Nelle — Personal AI Assistant

🏠 Dashboard

Your daily overview in one place — upcoming events, pending tasks, a stocks/budget/Bitcoin snapshot, your shopping lists, and a Quick Chat box for a fast question without leaving the page (works the same as the full Chat, including attaching photos). A System Health card also shows a quick red/green status for whether the server and its connected services are all running normally.

💬 Chat (AI Assistant)

Talk to your personal AI. Ask questions, get advice, brainstorm ideas, or just have a conversation. Your chat history is saved between visits. Tap the 🎤 mic button to speak instead of type.

📷 Attach photos — up to 4 at once, in any chat (the main Chat, the Dashboard's Quick Chat, and the Writing Center's chat all work the same way). Nelle can look at them and describe what she sees, answer questions, or give feedback. Add a caption to ask something specific, or just send the photos on their own.
🎨 Attach from Image Studio — pick a photo you've already generated or edited instead of uploading a new one, so Nelle can look at and discuss something you made.
Ask her to draw something — say things like "draw me a corgi in a spacesuit" or "make an image of a sunset over mountains" and Nelle generates it right in the conversation, then saves it to your Image Studio gallery. She defaults to Low quality to keep it fast and cheap — ask for "high quality" if you want more detail.
Ask about the weather — "what's the weather like in Las Vegas this weekend" or similar for any city — she'll pull a live multi-day forecast.
Budget, Vegas Trip, and My Files — she can add, update, and delete things in all three directly from a conversation now, not just answer questions about them. My Files access is always limited to your own private files, never the other household member's. See those sections below.
🔊 Voice orb — a small colorful, glowing ball appears above the mic button whenever she's speaking a reply out loud, pulsing along with the actual sound of her voice.
👂 Wake word — say a chosen word out loud to start talking to her, instead of pressing the mic button first. Defaults to your AI's name, but you can set it to anything in Settings under "Wake Word" — a short, common word (like "Hello") is often heard more reliably than a less common name. Off by default. Only works in Chrome or Edge, and only listens for that one word in the background; typing and the mic button both still work exactly the same either way.

A conversation is automatically deleted 30 days after the last message in it — the small line above the chat always shows exactly when that will happen. Sending a new message resets that 30-day timer.

📅 Calendar

Add and manage events. Tap any day to create a new event. Turn on Shared to make an event visible to everyone in the household.

✅ Tasks

Your personal to-do list. Tap a task to check it off. Set a priority level and an optional due date. Use the toggle at the top to show or hide completed tasks.

🛒 Shopping

Create multiple lists (e.g. Groceries, Hardware). Check items off as you shop without removing them — tap Clear Checked when you're done. Lists can be shared with other household members.

📝 Notes

Jot down anything — ideas, recipes, reminders. Use the search bar at the top to find notes by keyword. Everything saves automatically as you type.

🎰 Vegas Trip

Ideas for the Sept 5–10, 2026 trip, grouped by day. Use the tabs at the top to filter by category (Shows, Dining, Day Trips, Nightlife, Casino, Relax). Each idea's status can be moved from Idea to Planned to Booked to Done as decisions get made — the colored left edge of each card reflects its status. Tap "+ Add Idea" to add your own finds; anyone can edit or remove any idea.

Ask Nelle — in chat, ask her to add an idea ("add the Bellagio Fountains for Friday evening"), mark something done or booked, or ask what's planned for a given day — handy for handling things on the fly while you're actually there.

📈 Stocks

Track your holdings, watchlist, and stocks you're watching. Live prices come from Yahoo Finance. Each stock also shows recent news and key price numbers. Add a stock with the + button.

Small badges on a stock card — quick technical signals, not something you need to act on:

RSI — a 0-100 number showing how heavily a stock has been bought or sold over the last couple of weeks. Above 70 (📈) means it's been bought up a lot lately and could be due to cool off; below 30 (📉) means it's been sold off a lot and could be due to bounce back. Anywhere in between is unremarkable — no strong signal either way.
✨ Golden Cross / ☠ Death Cross — compares the stock's average price over the last 50 days against its average over the last 200 days. A Golden Cross (the shorter average has climbed above the longer one) is generally seen as a bullish sign; a Death Cross (the shorter average has fallen below) is generally seen as bearish. Neither is a guarantee — just a signal traders watch.
🔥 Vol [number]x — today's trading volume compared to its normal 30-day average. Only shows up once volume is at least double normal, which usually means something (news, an earnings report, etc.) has people paying unusual attention to it today.

📰 News

Top headlines from major news sources. Updates automatically every few minutes. Tap a story to read the full article.

🚗 DoorDash

Tracker — Start and end a dash to log your shift automatically.
Shifts — Record earnings, miles driven on deliveries, and odometer readings to make tax time easier.
Expenses — Keep track of gas, repairs, and any gear you buy for work.
Proof of Delivery — A photo log with your location recorded at each stop. Tap a card to see the details and photo. Tap the address to open it in Google Maps.

💰 Budget

Accounts — Add your checking and savings accounts with a starting balance. The balance shown updates automatically as you log transactions — there's no bank connection of any kind, everything here is typed in by hand.
Investment accounts — Add Robinhood, Fidelity, or any brokerage account as an "Investment" account type. These work differently: no transactions, just a balance you update yourself whenever you check your brokerage app — since the value moves with the market, not from deposits/expenses you'd log one by one.
Categories — Set up spending categories like Groceries or Gas, each with an optional monthly target. Mark a category as "Money in (income)" — like a paycheck category — and its bar fills as money comes in instead of goes out. An expense category's bar turns red once you go over its target.
Transactions — Log money in (a deposit or paycheck) or money out (an expense), pick which account and category it belongs to, and a date. Use the month picker at the top to look back at a previous month.
📷 Scan Receipt(s) — Instead of typing a transaction by hand, snap or upload a photo of a receipt and Nelle reads it — merchant, total, date, and a best-guess category — then pre-fills the Add Transaction form for you to check over and save. You can pick several photos at once (either multiple separate receipts, or a few shots of one long receipt) and she'll walk you through reviewing each one in turn. Nothing is ever saved automatically — she only fills in the form, you still have to hit Save. The photo itself isn't kept anywhere, just read once and discarded.
Filters — Above the transaction list, narrow it down by account, by category, or by typing part of a description — handy once the list gets long.
Recurring bills — Check "Recurring" when logging a bill that repeats every month (rent, a subscription). Next month, tap "🔁 Add Recurring Bills" and it re-logs each one on the same day of the new month automatically — safe to tap more than once, it won't create duplicates.
Pending transactions — Check "Pending" for a charge or deposit that hasn't cleared the bank yet. It shows up in its own "⏳ Pending" section, separate from your regular transaction history, same idea as the pending area on an actual bank statement — and it still counts toward that category's spending for the month, but the account's balance doesn't change until it clears. Tap "✓ Cleared" on a pending item (or uncheck "Pending" from its edit form) once it actually posts, and it moves itself down into the regular Transactions list automatically. Any account with pending transactions shows a small "pending, not yet available" note under its balance so you know why the total hasn't moved yet.
Deleting a category doesn't delete its past transactions — they just become uncategorized. Deleting an account removes its transactions too.
Ask Nelle — in chat, ask things like "how much have we spent on groceries?" or "are we over budget on anything?" and she'll pull your live current numbers. She can also log or delete a transaction for you ("log a $40 grocery expense") — if she's not sure which transaction you mean to delete, she'll ask rather than guess.

