Builds

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, Nathan, Rose Loops, Shawn Porter, Tia

A brutalist guide + game on the real agentic coding workflow. It opens with a 6-step interactive walkthrough (Plan -> Prompt -> Diff -> Test -> Commit -> Ship), each paired with the specific mistake that sinks a vibe-coded build. Then a 20-round timed Gauntlet: you get a real scenario (a plan, a diff, a claimed test result) and 14 seconds to call SHIP or REVERT, with an instant lesson either way. The centerpiece is Live Lab: you write an actual prompt telling an AI agent what function to build, it's sent server-side to a real LLM via OpenRouter constrained to return exactly one named function, the response is scanned for scope violations (fetch, globals, eval, infinite loops) before anything runs, then it's tested against real test cases inside a sandboxed iframe with a freeCodeCamp-style pass/fail checklist. Vague prompts get vague code, on purpose, no scripting involved.

Models

Claude Sonnet 5 (built the app); openai/gpt-4o-mini via OpenRouter (the live agent players prompt inside Live Lab)

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, Nathan, Rose Loops, Shawn Porter

Not sure one shot prompt.

Models

Grok build

Cost

10 minutes

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, Nathan, Rose Loops, Shawn Porter

It uses my proprietary custom custom LLM to talk you through different goblin meme agents

Models

TRiAD Coder and GoblinGLM1.2

Cost

$0

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, Nathan, Rose Loops, Shawn Porter

shuhari maieutik lehrgangsweg praxisnah durchspielung handreichung kata dojo spielwiese sandkasten anklickbar mitmachpfad pruefschleife kleinschrittig bisektion regressionsnetz werkzeugaufruf zwischenstand kaizen genchibutsu fensterdrift kontextfaeule kompaktierung planmodus phantomapi riesendiff blindannahme testlos reviewlos commitlos stillbruch gebetsdeploy schuldenlawine geheimnisleck entzauberung schonungslos stolperstein fallgrube podvodnyekamni praemonitus festinalente nuechternheit Build something that teaches someone how to code with AI agents — a guide, tutorial, or interactive walkthrough that shows the real workflow and calls out the mistakes that sink a vibe-coded build. You have 15 minutes

Models

Claude fable 5.1

Cost

113k tokens

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, Nathan, Rose Loops

Press play and listen

Models

Notepad, Akai MPC and Suno

Cost

1 million

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, Nathan, Rose Loops

you type what you wanna build. it doesnt let you. it asks 5 things first. what exists when its done. what it has to do, one per line. whats the obvious version and what actually goes wrong for the person using it. what youre building it with and what its NOT allowed to need. and how youll prove it works when you open it cold. every question has a red box that tells you exactly how skipping it sinks your build. hit "show me a filled one" if you dont wanna type. at the end you get a brief to paste into claude / cursor / whatever, or hit the bot button and it writes you a one pager: name, must dos, build order, what will sink it, and the first message to send your agent. last screen is what to do AFTER the prompt. 4 agent replies that sink you next to the 4 that ship. no login. nothing leaves your browser.

Models

claude fable 5.1

Cost

~90k

Voted by ₥AN$A, Air Planes, kaleb, Nathan, Rose Loops

It works by turning an agentic-coding guide into an explorable environment: 1. You move r3plic4nt through the room using WASD. 2. Each terminal represents one lesson or stage in an agentic workflow. 3. When you approach a terminal, you press `E` or `Enter`. 4. The game cuts to a focused terminal view. 5. That terminal can present: - The coding concept - A practical example - A prompt or workflow - Common mistakes - Supporting video or media - A small interactive challenge 6. You return to the room and continue to the next terminal. The room acts as the guide’s table of contents, while terminals contain the actual lessons. Eventually, completing terminals could unlock later sections and turn the guide into a structured learning journey rather than a conventional webpage. Right now, the exploration and terminal system work, but the six agentic-coding lessons still need to be written and installed.

Models

GPT Sol & Astra

Voted by ₥AN$A, Air Planes, Auny 🧡, Hovhannes Mkhitaryan, Is Wilson, kaleb, rookepoole, Rose Loops, senseless, Smiley “MakeShif”, Tia

Third Bass is a collaborative planning and review workspace where anyone can create a plan, invite others to review it, and independently flag anything they believe needs attention. There are no turns or room codes, and everyone can create and review multiple plans at the same time. Red flags are shared with the group, with independently flagged concerns becoming seconded when others identify the same issue. Plans, concerns, flags, responses, and resolutions remain available as part of the shared session history, creating a persistent record of how the group evaluated and improved each plan. Third Bass is remixed by Airplanes. Remix of SECONDED by AunySillyMe

Models

Grok Bolt

Cost

300k

Voted by ₥AN$A, Air Planes, Auny 🧡, Debbie, Hovhannes Mkhitaryan, Is Wilson, kaleb, rookepoole, Rose Loops, senseless, Smiley “MakeShif”

Create a room, give it a purpose, and share the room code with the group. Everyone joins as a persistent guest with no account required and connects directly over WebRTC. Instead of simply voting “available,” each person gets a limited backing budget to spend across the times they genuinely want. They can also mark conflicts as busy or maybe. Every action syncs live across the room. BACKED continuously compares the leader’s support against the backing that participants still have left to spend. Once the leader can no longer mathematically be caught, the room automatically locks the decision. If someone later marks a backed time as busy, that backing is automatically withdrawn. The result is a scheduling decision based on commitment, not checkbox availability, with the winning time and group reasoning preserved as a shared record.

Models

GPT-5.6 Sol

Cost

Built with ChatGPT. Exact token usage and per-build API cost were not exposed/tracked in the ChatGPT interface.

Voted by ₥AN$A, Air Planes, Auny 🧡, durdy, kaleb, rookepoole, Rose Loops, Smiley “MakeShif”, Wil

Air Traffic Control
by airplanes A shared board for one meeting. Everyone uses the same link. No accounts. Paint when you can, clear a time, confirm you’ll show up. How to use it 1 File the flight.
Give the meeting a name and a purpose. Set how long it is, which days to search, the event timezone, and a place or video link if you have one. Open the room and copy the link. 2 Get on the board.
Open the link. Type your name and choose your timezone. You are painting in your local time; the room stores one shared schedule underneath. 3 Paint availability.
Drag the grid. ◦ Hard yes = you can be there ◦ Maybe = you can make it if that’s the only slot ◦ Clear = remove a block
Use working hours if you want weekdays filled in one tap. Two people have to paint before the board can rank times. 4 Read the departures.
Ranked cards are the best overlaps. Each card lists the time in every person’s zone and whether the slot is Fair, Uneven, or Harsh. Harsh means someone is outside normal hours. 5 Clear one time.
The organizer locks a card (or picks a start on the grid). That is the hold. Everyone should see the same locked time. 6 Confirm.
Tap I’m in, Maybe, or I can’t. Download the calendar file and add it. Copy the invite if you need to send it outside the board. If someone hasn’t painted or confirmed, send them a nudge from the room. 7 Show up.
Wheels up means enough people committed. From shortly before the start, mark who arrived. That’s the roll. What the board is saying • Holding — still collecting availability • Ranked — enough data to pick a time • Cleared — the time is locked • Confirming — waiting on pledges • Wheels up — the meeting is real • Done — it already started; record who was there House rules • One link is one meeting. Share /e/..., not the home page. • Times always include a timezone. If a time has no zone, ignore it. • I can’t does not cancel the hold. The organizer unlocks or reschedules. • If your grid doesn’t move when someone else paints, you’re on a different link. Open the exact room URL again.

Models

Grok Bolt

Cost

300k

Voted by ₥AN$A, Air Planes, kaleb, Khalil, rookepoole, Smiley “MakeShif”, Striker9, Wil

SAMEFRAME is a shared watch-and-listen room that keeps everyone on the exact same moment, no matter where they are. Remote watch parties usually fall apart because someone is ahead, someone is behind, and everyone ends up coordinating playback instead of enjoying it. SAMEFRAME synchronizes play, pause, seeking, and skips across the whole room, with support for YouTube and direct video/audio URLs. Live chat, emoji reactions, and timestamped Moment Pins give everyone ways to react to the same moments together. Create a room, send the invite link to a friend, load something to watch, and try controlling playback from either device — the fastest way to understand SAMEFRAME is to experience it together.

