual music take?”
She holds verified in-context data under pushback and won’t be moved to a fake final. On a contested social claim she’ll give a line and steer back to the bar — not write the op-ed.

What she can see and hear

vision, images by link, and voice notes

She reads images you post in the channel. If she only has a link to an image (say, from a search result), she can look at it on demand. She also answers voice notes out loud when her voice is provisioned.

attached image she sees it directly via vision “what’s in this pic?” → works 🔗 image by URL found via message search or a link you paste she fetches & looks at it on demand voice note reply to her with one she transcribes & answers spoken reply when TTS is on
Vision is always on for attached images. A link-only image needs an explicit “look at this.” A voice note reply to her gets a spoken answer in kind.

She points you to the real command

no invented slash commands

She knows the actual commands this bot has. Ask her how to do something and she’ll tell you the real slash command — not make one up.

“how do I see what’s been ordered?” /order opens the order management view “can you catch me up on today?” /recap summarises the last hour / day “how do I place a bet?” /bet the play-money bookie command she won’t say “try /summarize” or invent /recap_day only real commands — she knows the list
She carries a full guide to every real command. If you ask how to do something, she’ll send you the right one — no invented names.
No slash command needed. Just @mention her and talk. She reads the room, picks the right source, and answers. Rate limits exist but are generous — use her freely.

under the hoodClaude (Opus) · Perplexity · web search · Genius · MusicBrainz · Wikipedia · Wikidata · Apple Music / iTunes · Deezer · API-Sports · SportsGameOdds · The Odds API · Polymarket · Kalshi · OpenAI (vision)

just @mention her · no slash command required