📖 Writing Center

Library — All your books in one place. Tap a book to open it. Tap the 🗑️ icon on a card to delete it.
Chapters tab — Add chapters, see how many words each one has, and open the writing editor. A progress bar appears if you've set a target word count for the book.
Notes tab — A place to keep character descriptions, world details, and plot notes. The AI reads these so it understands your story when helping you write. A character note can also have a voice assigned to it (pick from ElevenLabs' full voice library, with a preview button) — once at least one character has a voice, whole-chapter Read Aloud automatically narrates in multiple voices instead of one flat reading. See 🔊 Read Aloud below.
Streak badge — Next to the book title, shows your current writing streak and how many words you've written today. Appears once you've saved a change to the book.
⬇ Export (book view) — Downloads the entire book as one .docx file, with a title page and each chapter starting on a new page — see Standard Manuscript Format below.
🎨 Art tab — Shows any AI-generated art attached to this book (cover art, character portraits). Generating and attaching art itself happens in the separate 🎨 Image Studio page — see below.
📖 Story Bible tab — A reference document for this book: characters, locations, timeline, world rules, and anything else you want to establish as canon, organized into 14 sections (pre-filled as a starting template). Below it, a separate 🤖 AI Working Memory — Non-Canon box is a short working summary Nelle keeps and updates herself as she works (new characters, current plot situation, unresolved threads). You can edit or clear it too, but it never counts as established fact on its own — it's Nelle's notes, not canon.
📚 Series Bible tab — Appears once a book is assigned to a series (set this from the book's Edit form — "+ New Series…" creates one on the spot). Holds canon that spans the whole series — recurring characters, shared world rules, series-wide timeline — separate from any one book's own Story Bible. When writing a book that's part of a series, Nelle checks both the Series Bible and that book's own Story Bible for conflicts.
Continuity checking — Whenever Nelle continues, critiques, edits, or checks your story, she compares new material against the Story Bible (and Series Bible, if this book belongs to one) — and separately checks whether her own Working Memory still agrees with it. If something conflicts — a character's age, a location's geography, an established rule, or even her own memory disagreeing with the Bible — she flags it and quotes the conflict instead of silently changing your story, the Bible, or her memory to make it match. You decide which one is right.
Editor — Your full writing area. Everything saves automatically as you type. Use the toolbar at the bottom for AI help:

📂 Import	Bring in a file from your device to use as chapter text
⬇ Export	Downloads just this chapter as a .docx file in Standard Manuscript Format
🎙️ Dictate	Speak and have your words typed directly into the story — a rest from typing. Tap it, talk, and it types automatically as you pause between sentences. Tap it again to stop. Say "new paragraph" or "new line" to add a break without touching the keyboard.
📝 Scratchpad	A separate freeform space (one per book) for jotting down ideas, pasting research, or drafting something before it's ready to be real prose. Select text and tap "Insert at Cursor" to move it into the chapter you're currently working on.
💬 Import Chat	Bring in a conversation from your ChatGPT data export. Pick the .zip (or conversations.json), choose a conversation, then tick whole messages or highlight just a passage and send it to the chapter, Scratchpad, a new Note, or the Story Bible. Every part is marked Human or AI. It all happens on your device — nothing from the export is saved unless you send it into your book. Passages sent into a chapter get the same soft purple AI highlight if they came from the AI.
🤖 AI text marks	Anything Nelle writes into your chapter (Continue, Rewrite, Line Edit, or when she adds a paragraph in Chat) gets a soft purple highlight so you always know what is yours and what is hers. Tap the 🤖 button at the top of the editor to switch between three views: highlighted, marks hidden, and 👤 Mine only (her text fades out so you can see just your own words). Tap ✔ Accept as mine to make her text count as yours — select some text first, or just put the cursor inside her passage; with neither, it offers to accept the whole chapter. Stats shows the split, like "82% yours, 18% Nelle", for the chapter and the whole book. The marks never show up in your exported .docx.
📊 Stats	Instant writing metrics for this chapter — word/sentence counts, adverb use, rough passive-voice count, dialogue percentage, a readability grade level, and your most repeated words. Computed directly from your text, no AI cost.
▶ Continue	The AI picks up where you left off and writes the next section in your style
🔍 Critique	Get honest feedback on what's working and what could be stronger
✏️ Line Edit	Cleans up grammar, word choice, and flow — gives you the improved version to copy in
🔄 Rewrite	Rewrites a passage to make it clearer or more powerful
🔤 Synonyms	Select a single word or short phrase first — suggests synonyms and antonyms that fit the surrounding sentence, not just a generic thesaurus list
📋 Blurb	Writes a short back-cover description of your book
💡 Titles	Suggests five possible chapter titles
🔊 Read Aloud	Reads the selected text (or the whole chapter, if nothing's selected) out loud — great for catching sentences that read fine on the page but sound clunky when spoken. If any character in your Notes has a voice assigned, reading the whole chapter automatically splits it by speaker and narrates in multiple voices instead of one flat reading. Tap it again to stop.
🔬 More Tools	Opens a menu of deeper writing tools, grouped by category (see below)
💬 Chat	Have a free-flowing conversation about your story with the AI. Tap 📷 in the chat bar to attach a reference photo — a character, an outfit, a location — and the AI will look at it and help you describe or use it in your writing. Tap 🎨 to attach a photo you've already made in Image Studio instead of uploading a new one — handy for reusing a character portrait you generated earlier. Tap 🎤 to talk instead of type — Nelle answers back in the chat in writing, and (unless muted) reads her reply out loud too. Tap 🔊/🔇 next to the mic to mute or unmute her spoken replies in this chat — it's off by default, and only affects this chat's voice replies, not Read Aloud or Dictate. If you ask her to add something to your story (like "add a paragraph about the storm"), she'll write it directly into the open chapter for you, formatted into proper paragraphs.