Models

GPT-5.6 Sol

Cost

Built during a ~30-minute hackathon session. Exact token usage/cost was not tracked.

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Smiley “MakeShif”, Tia, Wil

GiveCircuit coordinates an entire community drive in one shared system. Recipients self-register for assistance, choose a pickup location, and track their request. Volunteers create accounts, sign up for shifts, check in/out, and have their service hours tracked. Organizers get an admin command center showing recipients, volunteers, inventory, fulfillment status, activity, and impact metrics in real time. Admins can configure pickup sites and custom registration fields, export records, manage privacy/retention policies, and review an append-only audit trail. The app uses persistent shared database state, so every role is participating in the same live drive rather than separate single-user experiences.

Models

GPT-5.6 Sol

Cost

Built in ChatGPT Pro; exact token usage/cost was not surfaced for the session.

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Smiley “MakeShif”, Tia, Wil

Let’s all participants sir help

Models

Phi-3.5

Voted by ₥AN$A, Air Planes, kaleb, Khalil, rookepoole, Smiley “MakeShif”, Striker9, Wil

How to use Roto Vibes Open the site. Type a name. Create room to start a session, or Join room and enter the 6-character code. Share the link or the code with the group. Everyone in that room hears the same station. The host plays, pauses, and changes channels — it updates for the whole group, not just them. Chat live. Send emoji as it plays. Comments stamp to the moment you heard them.

Models

Grok Bolt

Cost

300k

Voted by ₥AN$A, Air Planes, Auny 🧡, Hovhannes Mkhitaryan, Is Wilson, kaleb, rookepoole, senseless, Smiley “MakeShif”

Fork from Mansa’s Maith. Answers math questions, categorizes them, and shows proofs, allows PDF download for notes.

Models

Grok

Cost

$5 in compute

Voted by ₥AN$A, Air Planes, Auny 🧡, durdy, kaleb, rookepoole, Smiley “MakeShif”, Wil

An organizer creates a meeting room and shares the link. Everyone joins as a guest and marks each proposed time as Available, Can Flex, or Unavailable. ARRIVE calculates the live intersection of everyone's availability and only allows the organizer to lock a time that the entire submitted group can attend. If there isn't a shared time yet, it identifies the smallest conflicts and who could flex to resolve them. Once a shared time is locked, everyone confirms attendance, can add the meeting to Google Calendar or download an .ics event with reminders, and receives a shared countdown. Organizers can nudge people who haven't responded or confirmed. If someone drops out after locking, ARRIVE detects that the shared time has broken and can reopen scheduling while preserving everyone's availability. At meeting time, participants check in so the loop ends with who actually arrived — not just who voted.

Models

OpenAI GPT-5.6 Sol

Cost

Built during a ~30-minute ChatGPT hackathon session. Exact token usage/cost wasn't exposed in the ChatGPT client.

Voted by ₥AN$A, Air Planes, kaleb, Khalil, rookepoole, Smiley “MakeShif”, Striker9, Wil

Screening Room is a watch party that actually feels like being in the same room. Send one link, everyone takes a seat, and the film starts on a shared countdown so every screen hits the same frame. Play, pause, or skip and it happens for the whole group. You react live, chat, and pin comments to the exact moment they happen. And the Usher, an AI host running on our own cheap GPU, runs the room: it welcomes people by name, calls them out when it goes quiet, runs hot-take polls, and can pause or jump the whole group on request. Anyone can host it for pennies, and it drops onto any video page with one bookmark.

Models

Claude fable 5.1

Cost

100k

Voted by ₥AN$A, Air Planes, Auny 🧡, Hovhannes Mkhitaryan, Is Wilson, kaleb, rookepoole, Smiley “MakeShif”, Tia

A group passes one message down a chain. Each person sees ONLY what the last person handed them, then the app shows exactly how it mutated and who did it. The fork fixes three things in the Ep.8 build: 1. THE MESSAGE NOW DISAPPEARS. Before, it stayed on screen while you typed — so it measured deliberate paraphrase, not memory. Now you tap to reveal, get a countdown scaled to length (4-15s), the composer is LOCKED while it's up, then the text is removed from the DOM. Enforced server-side, so reloading isn't a free second look. Paste is blocked and typing speed recorded — a hop that comes back near-verbatim typed faster than anyone composes gets flagged LIKELY COPIED. 2. ONE BAD LINK NO LONGER POISONS THE RUN. Every claim used to be scored against the original seed, so one person going off-script marked every ribbon "dropped" and made the rest meaningless. Breaks are now detected (MAD outlier, robust to the single wild hop that inflates sigma), named, and the analysis re-baselines: per-segment fidelity, a second decay line, and a second flow diagram from where the chain actually restarted. Host can undo a turn. 3. THE BLINDING WAS BYPASSABLE. Player IDs were broadcast and join trusted a client-supplied one — you could read the host's ID off the wire, rejoin as them, and be handed the seed. Seats now carry a secret token, never broadcast. Live run from tonight: 13% survived against the original, 34% re-baselined after the break, break correctly attributed. The abduction called it "vividness, 78% confidence." The salience-5 detail (a cat asleep in a pastry box) outlived every other claim — textbook Bartlett sharpening.

Models

Claude Opus 5 in Claude Code to build it. anthropic/claude-haiku-4.5 via OpenRouter at runtime for proposition decomposition, per-hop NLI entailment, abduction and narration.

Cost

Runtime: ~15 model calls per 7-player game, roughly 1-2 cents. Build session tokens I didn't track.

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Sharon, Smiley “MakeShif”, Tia

Open the site, type a name, and you’re in. No code. You’ll see who else is on frequency at the top. One person files a short idea and can add a seed sketch. Everyone else taps Join Line. When you’re on deck you only get what the last person handed you. Write one line for what you think it is, draw that, hit Done. Then it moves to the next person. You will not see anyone else’s pages while it’s moving. That’s the game. After a couple of handoffs the starter hits Reveal (or it opens itself when the line is empty). Then the whole group sees the original idea next to the final version, and every caption and sketch in between — so you can tell what got kept, dropped, or invented.

Models

Grok Bolt

Cost

300k

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Sharon, Smiley “MakeShif”, Tia

A rumor's chain of custody. Two games about what memory does to a sentence. Telephone: one person writes a rumor and it passes through everyone in the room, one at a time. You get one look at what reached you (12 seconds, shorter every hop, no pasting) and retell it from memory. The room only ever hands you the last version, enforced on the server. At the end: original beside final, then every hop word-diffed and tagged exaggerated / dropped / invented, pinned to whoever did it, plus a per-player ledger with a verdict. Before the record opens, a recall round scores each receiver on the first message that reached them. Memory ladder: whoever is playing sees one sentence for a few seconds and types it back. Accuracy is the score. Levels get longer and faster until nobody clears 40%. Three modes: solo, 1v1 (round wins decide), group. Four gossips with known habits (Mo exaggerates, Dee forgets, Fabi fabricates, Sal nearly gets it right) fill empty chairs, so it plays alone too.

Models

Claude Fable 5.1, Grok

Cost

90k tokens

Voted by ₥AN$A, Air Planes, Auny 🧡, durdy, kaleb, rookepoole, Smiley “MakeShif”, Wil

Pitch

Models

Transformers of San Francisco

Cost

$10

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Sharon, Smiley “MakeShif”, Tia

Every organization has a game of telephone running inside it. A decision, a complaint, a number — by the time it reaches the far end, nobody can say where it changed. Stemma shows you. Your group passes one message down a chain, each person seeing only what they were handed. Every version is captured, then laid bare: which claims survived, which got exaggerated, which were invented outright — and at exactly which person. And you can seat AI players with known biases into the chain, to watch a specific distortion do its work.

