It writes the ads. It runs the money. It learns what works.
About 2 minutes to see the plan we'd run — no card. Dry run by default; nothing is spent until you turn it on.
Cycle 14
12 Mar · next check in 2 days
Founder story · 15s
$32→$46/day · $21.40 each
Promoted$21.40
cost per purchaseFeature carousel
$28→$0/day · $103.00 each
Paused$103.00
cost per purchaseCustomer testimonial · 30s
31/50 conversions · day 4/7 · budget unchanged
Holdingbudget unchanged
Why it paused the carousel
$412 spent, 4 purchases, $103 per purchase against your $45 target. It passed the 7-day window on 10 Mar, so this is a decision, not a guess.
Written to your activity log. Nobody can edit it later — including us.
Example ad creatives for invented brands, generated to show the layouts Zolvin makes.
- Writes the copy, the image and the video script from your brand
- Cannot spend past the cap you set
- Logs every decision in plain English, permanently
Four jobs, one agent, one loop.
Most tools do one and hand you the rest.
It builds the strategy
Your brand and offer in, a campaign with targeting and budget out — created paused.
See itIt writes the image ads
Headline and image together, in the formats that win. Brand-checked before it ships.
See itIt scripts the video
Scripted from the ad already winning. You approve it twice before it goes out.
See itIt runs the money
Pauses losers, scales winners, moves budget to your cheapest result.
See itAnd it learns from all four
It reads its own results and writes what worked into the next brief.
From your brand to a live account, then round again.
The same loop every few days, with the reasoning left behind.
It starts by learning your brand, not your budget.
What you sell, what you promise, what you will never say. That becomes the brand guide — and the brand guide becomes the campaign.
Your brand guide
Answered once, applied to everything after
- Palette
- Voice
- Plain, warm, no hype
- Promise
- Sleep in 20 minutes, or your money back
- Never say
- “miracle”, “cure”, “guaranteed”
The strategy it proposed
Cold traffic, 25–44, US and Canada, interest-led rather than lookalike — you have no purchase history to model from yet. One campaign, one ad set, three creatives to start, at $40/day.
Created paused. Nothing runs until you read it and say go.
Then it writes the ads, and checks its own work.
Words and picture generated together, so they argue for the same thing. The brand check fails closed — anything doubtful waits for you.
Three variants, written and checked
Copy and image, generated together
- Passed brand check
The 20-minute wind-down
Three drops, twenty minutes, actual sleep. No grogginess in the morning.
On palette, on voice, no banned claims.
- Passed brand check
Sleep that finally sticks
The nightly ritual that replaced my melatonin habit for good.
On palette, on voice, no banned claims.
- Held for you
The cure for restless nights
Guaranteed to knock you out — miracle drops your body will thank you for.
Uses "cure", "guaranteed" and "miracle". Your guide bans all three.
The gate fails closed: anything it cannot clearly clear waits for you rather than going out. Let the clean ones publish themselves, or hold every one for review — that switch is yours.
Then it runs the money, and refuses to guess.
It waits until the numbers mean something, then acts. Losers paused, budget moved to your cheapest result — clamped, so one bad reading cannot swing the account.
Daily budget, this ad set
$120 / day · your cap
Founder story · 15s
$32→$46
raised to the cycle limit
Feature carousel
$28→$0
paused
Customer testimonial · 30s
$60→$60
unchanged · still holding
$14 of the $28 freed by the pause was left where it was. An ad still below the threshold cannot receive budget, and no single cycle may shift more than the share you allow — so the rest waits for the next one. Nothing here can raise the total above your cap.
Image ads are the volume game. It plays it.
Statics are where testing happens: cheap, fast, and you need many.
- The layouts that keep working, not a template library
- UGC-style too — person to camera, shot on a phone
- Your palette, type and voice on every one
- Held for review unless it clearly passes
UGC here means the look, generated to the brief — not a roster of licensed avatars. We do not cast anyone.
Real images, generated by Zolvin's own pipeline. The brands are invented — the pictures are not.
Video, with your hand on it at two points.
Scripted from the ad already winning — then it stops and asks. A render costs real money and carries your brand, so this gate cannot be switched off.
Video never runs on autopilot
Two of these five steps are yours
- 1
It writes the script
From the ad already winning at $21.40, not from a blank page.
- 2
You approve the scriptyoubeing built
Nothing is spent until you do.
- 3
It renders the clip
Voice, captions and cut, against your brand guide.
- 4
You watch ityoubeing built
The whole clip, before anyone else sees a frame.
- 5
It publishes
Uploaded to Meta with the approved frame as the thumbnail.
Video costs real money per render, so it is the one thing the agent is never allowed to decide alone — by design, not because the switch is missing. The agent writes the script today; the two approval screens are being built now, so an approved script does not reach Meta yet.
The part that compounds: it learns from its own results.
Anything can generate an ad. What matters is whether the thing that made it ever found out how it did.
Scoped to your campaign, built from your money — not a general theory of advertising.