Tip: Highlight a section of text before tapping Line Edit, Rewrite, Critique, or most tools in the More Tools menu to focus on just that part. Otherwise the AI works with the whole chapter (or its most recent portion).

🔬 More Tools menu:

🧬 Consistency Check	Flags contradictions against your Notes and Story Bible — a description, name, or fact that doesn't match
🕒 Timeline Check	Flags timeline and continuity slip-ups against the chapter and your Story Bible — inconsistent days, times, or events
🔁 Repetition Check	Points out words or phrases you lean on too often, with alternatives
📈 Pacing Check	Flags sections that feel rushed or that drag
🎭 Show, Don't Tell	Finds passages that tell the reader something instead of showing it, with suggestions
🎯 Genre Beat Check	Checks the chapter against the story beats readers typically expect for your genre
🗺️ Chapter Outline	Suggests a chapter-by-chapter outline based on your book's story summary
🧩 Suggest Notes	Reads the chapter and lists characters, places, or story elements you haven't added to your Notes tab yet
✉️ Query Letter	Writes a query letter pitch paragraph for submitting to literary agents
📄 Synopsis	Writes a one-page synopsis for agent submissions — unlike the Blurb, this includes the ending
🗣️ Dialogue Polish	Makes each character's dialogue sound distinct and natural, without changing what's said
🎭 Voice Check	Checks selected dialogue against a character's personality and voice as described in the Story Bible and Notes, and flags lines that feel out of character. Only compares against what's written there — not against how the character has actually spoken in other chapters.
⏪ Chapter Recap	A short "previously, in your story..." summary to help you jump back in after time away

Editor toolbar extras:

↩ ↪ Undo / Redo	Step backwards or forwards through your changes — works on mobile too
⬜ Focus mode	Narrows the page to a comfortable reading width so it feels more like writing on paper
🔍 Find & Replace	Search for any word or phrase and swap it out — one at a time or all at once. Also opens with Ctrl+H
🕐 Version History	Recover an earlier draft of this chapter. A version is saved automatically right before the AI edits your chapter, or tap "Save a version now" to save one yourself anytime. Restoring always saves your current text as a version first, so nothing is ever truly lost.
Line spacing	The dropdown near the alignment buttons at the top of the editor (labeled Single by default) — switch to 1.5× or Double, e.g. for manuscript formatting

Standard Manuscript Format: both Export buttons produce a .docx file formatted the way literary agents and publishers expect submissions — Times New Roman, 12pt, double-spaced, 1" margins, with a running header showing your name, the title, and the page number.

Tip: In Find & Replace, press Enter to jump to the next match or Shift+Enter to go back. Press Escape to close it.

🎨 Image Studio

The gallery is shared across the household, like the calendar and shopping lists — everyone sees everything generated here. A card shows whose image it is when it isn't yours.

Generate — Describe what you want (a book cover, a character portrait, a scene, or just something silly) and tap Generate. Takes up to about 30 seconds.
Style buttons — Tap Cartoon, Claymation, Pixel Art, Renaissance, or Coloring Page to add that style to your description instead of typing it out.
🎲 Surprise Me — Fills the prompt box with a random silly idea if you're not sure what to make.
🖌 Edit a Photo — Upload a real photo and describe how to change it, like "turn this into a cartoon" or "add a birthday hat." The style buttons work here too, and it uses the same generation budget as regular image generation.
✏️ Edit (on any gallery image) — Loads that image straight into the Edit-a-Photo form so you can make further changes without downloading and re-uploading it — handy for chaining tweaks like "now add sunglasses" after a first pass didn't quite land. Nothing is overwritten: this always creates a brand-new image, so the one you started from is untouched and still there if you want to go back to it.
Gallery — Tap any thumbnail to see it full-size. Use the dropdown on any image to attach it to a Writing Center book, where it'll also show up on that book's 🎨 Art tab. Attach it to a different book (or "— Not attached —") anytime — nothing is ever deleted by attaching or detaching.
😂 Meme — Add top/bottom caption text to any generated image and download it as its own picture. This happens entirely on your device — no AI or cost involved.

Quality	Defaults to Low — the fastest and cheapest option. Bump it up to Medium or High for extra detail.
Photo uploads	JPG, PNG, or WEBP. Larger photos are automatically shrunk on your device before uploading, so a typical phone photo works fine — anything still over 20MB after that will be rejected.
Card details	Each card shows the date, shape, quality, and file size in MB — handy for keeping an eye on disk space.
⬇ Download	Saves the image to your device
🗑 Delete	Removes the image permanently — this can't be undone. If a book gets deleted, its attached art is kept and just becomes unattached again, not deleted.

This needs a separate AI image account connected by an admin — if you see a message saying it isn't set up yet, that's why.

📁 My Files

Your own private file storage — unlike everything else in this app, files here are NOT shared with the rest of the household. Only you can see, download, or delete what you've uploaded, even from your own account's login.

⬆ Upload a File — accepts common documents (PDF, Word, Excel, PowerPoint, text, CSV) and images (JPG, PNG, GIF, WEBP, HEIC). Executable and script files (.exe, .bat, and similar) are always rejected, and every upload is checked that its actual content really matches what it claims to be — not just its file extension — before it's accepted.
Storage bar — shows how much of your 500MB personal allowance is used. Each individual file can be up to 100MB.
Thumbnails — image files show a real preview; other file types show an icon for their type.
✎ Description — add a short note about a file, either when you upload it or any time afterward.
⬇ Download — always saves under the file's real name, even though it's stored on the server under a different, randomly generated name for safety.