Models

Claude Opus 5

Cost

80k tokens

Voted by ₥AN$A, Air Planes, Auny 🧡, durdy, kaleb, rookepoole, Sharon, Smiley “MakeShif”, Wil

Create a room and paste the site you want to test. Guests join without accounts, choose a name/avatar, and appear in the live room. Each tester uses the “THRASH THIS” bookmarklet on the actual target website. Click PIN, select any real DOM element, and report a bug or comment. THRASH automatically captures the page URL, DOM selector, viewport, browser info, severity/category, and recent click trail for reproduction. Everyone in the room sees shared issues, live presence, page pins, activity, and can resolve or reopen findings together.

Models

GPT-5.6 Sol

Cost

Built in one ChatGPT session. Exact token count and per-session cost were not exposed to me.

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Sharon, Smiley “MakeShif”, Tia

A telephone game for one phone — someone starts a rumor, it gets said aloud and shown on-screen just long enough to read once, then vanishes. Each player writes down what they remember, passes the phone on, and by the end you get a side-by-side reveal of exactly how the story mutated.2. "Deploy it to railway so we can have a backend and it can persist if need me, also lets submit this"

Models

GroupGPT / Sonnet 5

Cost

$4.83

Voted by ₥AN$A, Air Planes, anakhuagbor sule, Auny 🧡, Hovhannes Mkhitaryan, kaleb, rookepoole, Smiley “MakeShif”, Wil

Players have to answers questions and get to know each other.

Models

GroupGPT/Claude Sonnet 5

Voted by ₥AN$A, Air Planes, durdy, Hovhannes Mkhitaryan, kaleb, rookepoole, Sharon, Wil

A crew points it at any URL and red-teams it together without all finding the same bug. Everyone is auto-assigned a different lane — forms, small screen, keyboard, edge input, the flow, slow+cold, links, words, look, what's missing. Coverage is inverted: covered lanes recede, untouched ones stay dashed orange saying "nobody tested this", and the header counts lanes still dark rather than percent done. File a finding and it fingerprints your words. If someone already reported it differently, it stops you and makes you either +1 theirs or say something new. Bugs, feature ideas, design feedback and real code patches all pin to the same board as paper notes. "Open a pull request" fetches the target's live source through the Worker, you edit it in place, and it diffs your edit into a patch. "Red team with AI" fetches the page and runs real checks — CSP, clickjacking, HSTS, nosniff, tab-nabbing, mixed content, viewport, alt text, password-over-GET — pinning every hit through the same duplicate gate. An AI Reviewer reads every patch automatically, answers @bot in chat, and can be asked about any finding. Plus a shared whiteboard, sticky notes, a plan list and chat. The target opens in your own tab, not an iframe, so logins and paywalled pages work.

Models

Claude Opus 5, Llama 3.1, GPT-5 Codex

Cost

~55k

Voted by ₥AN$A, Air Planes, kaleb, Khalil, rookepoole, Smiley “MakeShif”, Striker9, Wil

Load a YouTube signal, project it onto a Three.js theater stage, populate the auditorium with simple luminous avatars, synchronize host controls across local tabs, collect reactions, and let the local Stage Director adjust cinematic cues from audience energy. The room URL can be copied for sharing.

Models

Gpt Sol

Cost

20$ sub

Voted by ₥AN$A, Air Planes, Auny 🧡, durdy, kaleb, rookepoole, Sharon, Wil

Uers join a collaborative room and talk trash about a website

Models

Claude Opus 5

Cost

87k tokens

Voted by ₥AN$A, Air Planes, Auny 🧡, durdy, kaleb, rookepoole, Smiley “MakeShif”, Wil

Scroll to ottom use organizer tools

Models

Claude Opus 5

Cost

87k tokens

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Tia, Wil

CampAI reimagined as a welcoming, woodsy workspace. It opens in dark forest mode, with a remembered light-mode option, clear sidebar navigation, the current challenge, community stats, and real builds. The design carries through builds, voting, the leaderboard, builder profiles, and account pages. Profile photos and initials share consistent sizing, and leaderboard rows keep names and scores aligned. This is a fork of the existing platform. The production code retains PostgreSQL, accounts, sessions, submissions, four-category ratings, and existing URLs. Adoption requires no schema migration or account transfer. A read-only preflight script and staging/cutover/rollback guide are included. Security improvements isolate uploaded HTML with an opaque sandbox origin, add response headers, and normalize login error messages. The ChatGPT Sites preview shows actual public CampAI content. Account and submission actions continue on the existing service; the PR contains the database-backed application changes.

Models

GPT-6 through Codex.

Cost

Built using my remaining Codex subscription allowance. Exact token count and dollar cost were not tracked.

Voted by ₥AN$A, Air Planes, durdy, rookepoole, Smiley “MakeShif”, Wil

BRAID is a collaborative creative-writing room built around how groups actually make stories, scripts, lyrics, and worlds together. Writers join as persistent guest profiles with their own unique IDs, then co-author the same structured project in real time. Instead of one giant shared textarea, projects are organized into chapters, scenes, verses, beats, characters, and world notes. Collaborators can see who is present and where they are writing, leave inline comments and rewrite suggestions, accept changes, create version checkpoints, restore earlier states, and branch the entire manuscript to explore alternate directions without destroying the original. A built-in Muse system can also generate contextual brainstorming prompts. Room/project state and guest identity persist between visits, so it feels like an actual creative workspace rather than a temporary multiplayer demo.

Models

GPT-5.6 Sol for product design, implementation, debugging, and iteration. BRAID itself does not require a paid AI model/API to function; its current Muse brainstorming system runs locally in the app.

Cost

Built using my existing ChatGPT access. Exact session token usage/cost is not exposed to me. The deployed BRAID app itself currently has $0 required AI/API cost per user because its core collaboration and Muse features do not depend on paid model calls.

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Smiley “MakeShif”, Wil

Allows org, vols, and recipients to register and manage their drive using Google routes and flyer management.

Models

Opus

Voted by ₥AN$A, Air Planes, rookepoole, Smiley “MakeShif”, Tia, Wil

Migrated Third Bass — a collaborative planning app where teams raise "red flags" on plan steps and a concern only becomes visible once someone seconds it — off Supabase onto Railway Postgres, then added the editing controls it never had. On theme: it's a fork of a fork, since Third Bass is itself a remix of SECONDED.

Models

GroupGPT/Claude Sonnet 5

Cost

$16

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Tia, Wil

Camp Commissary — 45 seconds Right now a 🏕️ AI build lives for exactly one night. You ship in thirty minutes, everyone votes, and then it's a row on a leaderboard. Two seasons in, that's 158 builds we've thrown away. Camp Commissary turns them into stock. A builder lists what they shipped as a seed — not a product, a starting point — and sells the build log, the export, or a single run. Anyone can fork it, and the original builder gets credited permanently plus 5% of whatever the fork earns. Builder keeps 80%, camp takes 15%. Not everything gets in. Two-thirds of our builds don't clear the bar today, and that's the point — it's a shelf, not an archive. The league doesn't change. No wallet, no checkout, nothing to pay before you compete. The thirty-minute window is exactly what it was. The archive stops being a graveyard and becomes inventor

Models

Claude fable 5.1

Cost

450k tokens

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Smiley “MakeShif”, Wil

A community drive has three groups who normally never see the same information. Families in need fill out a paper form. Volunteers text a coordinator to ask when to show up. And one exhausted organizer keeps the truth in a spreadsheet nobody else can open. Drivehouse is one page with three doors into one shared record. A family walks through the first door and registers themselves — two minutes, no proof of need, no account. They get a four-character code instead of a password, and can check their status or withdraw any time. A volunteer walks through the second door, picks the jobs they're up for, claims a shift off the board, and checks in when they arrive. Their hours count themselves. The organizer walks through the third door into a console that already knows everything: who registered, who's on the floor right now, what's on the shelf, which shifts are short. One button walks the shelf, reserves gifts for the families still waiting — oldest request first — and assigns each delivery to a driver in that neighborhood. They review the plan and apply it. The trick is that all three doors write to the same page. When a family registers, the organizer's tally moves. When a donation is logged, the family's status advances. Nobody reconciles anything, because there's nothing to reconcile — there's one record, and everyone is looking at it.