What it learned this cycle
Example dataRead from your own results
- Founder story · 15s — a face, first person, no product shot until 0:06$21.40
- Feature carousel — product grid, no person on screen at all$103.00
The conclusion it drew
On this account, a person on screen in the first six seconds beats a product grid by roughly 4.8× on cost per purchase. The next thing it makes will open on a face.
Scoped to this campaign. What it learns about your sleep drops is never applied to somebody else's running shoes.
The brief it wrote next
Generated from the cycle above, not from a template
- Open on
- A person, face to camera, within the first two seconds
- Hold back
- No product shot until 0:06
- Angle
- The thing it replaced, not the thing it is
- Avoid
- Product grids — $103 per purchase on this account
This brief is what the next image and the next video script are both generated from. The loop closes here: what your money proved becomes what it makes next.
Autonomous is not the same as unaccountable.
The agent runs on its own. It never runs unsupervised — four things stay yours.
- The ceiling is yours
- Daily cap and target cost per purchase. No path past it — not a warning, a wall.
- One switch stops it
- Pause the agent from the dashboard and no cycle runs until you resume. Your ads on Meta stay as they are.
- Any call can be overruled
- Change it. Your override lands in the log beside the decision it replaced.
- You can leave
- Disconnect and it stops. Tokens sealed with AES-256-GCM while we hold them.
Everything it does, written down where you can read it.
What it waited for, what it moved, and why.
Customer testimonial · 30s
Example dataConversions
31 / 50
Days running
4 / 7
No action taken, and no budget moved. It is the cheapest ad in the set so far at $10.90. That is still not a reason to trust it — neither threshold is met, so four days of good numbers are inside the range of noise.
Daily budget, this ad set
$120 / day · your cap
Founder story · 15s
$32→$46
raised to the cycle limit
Feature carousel
$28→$0
paused
Customer testimonial · 30s
$60→$60
unchanged · still holding
$14 of the $28 freed by the pause was left where it was. An ad still below the threshold cannot receive budget, and no single cycle may shift more than the share you allow — so the rest waits for the next one. Nothing here can raise the total above your cap.
Activity
Example dataPaused Feature carousel
$103 per purchase against a $45 target, over $412 spent and 7 days live.
12 Mar 09:04
Raised budget Founder story · 15s
$21.40 per purchase, the best of the ads with enough data behind them. $32 → $46 per day, the most this cycle is allowed to move.
12 Mar 09:04
Held Customer testimonial · 30s
31 of 50 conversions, 4 of 7 days. Cheapest in the set so far at $10.90, but below threshold — no call made, and no budget moved.
12 Mar 09:04
Held Feature carousel
6 of 7 days. One more day before the window closes.
09 Mar 09:02
Append-only. The database revokes update and delete on this table.
What is running today, and what is not.
We would rather lose the sale on this page than in your second month. Here is the honest line between the two.
Live today
- Connect a Meta account; set your cap, target cost and autonomy
- A brand guide captured once, applied to everything after
- A first campaign, ad set and targeting from your offer — created paused
- Copy and image generated together, brand-checked, failing closed
- Autonomous kill, scale and hold on real performance
- Budget reallocation weighted by cost per purchase, clamped per cycle
- An append-only log of every decision and its reasoning
- Pause, run-now, override, disconnect
Not yet
- Video's two approval screens — the agent writes scripts today, but an approved one does not reach Meta yet
- No customer account has run live with real spend. You would be among the first
- Self-serve plans and billing
- Connecting an account without us — Meta's app review is still ahead
Pricing
One agent, one price, three ways to pay it. No tiers, no usage meter, no free trial. Billing is not switched on yet, so early accounts run without a charge until it is.
No free trial. You pick the plan at checkout, after we've read your page and drafted your first campaign.
One workspace, one connected Meta ad account. Managing several client accounts under one login is not supported yet.
Questions founders actually ask.
Does it make the ads, or only manage them?
Both — that is the point. It writes the copy and image together, scripts video from whatever is winning, then manages what it made. It is judging its own work.
Can I approve everything before it goes out?
Creative, yes — widen the gate and nothing ships unread. Video always waits for you, twice. Budget decisions are different: the agent makes them inside your caps and tells you after, because approving every budget nudge is babysitting with fewer tabs. Pause it whenever you want it to stop.
What does “it learns” actually mean?
Each cycle it reads its published creative against real spend, finds what your cheapest results share, and writes that into the next brief. Scoped per campaign. A model of your account, not a theory of advertising.
Is UGC done with avatars?
No. We generate the UGC look — person to camera, phone-shot, captioned — from your brief. There is no avatar roster and no licensed creators. We do not cast anyone.
What if it kills an ad that was about to work?
Nothing is killed before 50 conversions or 7 days — Meta's threshold, not ours. Below that it reports progress. Disagree with a call and override it; both land in the log.
Do I need campaigns running already?
No. It builds the first campaign, ad set and targeting from your brand and offer, created paused, then makes the first ads from the same offer and shows them to you. Nothing runs until you press Go live. If you already have ads live, it reads those instead.
Connect your ad account. Watch it think.
It reads your real ads, writes the creative, and shows you the calls it would make. Nothing changes in your account until you say so.
Early access: we connect your ad account with you, by hand, while Meta's app review is still ahead of us.