Ask Nelle — in chat, ask "what files do I have," have her write something up and save it as a text file for you ("save this packing list as a file"), add or change a file's description, or delete one. This only ever touches your own private files — never anyone else's — and if she's not sure which file you mean to delete, she'll ask rather than guess.

₿ Bitcoin Trader

Overview

What it does A fully-automatic paper trading engine for Kalshi's BTC 15-Minute contracts. Every 15 minutes Kalshi opens a fresh YES/NO contract ("will BTC be above $X when this window closes?"), and the server compares its own probability estimate (how far BTC sits from that strike, and its recent volatility) against what the market is charging. If there's a meaningful gap, it automatically places a small paper trade — most windows won't have a clear enough edge, and skipping those is expected, not a bug.
Version picker every active version (currently v6, v7, v8 and v9) runs at the same time, side by side, each tracking its own account and trade history independently — switching one never pauses or resets the others. v1 through v5 were retired on 2026-10-30: they no longer trade and aren't in the picker, but all their past trades are kept and still show on the Dashboard's bot card and in the side-by-side comparison report. The dropdown next to "PAPER TRADING" picks which one this whole panel — balance, chart, trade history, Reset, and the report downloads — is currently showing.
- v2 adds a loss-streak circuit breaker on top of v1: halves its bet size after 3 losses in a row, pauses entirely after 6, and resumes full size the moment it wins again.
- v3 adds a market-favorite filter on top of that — it won't bet against whichever side the market itself currently favors unless its edge is large enough to justify going against the crowd.
- v4 is v3 plus one more change: instead of getting BTC's price from a single exchange (Kraken), it checks Kraken, Coinbase, and Binance together and reconciles them — closer to the blended, multi-exchange price Kalshi itself actually settles against, which matters most on close calls right near the strike.
- v5 (added 2026-10-07) is v4 plus confidence-tiered sizing — instead of every qualifying trade getting the same flat bet, it bets half-size on a marginal edge (under 15%), normal size 15-25%, and double size at 25%+, backtested at +15.3% over the same trades v3 already took.
- v6 (added 2026-10-09) is v5 plus two active-exit rules, both checked every tick while a position is open and more than a minute remains: a "ride the curve" profit exit — if the live resale price climbs at least 15c above what was paid, it sells and locks in that smaller, guaranteed profit instead of risking a last-second reversal (a green "⚡ LOCKED" flag marks these in the trade history) — and a stop-loss — if the live resale price falls far enough below the best price actually seen while the position's been open, it sells and takes that smaller, bounded loss instead of riding it all the way down (a red "🛑 STOPPED" flag marks these). It also won't open a new trade until at least 5 minutes into a window, or below a minimum entry price (shown as v6 MIN ENTRY in Bot Activity).
  
  Full v6 change history
  
  Added 2026-10-19: the stop-loss rule itself. Added 2026-10-20: neither exit fires on the first tick a threshold is crossed — it has to still be crossed on the very next tick (~10s later) too, confirming it wasn't just a brief price wick, before the real sell actually goes out (shown as a "👀 LOCK ARMED"/"👀 STOP ARMED" line in Bot Activity while waiting on that confirmation). Both place a real sell order on Kalshi when v6 is the live version, not just in paper — if that real sell doesn't fill, the position simply keeps riding to real settlement like normal.
  
  Added 2026-10-21: a real exit order's price buffer now widens a little on every consecutive tick it fails to fill (capped at 20c), instead of retrying at the same fixed 4c offer forever — found live, during a window where BTC's price swung roughly 40-95c within about a minute, when two stop-loss exit attempts in a row both missed because the price kept moving before the order could cross; that particular window still resolved in profit, but the stop-loss itself hadn't actually protected it. Also fixed the same day: an already-armed lock/stop-loss no longer gets abandoned mid-retry just because the window enters its final minute — that 60-second cutoff was only ever meant to stop a brand-new exit from arming that late, but it was also silently cancelling one already in progress, which let a real live stop-loss attempt get dropped and ride uncontrolled to a full loss.
  
  Added 2026-10-22: the lock/stop-loss price check now always reads a fresh, uncached quote instead of the normal 15-second-cached one — v6 ticks every 5s, so the old cached read could silently feed the same stale price to up to 3 ticks in a row, including the two-scan confirmation's "second" look, which is supposed to confirm against genuinely new data rather than the same snapshot twice. Also added that day: stop-loss no longer waits for the two-scan confirmation (the profit lock still does) and its starting price buffer is wider (12c vs the lock's 4c) — measured live, stop-losses were still averaging a 26.3c real loss distance against the 20c target even after the freshness fix, with 79% overshooting; a stop firing a few cents worse immediately beats one that fires correctly-priced but late, so it now acts the instant the threshold is crossed instead of waiting a tick to confirm. Also added that day: a live entry now re-checks its edge against the fresh price right before firing, not just the cost — found from a real case where the edge (11%, off a 52c quote) was likely already gone by the time the fresh price (70c) was read, yet the trade still fired anyway at a smaller size; it now skips entirely, same as a missed fill, if the re-checked edge has dropped below the threshold that justified the trade in the first place.
  
  Added 2026-10-23: the 5-minute early-window filter — a 49-trade analysis of live trades that rode to full settlement found win rate climbing cleanly the later a trade was placed (40.7% in the first 5 minutes vs. 66.7% in minutes 10-15, same real trades, same model).
  
  Added 2026-10-26: the stop-loss now trails the peak resale price actually seen while a position's been open, instead of only ever measuring from entry. A trade that ran up nicely and then gave it all back used to have to fall the full stop distance below ENTRY before stopping out — riding the whole reversal first. Now it stops that same distance below whatever the best price it reached was, catching a "ran up then gave it back" reversal earlier. A trade that never moves into profit behaves exactly as before.
  
  Added 2026-10-27, ahead of a planned real-money launch: three changes, all tested in paper first. (1) Won't buy a contract below 20c, later the same day narrowed down to below 15c after a closer look found the original cutoff was too broad (15-20c actually performs fine; the real soft spot is specifically 10-15c) — both our own trade data and a completely independent trader's public bot audit, on the same market, separately flagged this general price range as underperforming; a tiny move against a contract that cheap is a huge percentage swing. (2) Paper bets briefly sized as a percentage of the current paper balance (2%, with a floor matching the old flat $10) instead of a flat dollar amount — reverted back to flat sizing the same day after a real losing stretch (14 trades, 53.8% win rate, -$254.14) once bets jumped to $139-284 each; ordinary variance in a small sample looked much worse once bet sizes were 15-20x bigger than before. (3) A real order now checks how much is actually resting on Kalshi's order book near the intended price before firing, trimming the contract count if the book is thin — a likely contributor to the "PRICE MOVED" pattern seen on live days, where the real cost-cap trim already covers the quote moving but never checked whether enough real size was there to fill against in the first place.
  