Models

Claude Opus 5

Cost

87k tokens

Voted by ₥AN$A, Air Planes, Auny 🧡, Debbie, Hovhannes Mkhitaryan, Is Wilson, kaleb, rookepoole, Smiley “MakeShif”

SignTrail by @rookepoole is a browser-based PDF signing tool that produces byte-match integrity receipts, but left the audit trail trapped in a companion JSON file. This remix automatically appends a tamper-evident, DocuSign-style Certificate of Completion page directly to finalized PDFs. When you finalize a document, SignTrail generates the verification ID and completion timestamps, synthesizes an audit page detailing the original SHA-256 fingerprint, signer identity, and field inventory, and binds it permanently into the PDF before computing the final signed hash. To use it, drop any PDF into the workspace, place and complete your fields, and select Finalize document. The final PDF now carries its human-readable certificate on the last page while maintaining SignTrail's strict byte-for-byte cryptographic receipt verification.

Models

Antigravity (Gemini 3.8 Flash)

Cost

Not exposed in session UI (~15 turns, standard token rate)

Voted by ₥AN$A, Air Planes, durdy, rookepoole, Smiley “MakeShif”, Wil

Each song room is a single Yjs CRDT doc (one Y.Text per section for character-level concurrent editing, plus a Y.Array of comment threads per section), synced live over a custom WebSocket server built directly on y-protocols/sync + awareness. Upload a reference track and it renders as a wavesurfer.js waveform; song sections (Verse/Chorus/Bridge/etc) are literal timestamp regions on that waveform, and the active section highlights karaoke-style as the track plays. Multiple writers type in the same section simultaneously with live presence cursors/avatars, leave inline comments per section, save named version snapshots to roll back to (restores merge live via the CRDT, no reconnect needed), and pull AI rewrite/brainstorm suggestions per section from an OpenRouter-backed panel. Styled fully brutalist: thick borders, hard offset drop-shadows, Archivo Black + Space Mono type, dark high-contrast theme.

Models

OpenRouter → anthropic/claude-haiku-4.5 (server-side proxy for AI rewrite/brainstorm suggestions)

Cost

Didn't track exact tokens — a handful of test calls (~5-10), each capped at 400 max_tokens on Claude Haiku via OpenRouter. Well under $0.01 total.

Voted by ₥AN$A, Air Planes, Auny 🧡, Hovhannes Mkhitaryan, Is Wilson, kaleb, rookepoole, Smiley “MakeShif”

Red team work usually lives in chat threads, spreadsheets, and someone's notes, so findings get lost and nobody agrees on what's critical. CodeRed gives the crew one board. A tester files a finding with repro steps and evidence. A reviewer sets the severity, assigns an owner, and validates it. A maintainer resolves it once the fix is verified. Every step shows up in a live feed so the whole team sees the state of the engagement without asking. And between engagements, the crew sharpens up in the training lab: realistic scenarios where you write the report you'd actually file, get scored against a reference answer, and climb the leaderboard.

Models

Claude fable 5.1

Cost

292k tokens

Voted by ₥AN$A, Air Planes, Auny 🧡, durdy, Hovhannes Mkhitaryan, kaleb, rookepoole, Sharon, Wil

Everyone trashes one site. Tap and leave comments. Download md file for fixing.

Models

Grok

Cost

<$5

Voted by ₥AN$A, Air Planes, Auny 🧡, durdy, kaleb, rookepoole, Sharon, Wil

Air Audit Live by Airplanes is a collaborative website testing tool where guests join instantly with just a username—no sign-up, no fees, no backend. Teams load any URL, thrash the site together by clicking, commenting, and pinning issues across Design, Function, and Security audits, with everything updating live for everyone in the session using local browser sync.

Models

Deepseek Bolt

Cost

300k

Voted by ₥AN$A, Air Planes, Auny 🧡, durdy, kaleb, rookepoole, Sharon, Wil

A group points at one live website, tries to break it, and leaves hits in real time. No login. 1. Paste the URL. Name yourself. Start. 2. Copy the invite. Everyone opens the same link. 3. Poke the site. Drop a break, bug, note, or ok. 4. The feed updates for the group. Tap THRASH (top left) for a new session.

Models

grok build

Voted by ₥AN$A, Air Planes, kaleb, Khalil, rookepoole, Smiley “MakeShif”, Wil

Walking through it like a live demo: 1. **Join** — you land on a gate screen, type a name, and either generate a random room code or paste one a friend sent you. Hit "Enter the room." 2. **Share** — the room code lives right in the URL (`#room=ABC123`), so "Copy invite link" gives you something you can just paste in a group chat. Whoever opens it lands in the same room. 3. **Load something** — anyone pastes a direct video or audio link into the box at the bottom and hits "Play for everyone." It detects whether it's audio or video and loads it for the whole room simultaneously. 4. **Stay in sync** — from there, whoever plays, pauses, or drags the seek bar, everyone else's player jumps to match. It corrects for network lag using timestamps, so people don't slowly drift out of sync over a long video. 5. **React together** — tapping an emoji sends a burst that floats up over the video for everyone in the room at the same instant, so it feels like reacting in the same room instead of typing into a void. 6. **Chat** — a normal live chat panel alongside the player for running commentary. 7. **Pin a moment** — instead of just chatting past a great scene, "Pin this moment" drops a marker on the seek bar tied to that exact timestamp, with a note. It shows up in the Moments tab too, and anyone can click it later to jump straight back to that point — so the best bits of the watch don't get lost in scrollback. Under the hood, there's no backend server: browsers connect directly to a public real-time relay (PartyKit) keyed by the room code, and broadcast small JSON messages for playback state, chat, reactions, and moments. That's also why it can ship as a plain static file / Hugging Face static Space with nothing to configure.

Models

Gemini/Claude

Cost

I don't have access to the actual token counts for this conversation, so I can't give you an exact number — but here's a reasonable estimate based on what got generated: - The main build (single HTML/CSS/JS file, ~740 lines) plus the design-planning pass and the HF Space packaging round out to roughly **40–70k tokens** across the conversation. - Using Claude Sonnet 5 API pricing, that's very roughly **$0.30–$0.60**, though actual cost depends on your specific plan/interface and whether you're on API billing or a subscription (in which case there's no per-token charge to report). If you need a precise figure for the submission, the accurate way to get it is checking your Anthropic Console usage dashboard for this session rather than my estimate — happy to help interpret it if you pull those numbers.

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Smiley “MakeShif”, Wil

A simple app for groups to volunteer to match their missions.

Models

GroupGPT/Claude Sonnet 5

Cost

$4

Voted by ₥AN$A, Air Planes, rookepoole, Smiley “MakeShif”, Wil

FlagHive: a visual, collaborative planning workspace remixed from Third Bass, which builds on SECONDED. I replaced the aviation theme with emoji identities, three-color feedback, line-by-line discussion, and a complete decision-to-final-plan workflow. Open the public link, sign in with ChatGPT, and choose your name and emoji. The People directory shows recognizable names and stable account tags. Create a plan with one step per line—or open the sample to try it immediately. Reviewers flag individual lines and explain why: 🚩 Concern: the reason stays sealed until two distinct accounts raise a red flag. Flag authors stay anonymous. 🟡 Almost there: named feedback suggesting refinement. 🟢 Love this: named feedback explaining what works. Once a flag lands, the line gets a named discussion thread. The planner can update, accept, or remove a line, with a required justification—or stop the entire plan with an explanation. After every seconded-red and yellow concern is resolved, the planner gives the green light. FlagHive produces a copyable final plan with corrections applied and removed lines omitted. Changes sync every five seconds.

Models

Gpt6

Voted by ₥AN$A, Air Planes, Auny 🧡, Hovhannes Mkhitaryan, kaleb, rookepoole, Smiley “MakeShif”, Wil

One player hosts a room and guests join with a 4-letter code or invite link. Players answer fun prompts anonymously, guess who wrote each answer, react to reveals, earn points, and finish with group-voted awards.

Models

GPT-5.6 Sol

Cost

$0 additional cost — subscription-based plan

Voted by ₥AN$A, Air Planes, durdy, kaleb, rookepoole, Sharon, Wil

I built an app that allows multiple users to "Red Team" a website.

