✦ Safety & control

Mira can work. She cannot run wild.

"An AI that spends my money" is a fair thing to fear. So Mira's autonomy is a fence you draw and can collapse in one click — every move that costs money or touches your brand waits for you. Here is exactly how that's enforced, not just promised.

The controls you hold

Six guardrails, always on. None of them are optional, and none of them are things Mira can turn off herself.

₹

Budget envelope

You set the spend limit — overall, per channel, and optionally per campaign. She works inside it. Set the wallet to ₹0 and every paid action stops cold.

☷

Approval thresholds

New paid campaigns, any scale-up, positioning changes, public performance claims and large sends require an explicit yes from you before they go live.

△

Anomaly halt

Spend spikes, ad-account restrictions, broken tracking or public backlash are escalated to you immediately — she doesn't push through a red flag.

Ⅱ

One-click pause

Pause Mira and every wallet-consuming action freezes. The interface shows you exactly what's paused and what's still running.

▤

Decision logs

Every action carries its evidence, confidence and reason. Nothing happens off the record — your ledger shows what she did and why.

◎

Rival AI audit

A second, independent AI (ChatGPT) grades Mira's decisions for you — a built-in adversary, so you're never taking her word alone.

What an approval looks like

When a decision crosses the line, Mira doesn't bury it in a paragraph. She brings you a card: the decision, the spend, her reasoning, the kill rule, and why she needs you — with three clear buttons.

Needs your approval

Decision: Launch a Meta test campaign to founders.
Spend: ₹2,000/day for 5 days (₹10,000 cap).
Why: Your landing-page audit surfaced a sharp founder-pain angle that isn't being used in any live ad.
Kill rule: Stop automatically if CAC exceeds ₹2,500 after the minimum spend.
Needs you: It's a new money-spending campaign — outside what she runs on her own.

Three honest outcomes, every time:

Approve — she launches inside the cap and the kill rule, and reports back.
Reject — nothing happens, and she logs why you passed so she learns your line.
Ask Mira to revise — she reworks the angle, budget or audience and brings you a new card.

This is a UI demonstration of the pattern. Real approvals appear in your console and on the channels you've connected.

She labels how much to trust the data

The most dangerous AI is a confident one with bad data. Mira refuses to fake certainty.

Verified vs inferred alwaysShe separates a measured fact from a modelled conclusion. Attribution is a model, not gospel — and she says so.

Tracking health: healthy / partial / brokenIf the pixel or conversions are broken, the number is flagged — not reported as truth.

"Inconclusive" is a valid answerThin sample or conflicting sources → she says the data won't support a call, instead of over-deciding on noise.

No blind scalingNo clean attribution, no scale-up. She will not pour more money into a campaign she can't measure.

What Mira will not do

Credibility comes from the limits. A submissive agent is a dangerous one.

No fake proofNo fabricated testimonials, fake screenshots, invented client names or unverifiable claims. A wrong fact is worse than a missing one.

No spamNo high-volume sends without consent, suppression lists and frequency caps.

No silent spend or scale-upSpending up is never invisible — it always passes through an approval you see.

No going quiet on emergenciesSpend spikes, account restrictions or public backlash come to you immediately.

No repositioning the brand alonePricing, partnerships and positioning are yours to approve — never hers to decide.

Autonomy is a setting, not a default you're stuck with. You choose how much rope Mira has, raise it as you build trust, and pull it back to approval-only whenever you want. The fence is yours to move.

Who can do what — at a glance

The simplest version of the rules: what Mira does on her own, and what waits for you.

Action	Can Mira do it alone?	Why
Draft ad copy	Yes	Low risk, fully editable before anything ships.
Analyze your website	Yes	No external action — she's only reading.
Pause an obvious loser	Yes, if a rule allows	Pausing protects your budget — conservative by design.
Launch a new paid campaign	No	Spends money and carries brand risk — needs your approval.
Scale budget	No	Financial risk — you confirm the step-up.
Send a sensitive claim	No	Brand and legal risk — never hers to decide alone.
Delete data	No	Irreversible — always escalated to you.

Try her with the brakes on.

First month free, ₹1,000 in credits, and no ad spend without your approval. Judge her before you trust her with anything that costs money.