  Added 2026-10-28: corrected the 2026-10-26 trailing-stop change after it turned out to be the main driver behind a real win-rate decline (85.2% on 10-20 down to 64-69% by 10-27/28). Since the trailing stop went in, 69% of all stopped-out trades had briefly ticked above entry — often by just a few cents, nowhere near the 15c profit-lock threshold — before reversing and getting cut; the trailing reference was kicking in on any uptick at all, turning ordinary noise into a guaranteed loss. The peak now only counts as a trailing reference once it's cleared entry by at least 8c (half the profit-lock threshold) — below that, the stop falls back to the plain fixed-distance-from-entry behavior it had before 10-26.
  
  Added 2026-10-28 (later the same day): a tighter stop-loss (10c instead of the normal 20c) in just the final 3 minutes of a window, after a real example that same day — a position ran up to an 84c peak with 149s left, then collapsed to 20c in under a minute, blowing straight through the normal 20c stop before the next ~5s check could catch it. This close to resolution, the contract price swings bigger and faster on small BTC moves than it does earlier in the window. Only the final 3 minutes are affected — the normal 20c distance still applies everywhere else.
  
  Added 2026-10-29: once a trade is open, its exit is now checked every second instead of every 5 seconds, using a live price feed straight from Kalshi that updates the instant the price moves. Before this, a fast price drop could blow right past the stop-loss between checks (one stop meant to sell at 64c actually sold at 20c). Stops now land much closer to where they're supposed to.
  
  Added 2026-10-30: the profit lock now sells the moment the price reaches the lock line, instead of "arming" first and waiting for the next check to confirm (so the "👀 LOCK ARMED" line and sound no longer appear). The wait made sense when checks were 10 seconds apart, but at once-a-second checks it mostly just gave a few trades time to slip back and get stopped out instead — a net cost of about $42 over the 30 hours measured. Same day: fixed a timing bug where a check running a second after a window closed could record a fake stop at 0c on a trade that actually won.
  
  Added 2026-10-31: won't buy any contract priced at 70c or more (shown as "MAX ENTRY" in the settings tiles). Over two days, buys under 70c earned about 26c for every $1 bet, while buys at 70c and up earned about 2c — 40% of all trades for almost nothing. A contract that expensive can barely climb the 15c needed to lock a profit, so it usually just rides to the end for a few cents, while a loss still costs the full amount. Also applies to v7. Expect fewer trades, and skip lines in Bot Activity that say "Entry price (…) is 70c or more".
  