Models

GroupGPT/Claude Sonnet 5

Cost

$1

Voted by ₥AN$A, Air Planes, kaleb, rookepoole, Smiley “MakeShif”, Wil

Four roles, and the amount you have to prove scales with what you're trusted with, not with what you need. Volunteers and providers sign in with a name and no password. Organisers get a real gate because they see real names. Households never get an account at all — they're given a two-word emblem like "The Winter Wren" and that's their only key. A household asks for something through a form that wants to know who's around your table and what a child loses track of time doing, not their clothing sizes. That creates a square on the public quilt with no name on it. A volunteer opens a square and the delivery address is gated twice: signed out it isn't rendered at all, and signed in you still have to take the square on before it appears. Then they get the address, the handover method the family chose, and a QR code. The shelter sits on the same quilt — foster a pet, book a walking slot, or cover what the shelter is short of. And a sponsored-work lane pays people $18–25/hr or $20 a patch, funded by sponsors, open to anyone including people who are on the board themselves.

Models

Claude Opus 5, Gemini 3.6 Flash, Grok, GPT Luna

Cost

~830k

Voted by ₥AN$A, Air Planes, kaleb, rookepoole, Smiley “MakeShif”, Wil

The name is the message already sitting unanswered in your group chat. Finding the overlap is the part every tool already does. When2Meet finds it and stops, handing confirming and reminding back to you; Doodle and the open-source clones do the same. Meanwhile no-shows run 20-30% at corporate campuses and ~32% across events, and the research is consistent: the weakest formats are the ones where not attending costs nothing. A green box costs nothing. That's why nobody shows up. So this doesn't add reminders — it makes the yes cost something. 1 · Get in. The room code mints itself into the url, so the address bar is the invite. Name only, no account. 2 · Say what you can't make. Three ways, all optional: a Google sync, your calendar's secret .ics link, an .ics file parsed in your browser, or just tap the times. You only ever flag conflicts — leaving a time open is deliberately not a promise to be there. 3 · The board finds what's clear for everyone. One person's conflict is the table's conflict, so a green slot is green for all of you. The headline says how many times are clear and how many people have actually synced. Anyone can also propose a specific date and time, and the slot carries the name of whoever asked for it. 4 · Stake five chips on the times you'll genuinely turn up for, stacked if you mean it. You can't chip everything, and you can't stake a time you marked busy. The board closes itself the moment no chip left in anyone's hand can catch the leader — so nobody waits on a poll that's already decided. 5 · Show up, or don't. Turn up and your stake comes back. Don't, and it burns, and you're short next time someone asks when you available. Then the AI notetaker writes the meeting into the room's record — the time called, what it won on, minutes from first chip to close, who staked what, who showed — and deals a fresh board with chips carrying across. The room is the team; meetings are rounds inside it. Hit Deal a demo table to see the whole loop with four people already at it.

Models

Claude Opus 5

Cost

350k

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Sharon, Smiley “MakeShif”

Start a chain by naming 3–10 players and writing the original message. The device is passed to the next player behind a concealed handoff screen. Each person sees only the version immediately before theirs, rewrites what they remember, and passes it forward. Groups can play on one device or use remote handoff links. When the final player submits, HEARSAY places the original and final messages side by side, calculates a drift score, identifies what was dropped, invented, or amplified, and exposes every version through a complete chain-of-custody timeline. No accounts or sign-in are required.

Models

OpenAI Codex (GPT-5.6) for design and implementation. HEARSAY does not call an AI model at runtime; its drift analysis runs locally using custom comparison logic.

Cost

$0 in external API or runtime costs. Built within my ChatGPT plan; Codex did not surface an exact token count.

Voted by ₥AN$A, Air Planes, durdy, rookepoole, Smiley “MakeShif”, Wil

Collects writer inputs

Models

ChatGPT

Voted by ₥AN$A, Air Planes, Auny 🧡, Hovhannes Mkhitaryan, kaleb, rookepoole, Smiley “MakeShif”, Wil

Bar based icebreaker drinking game

Models

Claude Opus 5

Cost

32k tokens

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Sharon, Wil

Sandbox DOM editor. Keep track of comments and edits from a central plane. Load the page within the site.

Models

Opus 5

Voted by ₥AN$A, Air Planes, Auny 🧡, Hovhannes Mkhitaryan, kaleb, rookepoole, Smiley “MakeShif”, Wil

People submit answers to questions and vote on others after reading their submissions. We see how much collective ice we broke.

Models

Grok

Cost

<$5 in credifs

Voted by ₥AN$A, Air Planes, Auny 🧡, Is Wilson, kaleb, rookepoole, Smiley “MakeShif”

The flow is basically: signed PDF ↓ calculate SHA-256 ↓ compare with receipt ↓ receipt ↓ recalculate receipt digest ↓ compare with stored digest ↓ both match ↓ VERIFIED

Models

Codex Spark

Cost

$.0

Voted by ₥AN$A, Air Planes, Auny 🧡, durdy, Hovhannes Mkhitaryan, kaleb, rookepoole, Smiley “MakeShif”, Wil

I built a red teaming collaboration site for a builder crew to test and work together in real time. When users land on the homepage, they can click LOGIN (DEMO), which uses a fake login flow (no auth backend yet) and routes them to a mission dashboard. The landing page is fully re-themed for red teaming, with sections focused on reporting findings, triage flow, collaboration, and team training. The dashboard gives the crew quick access to Bug Reports, Project Rooms, Training Labs, and Team Activity. This version is intentionally no-backend for speed, so we can test UX and team workflow first, then add real auth + database once the flow is validated.

Models

GPT-5.3-Codex (Factory Droid)

Cost

Estimated ~60k–120k tokens total across design + implementation iterations.

Voted by ₥AN$A, Air Planes, rookepoole, Smiley “MakeShif”, Wil

HOME RUN is a collaborative planning and decision tool. Create a plan, let others review it, and flag risks or concerns. Use Premortem to imagine how the plan could fail, and Challenge Flight to test it from different perspectives. Home Run automatically maps risks, disagreements, consensus, and failure points through visual diagrams. Review the Decision Cockpit, resolve concerns, compare plans, and use the final Decision Record to understand what was changed and why. Everyone can participate at the same time—no turns and no room codes.

Models

Chat gpt bolt

Cost

300k

Voted by ₥AN$A, Air Planes, Hovhannes Mkhitaryan, kaleb, rookepoole, Smiley “MakeShif”, Wil

Everyone opens the link and lands in a room that already exists — a code is generated for you, so the only thing it asks is your name. You get assigned a silly face you don't get to pick, and you're in the chat immediately. Then you confess. Twelve food crimes (pineapple on pizza, boneless wings, replying "k"), answered privately on your own phone, everyone at the same time. The second you finish, a bot called The Judge reads you out loud using only your own answers: "@AUNY objected to 11 of 12. A genuinely difficult person." You get roasted alone before you know anyone else's votes. Then the trial. One crime at a time, the split is revealed and the Judge names whoever stood alone: "Boneless wings. 7 said abomination. One did not. @SAM stands alone. Explain yourself." Everyone piles on in the chat. You can @ each other and @judge, and it answers from the real tally. After twelve crimes the screen flips from black void to bright paper and hands out four verdict cards: most cursed individual, insufferable purist, the one thing you all agreed on, and unlikely allies — two people who were alone together against the whole room and didn't know it. That last card is the actual icebreaker. You get there by arguing first. Then there's an arcade: snake, nerve (hold and release at exactly ten seconds, no clock) and reflex, all on one shared live leaderboard with the Judge calling the standings.

Models

Claude Opus 5

Cost

150k

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Sharon, Smiley “MakeShif”

You tell me. That’s the test

Models

GPT 5.5

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Smiley “MakeShif”, Wil

When you really need to meet.

Models

GroupGPT/Claude Sonnet 5

Cost

100k

Voted by ₥AN$A, Air Planes, Auny 🧡, rookepoole

A host creates a room and adds a plan manually or imports a PDF, DOCX, TXT, Markdown, CSV, JSON, HTML, or RTF document. FAULTLINE breaks the plan into reviewable sections and shares them with separate no-sign-in guest profiles over WebRTC. Everyone reviews independently and anonymously, flagging specific sections with severity, concern type, and an optional explanation. Crucially, nobody sees anyone else's concerns during review, preventing the first objection from anchoring the group. Once everyone is ready, the host reveals the results and FAULTLINE ranks the plan's real fault lines by anonymous reviewer count, severity, and section.

