A reply that breaks your rules never sends.
You set the limits — a discount cap, claims it can never make, refunds it can never approve. Before any reply leaves, a rule check reads it against those limits — the same verdict, every time. Break one and HyperDM blocks the message and hands the conversation to you. It doesn't hope the model behaves.
Trust the model alone, and it writes the reply you'd never approve.
An AI that writes its own replies can write the wrong one. To close a stubborn buyer it might conjure a discount you never offered. Asked whether your serum works, it might call it ‘clinically proven.’ Told a customer wants their money back, it might say the refund's approved. One reply like that at 3am and you're honoring a deal you didn't set, or answering for a claim you can't back. So most brands pick the other option — leave automation off and answer every DM by hand. One costs your margin. The other costs your nights.
- ✕Invents a discount to win the sale — and now you owe it
- ✕Makes a health or performance claim you can't stand behind
- ✕Tells a customer a refund is approved before you've seen it
- ✕Answers a question it should have flagged — confidently, and wrong
Every drafted reply passes a hard check before it can send.
HyperDM's agent drafts a reply grounded in your catalog. Before that reply leaves, a deterministic pre-send check reads it against the rules you set — not a second AI second-guessing the first, but a plain, predictable check that returns the same verdict every time. A clean reply goes out in about two seconds. One that breaks a rule is never delivered: HyperDM pauses the agent for that conversation and drops a note in your inbox saying exactly what it caught — blocked · 40% off > your 10% cap. Your prompt asks the model to behave. This is the part that doesn't ask.
Draft
the agent writes a reply grounded in your real catalog
Check
a rule check reads it before it can leave, the same way every time
Send or hand off
clean replies ship in ~2s; rule-breakers are blocked and routed to you
Four rules. You set them once. They hold on every reply.
These are the limits you hand the agent. Set them in a click, change them anytime — they're enforced on every automated reply from that moment on.
Discount cap
set your maximum. Any reply offering more than that is blocked before it sends. It's precise about it, too: ‘100% cotton’ and ‘true to size’ never trip the cap.
Forbidden claims
add the phrases you can't make, like ‘clinically proven’ or ‘guaranteed results.’ Any reply containing one is blocked, case-insensitive.
Refund approvals
off-limits. The agent never approves a refund; it tells the customer a teammate will follow up, then routes the thread to you.
Hand off when stuck
no confident answer, or a reply it just blocked? The agent stands down for that conversation and pings you in the inbox, so silence never wins the sale.
It's grounded before you set a single rule.
The four rules above are yours to tune. These hold from the first message, with no setup. HyperDM only states a price, a stock count, a shipping window, or a policy that lives in your catalog and knowledge base. If the answer isn't there, it says it'll check — it doesn't guess. It never invents a discount or promises a delivery date. And anything a customer types, or that HyperDM reads off your own site, is treated as data to answer from, never as instructions. Try to jailbreak it in a DM and it just keeps selling.
- ✓Only quotes prices, stock, and policy that exist in your catalog
- ✓Says ‘let me check’ instead of guessing when it doesn't know
- ✓Never invents a discount or a delivery date
- ✓Ignores injected ‘instructions’ hidden in a message or a scraped page
When it doesn't know, it taps you in — with the receipts.
Every operator asks the same thing before trusting automation: what happens when it's wrong, or out of its depth? It stops. A block pauses that one conversation and leaves a note in your inbox — what it caught and why — so nothing goes out behind your back and nothing quietly disappears. You take over in the same inbox, sort it, and hand it back when you're done. Your margin stays yours. Your claims stay defensible. Your brand keeps saying only what you'd say. And every send — and every block — is logged, on the official Instagram API, within policy on every message.
Every block and every send is logged
you always have the receipts
The conversation pauses; it never dead-ends on the customer
Take over any thread anytime, and hand it back when you're done
Runs on the official Instagram API, within policy on every send
The honest FAQ.
Automate the DMs. Keep your word.
Your agent can be live in about ten minutes with the safe defaults already on — refunds escalate to a human, and every reply stays grounded from the first message. Add your discount cap and blocked claims in a click. Your first 50 conversations a month are free. No credit card.