  Added 2026-11-02: the bot now decides using up-to-the-second prices. Before this, it used a market price that could be up to 15 seconds old, and paper trades filled at that old price, so in a fast market paper could "buy" cheap and "sell" a few seconds later at the real price for a profit that never existed. The first real-money test caught it: real money refused a 54c buy because the real price was already 66c, while paper logged a win. Paper results from before this date look much better than they really were (mostly from trades that closed within about 15 seconds). Also applies to v7. Expect fewer trades and fewer instant wins from here on. Those were the fake ones.
- v7 (added 2026-10-30) is an exact copy of v6 with one change: a smarter stop-loss. Before a stop sells, v7 checks the live Bitcoin price against the target line. If the odds still clearly say our side wins (by a 10-point safety margin over what selling right now would bring), it holds instead of selling, and keeps checking every second — so if Bitcoin really does turn against the trade, the stop still sells right away. A "🛡️ STOP HELD" line in Bot Activity shows when this happens. Built after finding that more than half of v6's stopped-out trades were actually on the side that went on to win. Runs on paper next to v6 so the two can be compared directly.
- v8 (added 2026-11-02) is a different bot, not a v7 tweak. It trades the ETH, SOL, XRP and DOGE 15-minute markets instead of Bitcoin — all four at once, at most one trade per coin per round. Why: an honest replay on Kalshi's own recorded prices showed the Bitcoin market is so crowded with professional bots that its price is already as accurate as our formula, while these quieter coin markets (10–35× less trading) sometimes fall behind for a moment. How it works: it watches each coin's live price from Coinbase and re-checks 4 times a second. When its fair-price formula says a side is worth 12+ points more than Kalshi's real price (priced 10¢–90¢, from 5 minutes into the round until 45 seconds left), it buys and holds to the end — no profit lock or stop-loss, so no second fee. On 5 days of real prices it had never seen, that made about +8¢ per $1 bet over ~1,300 trades. But it wins only about 4 trades in 10 (it mostly buys an underpriced underdog), so losing streaks are normal, and 1 of those 6 days lost money. The opening only lasts a second or two, which is why it reacts in a fraction of a second — the "reacted in …ms" part of each buy line in Bot Activity shows how fast. Every buy re-checks the real price first, paper included, so paper can never buy at a price that wasn't really there. Real-money size is 1 contract per trade (KALSHI_V8_LIVE_CONTRACTS). v8 has its own page: pick v8 in the version dropdown and the BTC cards switch to v8's dashboard (pick v6 or v7 to switch back; it remembers your last pick). It shows one card per coin with a live price graph that updates every second (shaded start = the first 5 minutes when it never buys, dashed line = the round's starting price, hollow ring = where it bought, solid dot = now), whether it's holding a trade or how close it is to the 12% buy line, a "Results over time" graph next to the pace the test predicted, a round-by-round grid of wins and losses, how fast each buy landed, your real Kalshi account, and a Paper / Real-money switch so the two are never mixed together. A health strip at the top shows whether v8 and both live price feeds are working, and loud banners appear if it's paused, the real-money loss limit is hit, or real money is stuck. Under the number tiles, Last round shows how each coin did in the round that just finished (results settle about 20 seconds after a round ends, so the P&L tile also says "just now: … from the 4:45 round" for a couple of minutes), and How it's doing says in plain words whether the results so far are within normal luck of the test — at low trade counts it will say "too early to tell" on purpose. Each held trade says whether it's On track or Behind right now, a coin that isn't trading says why, and "EDGE GONE" lines read "Skipped — the price moved before it could buy. No money spent." Hover over any number tile for what it means.
- v9 (added 2026-11-04) is v8 with three fixes, built after v8's first review found a run of 14 losses and 86% of its bets on NO. It started with v8's 12-point buying rule and runs next to v8 so you can compare them. Bigger bargains only (added 2026-11-07): it now waits for the formula to rate its side at least 15 points above the price, because the formula turned out to be about 10–15 points too sure of itself, and a bigger bargain leaves room for that. It trades less often, but each trade is a better bet. A coin card that says "Small bargain … it waits for 15+" is passing on purpose. NEAR added (2026-11-08): with the bigger-bargain rule it trades less often, so v9 now watches a sixth coin, NEAR, which Kalshi settles the same way as the others. On days the rule was never tuned on, NEAR tested about as well as the other five coins, so it should add roughly a fifth more trades without lowering the bar. Stronger bets only (added 2026-11-06): v9's long losing runs came from long shots, so it now only buys when the formula gives its side at least a 50% chance and the coin has been moving that way over the last 60 seconds. It trades ETH, SOL, XRP, ZEC and HYPE (plus NEAR since 11-08); DOGE is out because Coinbase prices it too rarely. It makes fewer trades, and each coin card says why it's passing on a bargain. 1. Price correction: Coinbase usually sits slightly below the index Kalshi uses to settle, which made v8 see bargains on NO that weren't there. At the start of each round v9 measures that gap against Kalshi's real starting price and corrects for it. Each coin card shows the correction, and its "Now" price and chart are the corrected ones. 2. It can sell: once a trade is up 30¢ it rides the gain and sells if it stays 15¢ or more below its high for a full second, so a price that only flashes down doesn't trigger a sale. (Until 11-08 it also kept holding past that line while the formula rated the trade higher. That let trades ride from a big gain back to a loss, like XRP going 41¢ up to 91¢ and finishing at zero, and on balance it didn't pay, so it was dropped.) It also sells if the market is paying 15+ points more than the formula says the trade is worth. After selling it waits 30 seconds and may buy again if a new bargain shows up, so a coin can trade more than once a round. Each sale costs a second Kalshi fee. 3. Instant checks: it re-checks the moment a price changes instead of 4 times a second, and just before every buy or sale it reads the real order book, because Kalshi's market snapshot was often several cents behind real trades. How well it tests: an earlier replay said about +31¢ to +44¢ per $1, but that replay let a buy happen at prices that were already gone. A stricter replay only counts a buy if a real trade happened near that price within 3 seconds, and it matches the real trades. On that test, across about 4,800 coin rounds at $10 a trade, the current rules (15-point bargains, selling on the ride line) made about +16¢ per $1 (+12¢ with every order 1 second late), against about +5¢ (+7¢) for the rules before 11-07. The longest losing streak was 5 (7 when late). On the rounds v9 really traded on 11-06 and 11-07, the 15-point rule would have kept 22 of the 67 real trades and made about the same money with them. These rules were the best of about 50 tried, so expect less live; the page measures against +12¢. Losing streaks and bad days still happen, and only real trades can show whether it makes money. Paper flatters its sales, especially "sell when overpriced," so judge it by real prices. During Kalshi's maintenance a round's results can take a couple of hours; a yellow banner says so and the trades settle on their own. The "Against v8" line on its page shows how the two compare since v9 started. A green or orange SOLD line in Bot Activity is a sale, and the Result column shows the price it sold at. This page plays the same sounds as the Bitcoin page: one for a buy, one for a win and one for a loss, and a sale counts as a win or loss by whether it made money. Paper checks that it works; the real-money test decides whether it makes money. Real-money size is about $1 a trade.

Controls

⏸ Pause Stops just the currently-selected version from opening any new trades, without touching the others. A trade already open when you pause still resolves normally — there's no way to force-close a live Kalshi position early, so letting it settle naturally is both the safe option and the only one. A banner appears on the panel whenever that version is paused, and it stays paused even through a server restart. Switch versions with the dropdown to pause or resume a different one; tap "▶ Resume" to pick back up where it left off.
⚠ Voided trades Very rarely, Kalshi itself fails to post a settlement result for a window in any reasonable time (this actually happened once, 2026-08-28, stuck for hours). Rather than freeze a version indefinitely waiting, it gives up after 15 minutes and voids that one trade — refunded as a wash, never guessed as a win or a loss. A yellow "VOIDED" badge appears next to AUTO when this has happened, and the voided trade shows up gray (not green or red) in the trade history, with the reason in a tooltip. It doesn't count toward wins or losses, but the small fee it cost does still count against the total P&L, since that part was real.
⚠ Data stale warning Added 2026-10-26. Invisible almost all the time by design -- only appears if the page's own account/trades data genuinely stops refreshing for 30+ seconds (a real network hiccup, a browser issue), so a silent failure never goes unnoticed the way a timing gap once did. The "Updated" clock next to the BTC price up top only reflects a separate, unrelated ticker poll, so this is the one honest signal that the actual trading data underneath is current.
🔴 Live trading Starting 2026-09-30, one version at a time can be switched from paper to real money (currently v3). When a version is actually live, its title turns into a pulsing red "🔴 LIVE TRADING" banner and its Mode field reads "🔴 LIVE — REAL MONEY" — impossible to mistake for a paper version by accident. Every other version keeps paper-trading completely untouched regardless. A real trade sizes itself from a separate, smaller setting than paper trades, and there's a hard dollar loss cap that stops all further real orders automatically (and does not resume on its own) if it's ever reached. Reset is refused outright for any version with real trade history on it — that can't be undone from a button click.
Daily paper loss limit Added 2026-10-26. Every version's PAPER trading pauses itself for new trades once that version's net paper losses for the day reach a set dollar amount (shown as DAILY LIMIT in Bot Activity) — separate from the loss-streak circuit breaker above, which only reacts to a losing streak and could miss a day that bleeds from a mix of wins and more losses without ever hitting that streak count. Unlike the real-money loss cap, this resets itself automatically at the start of the next calendar day — no manual reset needed. Live trading isn't affected by this; it already has its own tighter cap, which counts real trades from the live start date (moving that date forward resets it).
🔴 Make [version] live / ⬇ Step down to paper Added 2026-10-06. Lets you switch which single version is allowed to place real-money trades without needing SSH access or a server restart — pick a version in the dropdown and the button changes to whichever action applies to it. Switching a new version TO live is refused outright if the current live version still has an open real-money position, so two versions can never have real money at risk at the same time; stepping the current live version back down to paper has no such restriction since it only stops something, it never starts anything new. Both ask for confirmation first, and Resume does too now if you're resuming the version that's currently live, since that's what actually lets real trades start happening again.
Real Kalshi Account card Added 2026-10-05. Shows the actual available balance, portfolio value (the live value of any open real position, separate from cash), and number of open positions on the real Kalshi account, pulled live from Kalshi itself — not a simulation. When a real position is actually open, a detail line appears underneath showing exactly what's held (side, ticker, size, cost), pulled straight from Kalshi's own position data. There's only one real account, so this card stays the same no matter which version the dropdown is showing; it's styled in amber specifically so it never gets confused with the green/paper stats next to it. Sits side by side with the Window/Status card since both are compact. Added 2026-10-20: a LIVE REALIZED P&L / W-L / WIN RATE row appears underneath once the currently-selected version has placed at least one real trade — these are this version's own real-money results specifically, kept completely separate from the Paper Account card below so switching a version between paper and live never blends the two tallies together.

