An AI agent for Meta Ads
An Agent runs your Meta ads. You set the limits.
Roaswise reads your ad performance every hour and proposes what to pause, scale and publish. You approve from one queue, or let it act alone inside Guardrails you set. Every change is verified on Meta and can be reverted.
- Platform
- Meta Ads
- Agent runs
- Every hour, after the sync
- Every change
- Verified on Meta, revertible
- Billing
- Monthly or yearly, cancel any time
Queue
3 pendingPaid Ads Agent · Autonomy Level: Approve
- Budget changeSummer sale, broadadset:120211847302
ROAS 4.2x against an account average of 2.8x over the last 7 days.
82%€100 → €115 / day (+15%) - PauseRetargeting, 30-day viewersad:120211903377
Frequency 6.1 and click-through rate down 38% in 14 days.
91%Pause adCreative fatigued - Publish creativeStatic #214, Glow in 2 minutesasset:214
New angle for first-time buyers. No ad in testing covers it yet.
74%Publish into Testing Campaign - Budget changeProspecting, lookalike 2%adset:120211850144
Refused by Guardrails: the most a single change may shift a budget is 20%.
77%€200 → €280 / day (+40%)BlockedCannot be approved
Example data. The controls work: approve, reject or revert a row.
How a change is made
Every change takes the same five steps.
The model never sees your raw data and never writes to Meta itself. It proposes. The platform checks, executes and verifies.
- 01
Sync
Your Meta Ads data is pulled on a schedule. If the latest sync failed or is older than three hours, the Agent does not run.
- 02
Propose
The Agent reads a briefing of your goals, your limits and your top-ranked ads, and proposes changes with a rationale and a confidence.
- 03
Check
Every proposal is tested against your Guardrails before anyone sees it. What fails is recorded as blocked and can never be approved.
- 04
Decide
Approve or reject, one by one or in bulk. A Recommendation nobody decides on expires instead of acting on old numbers.
- 05
Verify
After the change is sent, Meta is read again. It counts as executed only when Meta shows it, and it can be reverted once.
Autonomy Level
Decide how much it does alone.
Each Agent has its own level. Moving one to Autonomous takes an Admin and an explicit acknowledgement, with the limits in force shown first.
Observe
L0The Agent proposes and nothing else. Read its reasoning for a few weeks before it touches anything.
- 1Proposal
- 2Guardrail check
- 3Queue
Approve
L1Each Recommendation waits for a person. This is the level a new Agent starts at.
- 1Proposal
- 2Guardrail check
- 3Your decision
- 4Write-back
- 5Verify
Autonomous
L2The Agent executes what is in bounds. A proposal that would reverse an earlier decision still goes to a person.
- 1Proposal
- 2Guardrail check
- 3Write-back
- 4Verify
Guardrails
Limits that hold, whatever the model proposes.
A Guardrail is not a prompt and the model cannot argue with it. It is a number the platform enforces on every change before it reaches Meta.
Three scopes, one rule
Set limits for a brand, for one Agent or for one Ad Account. Where they overlap, the most restrictive value wins.
Checked twice
A proposal is checked when it is made and again at the moment it is executed, so an approval cannot outrun a limit you tightened since.
Kill Switch
One control halts every Write-back for a brand or for a single Ad Account, immediately.
Effective Guardrails
Ad Account · EUR
- Largest budget shift per change
- 20%
- Largest budget shift per day
- €500
- Hours between budget changes
- 72
- Managed budget, at most
- No limit
- New creatives per day
- 5
- Allowed actions
- Pause, enable, budget change, publish, promote
- Blackout window
- Nov 27, 00:00 to Dec 1, 23:59 UTC
- Kill Switch
- Off
Example values. Every limit is yours to set.
Creative Engine
New ads, tested against your own.
The Agent that moves budget also keeps the ad account supplied. Every Creative Asset traces back to the angle and the references it was made from.
Glow in 2 minutes
Your routine, minus six steps
What my skin looked like on day 14
- 01
Research
Creative Angles come from real customer language: a persona, a pain point and the outcome they want.
- 02
Generate
Statics and 15-second videos, built on your Brand Style Guide and your real product images.
- 03
Publish
Publishing is a Recommendation like any other. Approved ads enter the Testing Campaign.
- 04
Measure
New ads compete for budget there, on the same metrics your dashboards read.
- 05
Prune
Ads that lose or fatigue are paused.
- 06
Remix
Winners move to the Winners Campaign and brief the next round of variants.
For agencies
Every brand you manage, one approval queue.
See pending approvals, sync health and results across the portfolio, then decide without opening each brand. The brand's own Guardrails still decide what may run.