Models

GPT-5.6 Sol

Cost

Built in ChatGPT. Exact token usage and per-build API cost were not exposed.

Voted by ₥AN$A, Air Planes, kaleb, Wil

Airplanes Mode is a local harness that takes an idea from plan to ship: someone writes a brief and a design, someone builds against that design, a checklist walks the real artifact without fixing it, then a gate must pass before anything ships, and two independent reviewers score the finished work from a fresh context so one person never builds and approves their own output.

Models

Glm 53 deepseek flash v41 opus 5

Cost

100 plus

Voted by ₥AN$A, anakhuagbor sule, Auny 🧡, durdy, kaleb, rookepoole, Smiley “MakeShif”, Wil

AI powered debate like decision making based o gorup consensus

Models

Claude Opus 5

Cost

61.2k tokens

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Smiley “MakeShif”, Wil

Guest join, no account. Pick an avatar from the strip. Type a guest name. Enter the circle. Camera orbits the ring. Sixteen stylized people, lanterns, a pit of embers. Select. Choose who leans next from the list, or tap Circle chooses. Everyone takes a turn. The faller steps to the inner edge. Chat opens. They plead. The rest talk, needle, and vote. 60-second vote. CATCH needs at least half the standing circle (ceil(n / 2) of 16 is 8). Abstain counts as not catching. Timer ends → cinematic Held (pulled back) or Taken (drops into the pit as a ghost). Then the next person is selected until all 16 have leaned. Finale lists who was held and who was taken.

Models

Grok build

Cost

Sub

Voted by ₥AN$A, Air Planes, kaleb, rookepoole, Smiley “MakeShif”, Wil

Create or join a room with a shareable code, paste any direct video/audio URL (.mp4/.mp3/.webm), and it loads for the whole group at once. Play, pause, and seek on the HTML5 player are captured and broadcast over Socket.IO — the server holds authoritative room state (currentTime + playing, drift-corrected against wall-clock time) so a late joiner snaps to the live position instead of starting at 0:00. On top of sync there's a live chat panel and six one-tap emoji reactions that float up over the player like a stream overlay, so it feels like everyone's actually in the room together instead of watching alone on staggered timers.

Models

Claude Sonnet 5 (via Claude Code)

Cost

Single Claude Code session, one pass, no separate token/cost tracking on this plan.

Voted by ₥AN$A, Auny 🧡, kaleb, rookepoole, Smiley “MakeShif”, Wil

You break the ice by being interactive

Models

Claude Haiku 4.6

Cost

12,500 tokens

Voted by ₥AN$A, Air Planes, anakhuagbor sule, Auny 🧡, kaleb, rookepoole, Smiley “MakeShif”, Wil

One device, passed around. Everyone answers privately and the screen clears between hands. Each person marks every option great / fine / can't. A "can't" is a hard limit and needs a stated reason the group sees. Options die on a named limit, not a popularity contest, so nobody has to argue anyone down. Survivors rank by their worst-off person, not the average. The winner is the option whose unhappiest person is least badly off, which protects whoever drew the short straw. If everything gets blocked, it finds the option blocked by exactly one person and asks only them. Nobody else moves, nobody is asked twice, and "no" is a legitimate answer. Then everyone signs. Refuse and your objection becomes a new limit and it re-runs. Unanimity, not four out of five. In the 3D scene each person is a bloom and each option a branch to everyone who can live with it. A limit snaps the branch and the option falls to the floor, tagged with who killed it and why.

Models

Claude Opus 5 (Claude Code), GPT-5 Codex

Cost

1.2M (that's what my dumb claude says)

Voted by ₥AN$A, Air Planes, kaleb, Wil

https://github.com/40verse/CAGE/blob/main/README.md

Models

Opus 5

Cost

n/a

Voted by ₥AN$A, Air Planes, Auny 🧡, durdy, kaleb, rookepoole, Smiley “MakeShif”, Wil

A team starts with a problem and discusses it normally. Messages can then be converted into a Decision Graph containing options, evidence, risks, and constraints. Teammates vote with confidence scores, while Converge tracks decision readiness and exposes unresolved reasoning. Team members can explicitly connect conflicting claims as contradictions; an unresolved contradiction blocks the final decision until the team records how it was resolved. Once the reasoning is clear, a named final decider can commit the winning option and assign the resulting action, owner, and deadline. The app supports shared rooms, persistent state, and live updates between connected users.

Models

GPT-5.6 Sol

Cost

A team starts with a problem and discusses it normally. Messages can then be converted into a Decision Graph containing options, evidence, risks, and constraints. Teammates vote with confidence scores, while Converge tracks decision readiness and exposes unresolved reasoning. Team members can explicitly connect conflicting claims as contradictions; an unresolved contradiction blocks the final decision until the team records how it was resolved. Once the reasoning is clear, a named final decider can commit the winning option and assign the resulting action, owner, and deadline. The app supports shared rooms, persistent state, and live updates between connected users.

Voted by ₥AN$A, Air Planes, Auny 🧡, rookepoole

Two player game, one player clicks 'create a case' and invites their friend to help solve the crime

Models

Austra OpenAI

Cost

???

Voted by ₥AN$A, Air Planes, Auny 🧡, rookepoole

A concern is heard when a second person has it too. Paste a plan, one line per line, and share the room code. Reviewers tap lines to flag them with an optional one-line why, and everyone must flag at least two lines before sealing, so having flagged at all identifies nobody. The owner reveals once two people have sealed. A line's flag text unseals only when two independent people flagged it; a lone flag stays a black redaction bar reading "sealed · 1 of 2", for everyone, the owner included, forever. Tally marks and a heat bar show frequency, the footer names the heaviest lines. Then the owner must answer every seconded line (fix / accept / stop) before the stamp flips from NO GO to GO.

Models

Claude Code, Grok, Codex

Cost

60k

Voted by ₥AN$A, Air Planes, Auny 🧡, rookepoole

Open the link. Type a name. You’re in. File a Plane if you’re starting it — six short steps, then it’s in the hangar. Everyone else taps that plane. Left side is the plan. Right side is the concern board. Flag the part that bothers you. Your name never shows. The red bar on that part gets longer. Hottest bar is where the group actually has a problem. That’s the point — not a comment thread.

Models

Grok bolt

Cost

300k

Voted by ₥AN$A, Air Planes, Auny 🧡, rookepoole, Smiley “MakeShif”, Wil

Jump into a live roulette table where everyone’s watching the same round and countdown. While betting’s open, you place your own virtual-credit bets, and you can see what everyone else is throwing down too. Once betting locks, that’s it—no changes. The server picks the winning number, we all watch the same wheel animation land on it, then winning bets light up and credits get auto-paid. Chat, react, send gifts, check who won, and peep the leaderboard for the running scores. Then it all resets and a new round kicks off automatically.

Models

Deepseek V4 Flash

Cost

60k

Voted by ₥AN$A, Air Planes, Auny 🧡, rookepoole

A proposal owner pastes or writes a plan (trip, project, decision); it's auto-split into sections. They get a share link for the group and a private owner-only dashboard link. Group members open the share link and tap any section to anonymously flag it as a "concern" or "dealbreaker," optionally with a short note — no login, one flag per person per section (tracked via a per-device anonymous ID, enforced server-side with a unique constraint). The owner dashboard shows the proposal text re-rendered as a heatmap — sections with more flags glow redder — with flag counts and notes attached. Notes only unlock once a section hits 2+ flags, so no single flag is attributable to one person. The dashboard polls live so results update as the group responds. Stack: Next.js (App Router) + Tailwind + Postgres, deployed on Railway.

Models

GroupGPT / Sonnet 5

Cost

$3.44

Voted by ₥AN$A, Air Planes, Auny 🧡, kaleb, rookepoole, Wil

Build Hangar is Airplane Studio Lab’s build-night board: live clock, one submission per episode, drafts, overall ratings, comments, and a season leaderboard. It covers the same job as Camp AI’s public Next.js league but is not a fork of it. It runs as a normal lab Node/SQLite project on airplanestudio, keeps the database off the publish path, enforces the window and uniqueness in the database, hardens sessions and uploads, adds webhooks and an audit log, and fixes the mobile drawer so choosing Builds actually leaves the menu.