Chart & Trade History

Chart Plots BTC's price during the current window against the strike (the dashed orange line), with a glowing gradient radiating from the strike line — green above it, red below — so which side is "winning" is obvious before even reading the line itself. The price line is smoothed and glows, the current price has a soft pulsing halo, and hovering anywhere on the chart shows a crosshair and a tooltip with the exact price and gap from strike at that point. A bright yellow ring marks every point where price actually crosses the strike. Small triangles mark when a trade opened (blue) and resolved (green/red/gray, matching win/loss/void), with a thin yellow ring around any early close — so you can see at a glance when in the window a trade fired. The bar along the bottom shows time left — orange, amber under 3 minutes, red in the final minute. The card above it shows strike, live BTC price, time left, our estimate vs. the market's, the current recommendation, and a 🔍 Edge Progress bar showing how close the current edge is to the threshold needed to trade.

Chart change history

Added 2026-10-12: trade-open/resolve triangle markers. Added 2026-10-20: the 🔍 Edge Progress bar. Added 2026-10-24: the full glow/gradient redesign — radiating strike-line gradient (replaced a flat tint), smoothed glowing line, pulsing live-price halo, and the hover crosshair/tooltip.
📡 Live Monitors card Sits next to the Window/Status card — live bars, all purely informational (none of them change a trading decision): Loss streak (vs. the half-size-at-3 / pause-at-6 thresholds that actually trigger), Live loss cap (how close real trading's P&L since the live start date is to the cap that halts further real orders), BRTI drift (how far the exchange-median approximation sits from Kalshi's real settlement index), Loop heartbeat (seconds since this version's trading loop last actually ran a tick), Market spread (how wide the current YES+NO ask gap is — wider means a thinner, less trustworthy market), Volatility (BTC's realized 1-minute volatility), Fee drag (what share of today's gross profit fees are eating), Participation (what fraction of the last 24 hours' windows got traded vs. skipped), No market found (consecutive ticks this version couldn't find any open window at all), Live entry trouble (consecutive live-entry attempts without a real fill), and Price source health (which of BRTI/Kraken/Coinbase/Gemini has failed the most times in a row, if any). Each bar runs green/amber/red depending on how close it is to actually mattering — hover any bar for the full explanation.

Live Monitors change history

Added 2026-10-20: the original five bars (loss streak, live loss cap, BRTI drift, heartbeat, participation). Added 2026-10-24, after Kalshi's own status=open filter started matching nothing for the BTC15M series specifically (a narrow Kalshi-side bug — confirmed the identical filter still worked for other series and globally, not a documented status-value change) and the loop went 2+ hours silently finding nothing to trade with zero errors raised: the No market found, Live entry trouble, and Price source health bars, built specifically to catch that same shape of problem going forward. Added 2026-10-25: every label was trimmed down to just the headline number, with the full sentence moved to a hover tooltip, after the card started feeling too text-heavy with all 11 bars in place.
📜 Bot Activity The row of small labeled numbers at the top of this card (live/paper bet size, loss cap, daily paper loss limit, price buffer, v6's lock and stop-loss thresholds, minimum edge, fade-favorite edge, minimum seconds before close, loss-streak half-size/pause points, stuck-settlement timeout, and a live "KALSHI PING" showing the round-trip time of the most recent real request to Kalshi) shows exactly what every version is actually running with right now — read-only here, since changing any of them still means editing the code and restarting the server. Below that, a live, color-tinted scrolling feed of the exact same narration that's always been in the server's own logs — every skipped window and why, every real order attempt (filled, missed, price moved since deciding to trade), every lock-in or stop-loss, every error — newest at the top, only keeping the most recent ~150 lines per version. It's a live glance at what the bot is doing right now, not a permanent record (the trade history and reports below are still the real archive).
Trade history One row per trade: date (with a "P"/"L" flag for paper vs. live), side, strike, entry price, contracts, BET (dollars staked), result, FEE (Kalshi's real taker fee, already subtracted from P&L), GOT BACK (proceeds from a self-closed sale — "--" for a trade that rode to full settlement), PING (round-trip time to Kalshi when that trade was decided), and the final P&L. A "Group by 15-min window" checkbox (on by default) collapses several trades from the same window — common since the fast tick can open/lock/reopen a few times while price holds still — into one row with the combined count, totals, and win-loss record; uncheck it to see every individual trade.

Trade history change history

Added 2026-10-20: the P/L flag, the PING column, and the "Group by 15-min window" checkbox. Added 2026-10-23: the GOT BACK and BET columns. A 📷 chart-snapshot icon was added 2026-10-25 and paused again 2026-10-26 -- being entirely browser-driven, it only caught a trade if this tab happened to be open at the exact moment it resolved, which missed the vast majority of real trades in practice. Off for now pending a more reliable approach.