Portfolio
Last 7 days · example data
- Acme Skincare2 pendingHealthy
- Currency
- Spend
- Revenue
- ROAS
- EUR
- 4,200
- 13,100
- 3.12
- USD
- 1,850
- 4,440
- 2.40
- Northwind CoffeeHealthy
- Currency
- Spend
- Revenue
- ROAS
- USD
- 9,600
- 31,700
- 3.30
- Fjord Outdoor1 pendingDegraded
- Currency
- Spend
- Revenue
- ROAS
- EUR
- 12,400
- 29,800
- 2.40
- Totals, per currency
- Currency
- Spend
- Revenue
- ROAS
- EUR
- 16,600
- 42,900
- 2.58
- USD
- 11,450
- 36,140
- 3.16
| Brand | Pending | Sync health | Ad Accounts | Currency | Spend | Revenue | ROAS |
|---|---|---|---|---|---|---|---|
| Acme Skincare | 2 pending | Healthy | 2 | EUR | 4,200 | 13,100 | 3.12 |
| USD | 1,850 | 4,440 | 2.40 | ||||
| Northwind Coffee | None | Healthy | 1 | USD | 9,600 | 31,700 | 3.30 |
| Fjord Outdoor | 1 pending | Degraded | 3 | EUR | 12,400 | 29,800 | 2.40 |
| Totals, per currency | EUR | 16,600 | 42,900 | 2.58 | |||
| USD | 11,450 | 36,140 | 3.16 | ||||
Money is never converted or added up across currencies.
Scoped access
Limit a member to the brands they work on. The rest of the portfolio does not exist for them.
Tighter, never looser
An Agency can tighten the Guardrails a brand has set. It cannot loosen them.
Agency-wide halt
One control sets the Kill Switch on every managed brand. Each brand releases its own.
Pricing
Plans follow the ad spend you manage.
Billed monthly or yearly. Pay yearly and two months are free.
Starter
$99/ month
- Ad spend managed
- Up to 10k / month
- Ad Accounts, Agents, seats
- 1 Ad Account, 1 Agent, 2 seats
- Autonomy Levels
- Observe, Approve
- Static creatives
- 20 / month
- Videos or winner-tier images
- —
Growth
$299/ month
- Ad spend managed
- Up to 50k / month
- Ad Accounts, Agents, seats
- 3 Ad Accounts, 3 Agents, 5 seats
- Autonomy Levels
- Observe, Approve, Autonomous
- Static creatives
- 60 / month
- Videos or winner-tier images
- 10 / month
Scale
$799/ month
- Ad spend managed
- 50k and up / month
- Ad Accounts, Agents, seats
- Unlimited
- Autonomy Levels
- Observe, Approve, Autonomous
- Static creatives
- 200 / month
- Videos or winner-tier images
- 40 / month
Prices are in US dollars. Ad spend is counted in your brand's reporting currency. When a month needs more creative than the Plan includes, credit packs start at $25 and the credits never expire.
Questions
Common questions, answered plainly.
What is Roaswise?
Roaswise is an AI agent for Meta Ads, the ad platform behind Facebook and Instagram. It reads your ad performance every hour and proposes what to pause, scale and publish. You approve from one queue, or let it act alone inside Guardrails you set.
Does Roaswise change my ads without asking?
Only if you let it. Each Agent has an Autonomy Level. At Observe it only proposes. At Approve, the level a new Agent starts at, each Recommendation waits for a person. At Autonomous it executes what is inside your Guardrails, and moving an Agent there takes an Admin and an explicit acknowledgement.
What are Guardrails?
Guardrails are limits the platform enforces on every change before it reaches Meta: the largest budget shift per change and per day, the hours between budget changes, the actions allowed and blackout windows. You set them for a brand, an Agent or an Ad Account, and where they overlap the most restrictive value wins. A proposal that breaks one is recorded as blocked and can never be approved.
Does the AI model write to my ad account directly?
No. The model reads a briefing of your goals, your limits and your top-ranked ads, and proposes changes. It never sees your raw data and never writes to Meta itself. The platform checks each proposal against your Guardrails, executes it and verifies it on Meta.
Can I undo a change the Agent made?
Yes. After a change is sent, Meta is read again, and the change counts as executed only when Meta shows it. An executed change can be reverted once. A Kill Switch halts every Write-back for a brand or for a single Ad Account, immediately.
Does Roaswise make ads as well as manage budgets?
Yes. The Creative Engine generates static ads and 15-second videos from your Brand Style Guide and your real product images. Publishing one is a Recommendation like any other. Approved ads enter the Testing Campaign, ads that lose or fatigue are paused, and winners move to the Winners Campaign.
Can an agency manage several brands?
Yes. An Agency sees pending approvals, sync health and results for every brand it manages, and decides from one approval queue. A member can be limited to the brands they work on. An Agency can tighten the Guardrails a brand has set, and cannot loosen them.
How much does Roaswise cost?
Plans follow the ad spend you manage: Starter at $99, Growth at $299 and Scale at $799 a month. Prices are in US dollars. Billing is monthly or yearly, paying yearly makes two months free, and you can cancel any time.
From sign-up to a live Agent in one sitting.
Connect Meta Ads, set your Guardrails and go live. Nothing changes in your ad account until you say so.