My Studio
Answer once — the Brandforge turns it into a complete brand: promise, positioning, PODs, taglines, voice, pitches and a brand sheet. Everything saves in this browser as you type. Pulldowns first (the easy part), then six sentences only you can write.
Part A · Pulldowns — pick, don't type
Not sure about value type or archetypes? Leave them on Let the AI recommend — the strategist will choose and justify in its critique.
Part B · The words only you can write — all six required
The rule of the sessionThe AI may only use what you write here — it never invents facts. Short, concrete sentences beat essays. Real client words are gold: paste them in Part C.
Part C · Optional — but this is where brands get good
AI Setup
Nothing to do here — the school's managed AI does the work. Advanced options below are entirely optional.
🎓 You're all set — no key, no accountThe Brandforge runs on the school's managed AI. Go straight to ③ The Forge. School-mode generations are logged on the school server (it pays for the tokens); everything you type still lives in your browser.
Advanced (optional) — bring your own OpenRouter key
Only for makers who want their own AI account: a free OpenRouter key keeps every byte in your browser. Ignore this box otherwise.
From openrouter.ai → Keys (free account). Stored only in this browser — never on our servers — and shared with CollabForge so you only ever paste it once. Your key always takes priority over school AI when present.
Any model your OpenRouter account can use — paid models draw from your credit.
Loading the live model list from OpenRouter…
Model realityFree models cost nothing but are rate-limited (≈20 requests/minute, 50/day) and occasionally busy — the Forge retries automatically. The paid picks are pennies per full run — they need credit on your OpenRouter account. A full run is 3 requests; the final document is a 4th.
The Forge
Three passes, each one nastier than the last. Pass 1 — a senior strategist drafts your full brand from your answers. Pass 2 — an adversary attacks the draft (competitor test, cliché scan, provability) and rebuilds it stronger. Pass 3 — a harsher adversary repeats on Pass 2, assuming a skeptical luxury buyer reads every word. Then you pick your favourite.
1
PASS 1 · The Strategist — draft the full brand package
Uses only your answers. Recommends value type & archetypes if you left them open.
queued
2
PASS 2 · The Adversary — attack, then rebuild stronger
Competitor test · customer test · delivery test · cliché & sameness scan.
queued
3
PASS 3 · The Adversary II — the harsh one
Assumes a skeptical luxury buyer, a trademark attorney and a tired fair visitor read it. Escalates specificity.
queued
Live from the model
⤓ You scrolled away from the live output —
The three forged brands — pick your favourite
Brand Document
Your chosen pass, written up as the finished brand document — the session's one-page brand sheet, grown to full size: promise, positioning, strategy, voice, expression, the four pitches, story, touchpoints and your first week.
No document yet — forge three passes, pick a favourite, then build.
✒
Writing the document — composing from your chosen pass
Streaming…
running
The human edit — the session's ruleAI is a brilliant intern with no taste. Before anything ships, add the details only you know: the stone's mine, the third attempt that finally worked, the client's own words. Specificity is the anti-slop.
FAQ & Privacy
What exactly gets sent where?
Your answers, and each pass's draft, are sent to the AI model you selected — through OpenRouter's API, over HTTPS, from your browser. On jdmis.edu.sg the school also collects anonymous usage counts only (sessions, runs completed, documents built — the same policy as the Atelier Pricer): never your answers, API key, model choices or content. Anywhere else this file runs, nothing is sent to anyone but OpenRouter. Your OpenRouter key, your answers, the three passes and the final document are stored in this browser's localStorage only. Clear your browser data (or press ⌫ Wipe above) and everything is gone — export first if you want to keep it.
Why three passes — what is "adversarial"?
One AI draft is usually pleasant and generic — the sameness trap from the session. So the Forge argues with itself: Pass 1 drafts; Pass 2 attacks that draft (could a competitor claim the same sentence? is every "because" checkable? which words are slop?) and rebuilds it; Pass 3 attacks the rebuild harder and sharpens again. Three genuinely different takes on your brand — you choose the one that sounds like you.
I don't have an OpenRouter key — can I still forge?
On jdmis.edu.sg: yes. The page detects the school's managed AI gateway and offers 🎓 School AI — no key, no account, runs on the school's key (GLM 5.3 Flash default). Because the school pays for those tokens, school-mode generations are logged on the school server (request + result under data/generations/; anonymous counts in the dashboard). Runs are fully async: jobs live server-side, and if you reload mid-forge the page re-attaches to the running job and finishes the pass. Bring your own key in ② AI Setup if you prefer everything stays in your browser — your key always wins when present.
Do I need to pay for AI?
No — the first group in the model list is :free models from OpenRouter; they cost nothing (an OpenRouter account and key are still needed). They are rate-limited (≈20 requests/min, ≈50/day): a full run uses 3 requests and the document a 4th. If a model is busy the Forge retries automatically. Prefer smoother runs? The paid picks (GLM 5.3 Flash — cheap & intelligent; GPT OSS 120B — cheap & fast) cost pennies per full run and are exempt from the daily limits — they need credit on your OpenRouter account. You can also type any model ID your account can use.
Why does it ask for these particular answers?
They are the intake of the JDMIS MK100 marketing methodology: what you make (offer), who really buys (target market), why they choose you (POD), what you can prove (reasons to believe), who you're compared to (frame of reference). The positioning statement is assembled from exactly these — To [target], [brand] is the [frame] that [promise] because [proof].
Can I edit the results?
Yes — sensibly. The passes live in your browser; the pragmatic path is to build the document, download it (HTML or Markdown), and edit by hand. Keep the banned-words discipline while you edit: exquisite, timeless, elevate your style, luxury redefined, craftsmanship (name the technique instead) are what the Forge removed — don't put them back.
Why did it used to time out, and what changed?
Reasoning models (GLM, MiniMax, Nemotron…) stream their thinking first — and thinking tokens count toward the completion budget. With the old small budget the model hit the ceiling mid-thought, the stream legally ended with no answer, and blind retries burned the same budget until timeout. Now: reasoning is switched off for these calls (brand JSON needs no chain-of-thought), the budget is 16K tokens, and if a model still thinks out its budget the Forge retries once at 32K — then tells you to switch models. Runs are also resumable: ▶ Continue from Pass N picks up exactly where a run stopped.
What if a pass fails or the model misbehaves?
Each pass has a ↻ rerun button (rerunning Pass 1 clears the later passes, since they build on it). If JSON comes back broken the Forge repairs or re-asks once automatically. Rate-limit (429) and busy models back off and retry. Try another model in ② AI Setup — free, a cheap paid pick, or your own typed ID — if one is consistently flaky.
Backup & export
Your answers, passes and document can be exported as one JSON file and re-imported on another computer — buttons below. Or just download the finished document itself. Remember: this browser is the only place your work lives.
Can I see the flow without an API key?
Yes — press ✨ Load demo on ③ The Forge (or open the page with ?demo=1). It loads the Riverbend Atelier worked example from the session: three fully-forged passes, a favourite, and the finished brand document — exactly as a real run produces, minus the API calls.
Self-test
Open this page with ?selftest=1 to run the built-in test panel — checks storage, validation, the tolerant JSON parser, prompt construction, the markdown renderer and the download builder without spending a single API request.