Reports

⬇ Report / ⬇ All-Time Downloads a visual summary of trading — color-coded win/loss record, profit and fees, and a trade-by-trade table — for whichever strategy version the version picker above is currently showing. The dropdown next to "⬇ Report" picks how far back it covers — anywhere from the last hour to the last 7 days. All-Time always covers every trade that version has ever made, regardless of that dropdown. Both open straight into any browser; scroll to the bottom for a plain-text copy of the same data, handy for printing or pasting elsewhere. The server also saves a daily 24-hour report to disk automatically for each version, so there's a running history even on days nobody downloads one by hand.
⬇ Compare Downloads every known strategy version's key numbers side by side for the time window selected above (win rate, P&L, profit factor, Sharpe ratio, max drawdown, and more), with the better number in each row marked green and a plain-English verdict at the top on which version is actually ahead on profit. Rows like trade count, fees, and participation aren't marked "better/worse" since more of those isn't inherently good or bad — they're shown for context only. If either version is still light on trades, the verdict says so rather than calling a winner too early.
Benchmarks Extra numbers beyond the basics: how much it wins per $1 it loses, win rate compared against the breakeven rate actually needed to profit (based on the average price paid — cheaper entries need a lower win rate to come out ahead, pricier ones need a higher one), how much of its winnings go to fees, a YES-vs-NO breakdown, average trade size, average time from placing a trade to it resolving, how many of the possible 15-minute windows actually got traded and how many were skipped (with a breakdown of why), the best and worst single trading day, max drawdown in dollars and as a percentage of peak, win rate vs. a pure coin flip, the current win/loss streak, win rate by entry price and bet size, spread stats, and win rate split by whether a trade agreed with or faded the market's own favorite. With at least 10 resolved trades, the coin-flip comparison shows whether the win rate is far enough from 50% to likely be more than luck.
🗒️ Nelle's Notes A short plain-English health check summarizing what the numbers show — not a prediction, just an honest read of how things are going right now.
💡 Ideas Worth Considering Patterns the data actually shows — a side or hour that's over/underperforming, a multi-hour losing stretch, fees eating a notable share of winnings, trading almost every window instead of being selective, and more. Only appears when something in the numbers actually stands out.
🔧 V2 Candidate Ideas A running log of things worth changing in a future version, added once a pattern shows up across more than one report — not every one-off blip. Nothing here changes anything automatically.
Daily / Hourly / Day-of-Week Trend Win rate and profit/loss broken down by calendar day, hour of day, and day of the week, so you can see whether certain times tend to go better or worse. The daily and day-of-week views only appear once there's more than one day of history.
Strategy timeline Shows when the selected strategy launched, how many days it's been running, its trading mode (paper or live), and its retire date — "TBD" until it's actually taken offline. The same info appears at the top of every downloaded report.
Sharpe Ratio (per trade) Average profit per trade divided by how much that profit bounces around trade to trade. A steady string of modest wins scores higher than the same total profit earned in a few huge swings. Not the traditional annualized version — just a consistency measure for this strategy's own results.

How the P&L number is calculated:

The P&L shown for a trade is your profit — what you gained on top of your stake — not the total money that came back. Every contract costs whatever price it's bought at (say 48¢), and if it wins, that same contract pays out a flat $1.00 no matter what you paid for it. So a cheaper entry price means more contracts for the same money, and a bigger payout multiple if it hits.

Example	20 contracts bought at 48¢ each, and it WON
Stake (what you paid)	20 × $0.48 = $9.60
Payout (since it won)	20 × $1.00 = $20.00
Gross profit	$20.00 − $9.60 = $10.40
Minus Kalshi's fee	−$0.35
P&L shown	+$10.05 — your actual gain

If a trade LOSES instead, there's no payout — you simply lose the full stake, plus the fee (which is charged either way, win or lose).

Still practice money throughout — nothing here places a real trade.

⚙️ Settings

Manage accounts, give your AI assistant a name, set its personality, and pick a voice for spoken responses. You can also add other household members here.

💾 Download Latest Backup — the server automatically backs up the database and all photos/images once a day; tap this anytime to save a copy of that latest backup to your own computer, for extra safety on top of what's already kept on the server.

💡 Tips

The 🎤 mic button works everywhere — try saying things like "Go to calendar" or "Add milk to the shopping list." Tap it once, talk, then just stop — it notices when you're done and answers automatically. You can also tap it again to stop it early.
Tap the small ‹ tab on the edge of the sidebar to collapse it and use the full screen width. Tap it again to bring the sidebar back — your choice is remembered next time you visit. (Not available on the mobile bottom-bar layout.)
Your data stays on your own server — nothing is shared with outside services.
If something stops working or looks off, a quick page refresh usually fixes it.

Good Morning! 👋

💬 Chat

📅 Calendar

Add Event

✅ Tasks

Add Task

🛒 Shopping

📝 Notes

🎰 Vegas Trip

📈 Stocks

Add a Stock

📰 News

🎯 Prediction Markets

🚗 DoorDash

Real Kalshi account

This round, coin by coin

Results over time

Round history

Bot activity

Reaction speed

Trade history

⚙️ Settings

💰 Budget

⚠️ Pending From Before This Month

Accounts

Checking & Savings

Investments

Spending by Category

⏳ Pending Transactions

Transactions

Add Transaction

Add Account

Add Category

📖 Writing Center

🎨 Image Studio

📁 My Files

Hi, I'm Nelle

Good Morning! 👋

💬 Chat

📅 Calendar

Add Event

✅ Tasks

Add Task

🛒 Shopping

📝 Notes

🎰 Vegas Trip

📈 Stocks

Add a Stock

Price History

📰 News

🎯 Prediction Markets

🚗 DoorDash

Real Kalshi account

This round, coin by coin

Results over time

Round history

Bot activity

Reaction speed

Trade history

⚙️ Settings

💰 Budget

⚠️ Pending From Before This Month

Accounts

Checking & Savings

Investments

Spending by Category

⏳ Pending Transactions

Transactions

Add Transaction

Add Account

Add Category

📖 Writing Center

New Book

🎨 Image Studio

📁 My Files

🎨 Pick from Image Studio

😂 Add Meme Text

❓ Help & Guide

🏠 Dashboard

💬 Chat (AI Assistant)

📅 Calendar

✅ Tasks

🛒 Shopping

📝 Notes

🎰 Vegas Trip

📈 Stocks

📰 News

🚗 DoorDash

💰 Budget

📖 Writing Center

🎨 Image Studio

📁 My Files

₿ Bitcoin Trader

⚙️ Settings

💡 Tips