Questions, answered.

Yeah, and we're not ashamed of it. We run Claude (Anthropic) for reasoning and text, Gemini Nano Banana 2 (Google) for image, and Fal AI for video — which routes to the best-in-class models per shot (Seedance 2.0, Kling, Veo, and others). These companies spend billions training frontier models — we're not gonna compete with that, lol. What we will do is take those models and get stupidly good at a few specific jobs. Hand-written skill files. Real output filters. Years of taste baked into every prompt.

One big AI that does everything falls back on the average. A résumé specialist has spent its whole existence thinking about résumés. Specialists read the room; generalists just finish the sentence.

Merl runs on three best-in-class providers, each for what they're best at: Claude (Anthropic) for reasoning and text — résumé bullets, cover letters, voice-matched writing. Gemini Nano Banana 2 (Google) for image — headshots and interior redesigns. Fal AI for video — which routes every Cutter cast to the best-in-class video model (Seedance 2.0, Kling, Veo, and more). But raw model access is half the value at best. Merl layers prompt-craft, skill files, and output filters on top of each — tuned for one job per agent — so you get "I have a résumé that lands" instead of "here's a résumé, sorry about the MBA-speak."

Each agent has a hand-written skill file — hundreds of lines of instructions, examples, counter-examples, failure modes, and taste rules. Editor's skill file knows what a startup PM résumé looks like different from an enterprise one. Echo's skill file knows how to preserve voice without flattening it. It's the difference between giving ChatGPT a one-line prompt and giving a specialist ten years of taste. We write these skill files, test them, and update them — you just pick the agent.

A second pass on every generation. Shine runs face-identity checks and rejects drifted results. Cutter checks video output (Seedance, Kling, Veo — whichever model it routed to) for face warping and motion smear. Editor flags invented experience before it reaches you. ~30–90% of raw model output gets rejected and re-rolled before you see the final result. You only ever see the cuts that pass.

Editor (résumés + cover letters) · Echo (writing in your voice) · Shine (AI headshots) · Cutter (AI video — Seedance, Kling, Veo via Fal AI) · Designer (interior redesign).

Yes. Merl is on Google Play with the same agents and the same pricing as iOS. Phones and tablets. Cross-device sync with iOS and web. Open on Play Store.

Open app.merl.app in any modern browser — Chrome, Safari, Firefox. No download. Sign in, pick an agent, cast. Scrolls and mana sync across iOS, Android, and web.

Yes. More highly-trained specialists are in the pipeline — but we only ship an agent when it hits the same quality bar as the existing crew. For now, the crew is Editor, Echo, Shine, Cutter, Designer.

Yes. One purchase, both iPhone and iPad. Same App Store listing.

Free first cast per agent (no card). Paid plans: Day Pass $4.99 · Weekly $9.99 · Pro $19.99/month. See pricing for mana breakdowns.

Text agents (Editor, Echo) are 5 mana per cast. Image agents (Shine, Designer) are 12 mana. Video (Cutter) is 30–60 mana depending on length.

Top-up mana never expires. Pro plan mana resets monthly — it doesn't roll over if you go over the 400.

Yes — through your Apple ID subscriptions. Cancel whenever, keep remaining mana for that period. No "hit us on email" traps.

No. Ever. Your inputs aren't used to train Merl, Gemini, Veo, or anyone else's models. Your scrolls (history) live on your device unless you choose to sync.

Scrolls (cast history) are stored on-device by default. Optional iCloud sync for backup. Account info is held server-side in a single encrypted store, separate from generations. See privacy for details.

Yes, within each model's license. Gemini and Veo outputs are yours to use in work, ads, and client projects.

Yes. Paste the job description and your existing résumé. Editor extracts what the role actually asks for (not just keywords), rewrites your bullets to mirror the posting's vocabulary, quantifies every line using your real experience, and flags anything that would require inventing experience you don't have.

Yes. Paste 3–5 samples of your natural writing once; Echo builds a persistent voice profile from cadence, vocabulary, sentence length, and idiosyncrasies. Uses that profile for every cast — not generic "AI writing."

Yes. Shine is tuned specifically to preserve facial features, skin tone, and bone structure. A face-identity filter rejects any drifted result automatically — ~90% of raw generations get thrown out before you see them.

Cutter routes each shot via Fal AI to the best model for the scene (Seedance 2.0, Kling, Veo) then runs a quality-filter layer specifically for face integrity and motion coherence. Most model-native failures are caught before you see them.

Yes. Designer extracts your room's architecture first — walls, windows, light direction, existing floor plan — before restyling. You get new furniture, palette, and textures placed where they actually fit your specific room. Not a generic room that vaguely matches the style.