Models

Sonnet Opus 5

Cost

100k

Voted by ₥AN$A, Debbie, rookepoole

Built live during Episode 7 — "Calendar" on 🏕️ AI. Deployed calendar build.

Models

Not specified in the original submission

Voted by ₥AN$A, Air Planes, kaleb, Wil

Live

Models

Claude bolt

Voted by ₥AN$A, Air Planes, Auny 🧡, rookepoole

Words with Planes puts you at a busy little airfield running freight instead of dogfights. Letters drop in as cargo crates and you turn them into words to load onto your plane, but nothing you claim is ever fully yours since anyone quick and clever enough can steal a word right out from under you by stretching it into something longer and different. Theres also a faster mode where you race the clock on short letter fragments and burn through engines if you cant keep up, with wider vocabulary earning you fuel back rather than just fast typing winning the day. Play solo against the tower for practice or jump into live matches against real people, sign in to keep your scores and climb the board, and enjoy some lighthearted callouts along the way.

Models

Opus Deepseek Flash v41

Cost

40k

Voted by ₥AN$A, Air Planes, Auny 🧡, durdy, kaleb, rookepoole, Smiley “MakeShif”, Wil

Camp Red Team trasher

Models

Mulberry Ai Studio

Voted by ₥AN$A, Air Planes, kaleb, Wil

Bizbo is a chat advisor whose entire knowledge base is one real founder's direct answers to 20 questions about building a business — not generic startup advice. You describe your business situation or question, and it answers through that founder's specific lens: validate with presales before writing any code, hold every expense (including your own time) accountable to incremental growth, be your own first market signal, distrust vanity metrics and "build it and they will come," and lead with positioning over paid channels when budget is $0.

Models

openai/gpt-4o-mini via OpenRouter

Cost

~2,000 tokens across test calls during development (system prompt + short conversations). At gpt-4o-mini pricing that's well under $0.01 so far.

Voted by ₥AN$A, Air Planes, Auny 🧡, rookepoole

Create a room and share the 6-character code or invite link with another player. Both players connect peer-to-peer over WebRTC and command a squad of Spike, Ghost, and Warden units on a 7×7 tactical grid. Each unit gets 2 AP per turn for movement, attacks, repairs, and fortification. Players fight over relay influence zones that generate Signal, with the first player to 10 Signal or the last surviving squad winning. Each class also has a once-per-match signature ability. New players can launch Tutorial vs AI, which walks them through movement, relay capture, AP chaining, defense, repair, ranged attacks, and then transitions into a full match against NULL//BOT.

Models

GPT-5.6 Sol

Cost

Built entirely in ChatGPT. No separate API spend; exact in-session token usage wasn’t exposed

Voted by ₥AN$A, Air Planes, durdy, kaleb, rookepoole, Smiley “MakeShif”, Tia, Wil

My app allows a host to ask a question and allows a group to answer it and vote on it to come to consensus.

Models

GroupGPT

Cost

200k

Voted by ₥AN$A, Air Planes, Auny 🧡, rookepoole

A two-player dice race on a 7-tile board. A coin flip seats one player directly on the Throne (worth 5 pts) — that's the entire "lucky start." Each turn, whoever is farther from the Throne rolls two dice and keeps the higher (comeback mechanic); landing exactly on your opponent bumps them back to their home tile. Highest score after 16 turns wins. Built as a Node/Express + Socket.IO app with server-authoritative game state (rooms keyed by a 4-letter code, in-memory), deployed on Railway with a public domain — anyone can create a room, share the code, and play live, no login required.

Models

GroupGPT / Claude Sonnet 5

Cost

$2

Voted by ₥AN$A, Air Planes, kaleb, rookepoole, Smiley “MakeShif”, Wil

One person creates a room and gets a short code. Everyone else joins with that code + a nickname (no accounts/emails). The host sets the decision and options. Each person picks their preferred option and rolls a 3D d20 — the roll becomes their voting weight. Higher totals win. Critical 20s double the weight; 1s cut it in half.

Models

grok bolt

Cost

400k

Voted by ₥AN$A, anakhuagbor sule, kaleb, rookepoole, Smiley “MakeShif”, Wil

gemini 2.5 live with seach grounding 2.5. non oficial api from X o get captions and speakers . at begin we chose the point in with we disagre and it helps reach concensus .

Models

chatGPT chat (not codex) with github , githubactions so it can test and selfrunner

Cost

0

Voted by ₥AN$A, kaleb, rookepoole, Smiley “MakeShif”, Wil

It lets people get rid of the hassle of what’s for dinner forever

Models

Claude Sonnet 5

Cost

$0

Voted by ₥AN$A, kaleb, rookepoole, Smiley “MakeShif”, Wil

End the age old probe

Models

Claude Sonnet 5

Cost

$0

Voted by ₥AN$A, rookepoole

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Deployed game build.

Models

Not specified in the original submission

Voted by ₥AN$A, rookepoole

Built live during Episode 12 — "Audio/Video Clipper" on 🏕️ AI. Deployed build, name matches clipper theme.

Models

Not specified in the original submission

Voted by ₥AN$A, Air Planes, durdy, kaleb, rookepoole, Wil

CIRCLE Vote this circle, or start your own poll. Photos on choices. Share the link. No login.

Models

Models: Grok 4.6 (Grok Build)

Cost

• Cost: ~610k tokens (local Grok Build, no OpenRouter bill)

Voted by ₥AN$A, kaleb, rookepoole, Smiley “MakeShif”, Wil

Sign in (Google, X, or email). File the case in plain language — we tested: “Vegas this Saturday or a house party in Atlanta? Two people are broke. Alex wants a strip-club detour. Sam is a hard no.” Grok, as clerk, parses the poles and writes the ballot: Vegas vs house party on a 1–10, importance 1–10, plus compromise questions (budget, strip-club opt-out, dealbreaker, “what would make this okay”). Share the link. Invitees vote with just a name — no account. If the group already texted you their take, record it as a proxy vote. Call the bench. The judge does not count heads. It weights stance by importance. Our test room: Wil 9/9 for Vegas, Alex 10/8 for Vegas, Sam 3/2 against. Sam’s shrug does not veto. The ruling: Vegas stands, Sam is heard, strip club is an optional side quest, house party is the rain check. Stamped: SO ORDERED — PACK THE BAGS, SKIP THE POLE.

Models

Grok 4.6

Cost

Grok build few thousand

Voted by ₥AN$A, rookepoole

Built live during Episode 6 — "Media Editor + AI + Metadata" on 🏕️ AI. Deployed media/studio build.

Models

Not specified in the original submission

Voted by ₥AN$A, rookepoole

Built live during Episode 10 — "Journal" on 🏕️ AI. Deployed journal build.

Models

Not specified in the original submission

Voted by ₥AN$A, rookepoole

Built live during Episode 11 — "Document Signer" on 🏕️ AI. Deployed build, generic name.

Models

Not specified in the original submission

Voted by ₥AN$A, rookepoole

Built live during Episode 11 — "Document Signer" on 🏕️ AI. Deployed build, generic name.

Models

Not specified in the original submission

Voted by ₥AN$A, kaleb, rookepoole, Wil

Lets people vote on things. See results.

Models

Grok build

Cost

$5 probably in grok build heavy

Voted by ₥AN$A, rookepoole, Wil

Risky text

Models

Opus 5

Cost

Idk

Voted by ₥AN$A, rookepoole, Wil

Don't know yet

Models

Alot of models mostly Gemini

Cost

Less than 10$

Voted by ₥AN$A, rookepoole

Allows user to upload documents or create fresh pdfs from scratch, send them for signature, and maintains a full audit log

Models

Claude Opus 5

Cost

292,000 tokens / ~$3

Voted by ₥AN$A, rookepoole

Built live during Episode 4 — "Game PT.1" on 🏕️ AI. Deployed game (Ricochet Rumble).

Models

Not specified in the original submission

Voted by ₥AN$A, rookepoole

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Deployed game (Ricochet Rumble).

Models

Not specified in the original submission

Voted by ₥AN$A, rookepoole

Built live during Episode 7 — "Calendar" on 🏕️ AI. Deployed calendar build.

Models

Not specified in the original submission

Voted by ₥AN$A, rookepoole

Built live during Episode 10 — "Journal" on 🏕️ AI. Claude artifact build.

Models

Claude

Voted by ₥AN$A, rookepoole

Built live during Episode 8 — "Coding Collaboration Platform + AI + Contribution Tracking" on 🏕️ AI. Claude artifact build.

Models

Claude

Voted by ₥AN$A, rookepoole

Built live during Episode 9 — "Coding Collaboration Platform + AI + Contribution Tracking 2" on 🏕️ AI. Deployed build, generic name.

Models

Not specified in the original submission

Voted by ₥AN$A, rookepoole

Built live during Episode 12 — "Audio/Video Clipper" on 🏕️ AI. Claude artifact build.

Models

Claude

Voted by ₥AN$A, rookepoole

Built live during Episode 13 — "Intake Form" on 🏕️ AI. Claude artifact build.

Models

Claude

Voted by ₥AN$A, rookepoole

Built live during Episode 13 — "Intake Form" on 🏕️ AI. Claude artifact build.

Models

Claude

Voted by ₥AN$A, rookepoole

Built live during Episode 13 — "Intake Form" on 🏕️ AI. Claude artifact build.

Models

Claude

Voted by ₥AN$A, rookepoole

Built live during Episode 14 — "x402" on 🏕️ AI. Same builder pattern (durdyh2o) — likely their x402 submission.

Models

Not specified in the original submission

Voted by ₥AN$A

The user talks or types to the AI and it updates their Calendar and can tell them everything about their Calendar without any searching.

Models

Claude Sonnet 5

Cost

237K

Voted by ₥AN$A

Users can record and add tracks and ask the AI to make edits and changes.

Models

Claude Sonnet 5

Cost

200K

Voted by ₥AN$A

You enter a question and the AI uses a calcluator function to compute the solution and then uses the LLM to explain the why behind the answer.

Models

Claude Sonnet 5

Cost

100k

Voted by ₥AN$A

talk or text basic journal entries

Models

Kimi

Cost

$2.71

Voted by ₥AN$A

A simple one page application that lets users sign documents and prove they signed them.

Models

Claude Sonnet 5

Cost

163K

Voted by ₥AN$A

Built live during Episode 1 — "Calculator" on 🏕️ AI. Universal Celestial Cynchronometer, developed by me and a friend" — explicit calculator-type build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 1 — "Calculator" on 🏕️ AI. Claude artifact, posted as a v2 calculator build.

Models

Claude

Voted by ₥AN$A

Built live during Episode 1 — "Calculator" on 🏕️ AI. Claude artifact build.

Models

Claude

Voted by ₥AN$A

Built live during Episode 1 — "Calculator" on 🏕️ AI. Deployed calculator app.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 1 — "Calculator" on 🏕️ AI. V2 Calculator, video coming soon.

Models

Claude

Voted by ₥AN$A

Built live during Episode 1 — "Calculator" on 🏕️ AI. V1 Calculator.

Models

Claude

Voted by ₥AN$A

Built live during Episode 3 — "Augmented Reality" on 🏕️ AI. Deployed AR app build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 3 — "Augmented Reality" on 🏕️ AI. Deployed AR app build (companion static site).

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 4 — "Game PT.1" on 🏕️ AI. Deployed game, posted by host.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 4 — "Game PT.1" on 🏕️ AI. Claude artifact game build.

Models

Claude

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Deployed game build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Deployed game build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Game-sounding custom domain, posted by host w/o description.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Claude artifact game build.

Models

Claude

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Deployed game build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Deployed game build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Deployed game build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Deployed game build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Deployed game build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Deployed game build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Deployed game build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Claude artifact game build.

Models

Claude

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Deployed game build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Deployed game build (Doom-style).

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Claude artifact game build.

Models

Claude

Voted by ₥AN$A

Built live during Episode 5 — "Game PT.2" on 🏕️ AI. Claude artifact game build.

Models

Claude

Voted by ₥AN$A

Built live during Episode 6 — "Media Editor + AI + Metadata" on 🏕️ AI. Claude artifact, media editor build.

Models

Claude

Voted by ₥AN$A

Built live during Episode 6 — "Media Editor + AI + Metadata" on 🏕️ AI. Deployed build, but "PromptRouter" name suggests different theme than media editor.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 6 — "Media Editor + AI + Metadata" on 🏕️ AI. Deployed build, name matches "metadata" theme.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 6 — "Media Editor + AI + Metadata" on 🏕️ AI. Deployed build, posted by host, name matches media editor theme.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 6 — "Media Editor + AI + Metadata" on 🏕️ AI. Claude artifact build.

Models

Claude

Voted by ₥AN$A

Built live during Episode 7 — "Calendar" on 🏕️ AI. Deployed calendar build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 7 — "Calendar" on 🏕️ AI. entry for a calendar" — explicit submission.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 7 — "Calendar" on 🏕️ AI. Deployed calendar build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 7 — "Calendar" on 🏕️ AI. Deployed calendar build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 7 — "Calendar" on 🏕️ AI. Deployed calendar build, posted by host.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 8 — "Coding Collaboration Platform + AI + Contribution Tracking" on 🏕️ AI. Deployed coding-collab build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 8 — "Coding Collaboration Platform + AI + Contribution Tracking" on 🏕️ AI. Deployed coding-collab build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 8 — "Coding Collaboration Platform + AI + Contribution Tracking" on 🏕️ AI. Deployed build, name matches contribution-tracking theme.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 8 — "Coding Collaboration Platform + AI + Contribution Tracking" on 🏕️ AI. Deployed build, name matches contribution-tracking theme.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 8 — "Coding Collaboration Platform + AI + Contribution Tracking" on 🏕️ AI. Deployed coding-collab build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 9 — "Coding Collaboration Platform + AI + Contribution Tracking 2" on 🏕️ AI. Deployed build, name matches collab-workspace theme.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 10 — "Journal" on 🏕️ AI. Explicitly labeled "Basic Journal.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 10 — "Journal" on 🏕️ AI. Deployed journal build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 10 — "Journal" on 🏕️ AI. Repo, plausible journal-related build via Telegram/Obsidian.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 10 — "Journal" on 🏕️ AI. Deployed build, name matches voice-journal theme.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 10 — "Journal" on 🏕️ AI. Claude artifact build.

Models

Claude

Voted by ₥AN$A

Built live during Episode 10 — "Journal" on 🏕️ AI. Deployed journal build.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 10 — "Journal" on 🏕️ AI. Deployed build, name matches journal theme.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 11 — "Document Signer" on 🏕️ AI. Claude artifact build.

Models

Claude

Voted by ₥AN$A

Built live during Episode 11 — "Document Signer" on 🏕️ AI. Deployed build, name matches document-signer theme, posted by host.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 11 — "Document Signer" on 🏕️ AI. Built through Telegram... document signer" — explicit submission.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 12 — "Audio/Video Clipper" on 🏕️ AI. Deployed build, name matches AV clipper theme.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 12 — "Audio/Video Clipper" on 🏕️ AI. Deployed build, name matches AV clipper theme.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 12 — "Audio/Video Clipper" on 🏕️ AI. Claude artifact build.

Models

Claude

Voted by ₥AN$A

Built live during Episode 13 — "Intake Form" on 🏕️ AI. Same builder (durdyh2o) as confirmed calendar/doc-signer entries — likely their intake-form submission.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 13 — "Intake Form" on 🏕️ AI. Deployed build, posted by host.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 14 — "x402" on 🏕️ AI. Deployed x402 marketplace build, posted by host.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 15 — "x402 PT2" on 🏕️ AI. Deployed x402 build, posted by host.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 15 — "x402 PT2" on 🏕️ AI. Deployed x402 build, posted by host.

Models

Not specified in the original submission

Voted by ₥AN$A

Built live during Episode 15 — "x402 PT2" on 🏕️ AI. Domain name directly references x402.

Models

Not specified in the original submission