HookSendOne moment
AI features
It answers your DMs from your own documents, moderates your comments, finds the brand deals in your inbox, writes your campaigns, runs the AI steps inside a DM flow, and connects to your own AI so you can ask for all of it in plain words. Four of them have a demo you can try on this site without signing up.
Upload your price list or brochure. Customers get answered the way you would answer them, in their own language, at three in the morning.
fees kitni hai?
Beginner batch ₹3,500 per month hai, hafte mein do class.
Two thousand messages a day, and ten of them are brands offering you money. Those ten get pulled out and put on their own screen before they scroll away.
Paid collab from @urbanthreads
2 reels, asked for your rate card
Abuse and scam links under your reel are hidden within seconds, so your customers never read them. Real questions are left exactly where they are.
cheap followers, DM me
hidden before anyone saw it
Run a giveaway that collects the email and the phone number by itself, then sends the link. Every button is counted, so you know who actually opened it.
Ask email → Ask phone → Send the link
1 click, 14 seconds after it was sent
Upload what you already have — a price list, a brochure, your website — and the AI answers questions from it. It is not a chatbot with opinions. Everything it says has to come out of your documents, and when they do not cover the question it says so instead of inventing an answer.
Upload a PDF, paste a URL, or paste the text straight in, and HookSend reads it, splits it into passages and answers DMs from those passages. Use it for the questions you answer forty times a week: price, timings, sizes, what is included. Each answer is built from at most four passages of your own documents, and the answer is capped short enough to read in one glance on a phone.
See it runThe AI is told to use only the passages it was given, and never general knowledge. If they do not clearly answer the question it returns nothing, and the code throws the answer away rather than sending an empty DM. On a keyword chat the person gets your fallback line instead; with Answer every DM switched on nothing is sent at all, so the message is still sitting in your inbox for you. This is the boundary that makes it safe to point at your own prices.
One switch in Settings, and the AI reads every incoming message rather than waiting for a keyword. It searches across all your live knowledge bases at once and the one that genuinely covers the question answers, rather than whichever was set up first. Anything it cannot answer is left for you, and a message that is only emoji never reaches the AI at all.
Every answer can carry four or five tappable follow-up questions, and the list they come from is written by the AI once, from all your documents at the same time — not from the few passages that answered this question. So an answer about what you do can still suggest the prices and the payment options that live in a different document. A tap sends the whole question, not the short label, so the next answer is a proper one.
Hindi in Devanagari, Hinglish typed in English letters, Tamil, Bengali, Urdu, Arabic, Russian, Greek, Thai, Japanese, French, German or Spanish — the reply comes back in the language the question arrived in, even when your documents are in English. The language is worked out in code and handed to the model as a direct instruction, because the smaller models left to themselves answered a Hinglish question in English. The suggestion buttons are translated too.
Write a line or two describing how you want your replies to sound and every answer follows it for tone, length and wording. Use it when the default reads too formal for your account, or too casual. The AI always talks about your business by name in the third person and never pretends to be you, so an answer never puts words in your mouth.
Switch this on and a long answer arrives as up to three short DMs with a typing indicator and a short pause between them, the way a person sends a thought in pieces. The split is done in code at sentence boundaries after the answer is written, so it never cuts mid-word or mangles a price. Each extra message counts as one more DM from your monthly allowance.
The AI is given the last four question-and-answer pairs, so "how much is it?" knows what "it" is. History is only there to work out what the person means — every fact in the reply still has to come from your documents. With Answer every DM on, it looks back over the last six hours without needing a keyword or an open chat.
The moment you answer someone yourself — in the Instagram app or in the HookSend inbox — the AI goes quiet in that conversation for 30, 60 or 120 minutes, whichever you chose. It also closes the open chat, so when the pause ends it does not pick up mid-thread as though nothing happened. This is the one behaviour customers notice: nothing is worse than a bot talking over you while you are mid-negotiation.
If somebody sends nothing but a stop word, the AI goes quiet for that person for 24 hours and sends your goodbye line if you wrote one. The whole message has to be the stop word, so "how do I cancel my order?" is still treated as a question worth answering rather than a request to be left alone.
A comment-to-DM campaign can carry a knowledge base, so when someone replies to the DM your link arrived in, the questions that follow get answered from your documents instead of hitting a dead end. If the documents do not cover it, the campaign's own written message is sent exactly as before.
Retrieval uses a multilingual embedding model rather than an English one, which is the difference between a Hindi question finding the Hindi paragraph in your brochure and finding nothing at all. A question typed in one language can also match a passage written in another.
Comment Guard reads the comments under your posts and reels and hides the ones that cost you customers — abuse, scam links, fake support accounts phishing for OTPs — while leaving real questions exactly where they are. The pattern rules are free and run on every plan; the AI rules are the ones that read meaning.
Three AI rules ship switched on: abuse and profanity, attacks calling your product a scam or a waste of money, and handles posing as you or your support team to phish your customers. Each is a question put to the model, which answers with how confident it is. Abuse and product attacks default to hiding, which is reversible; impersonation is the one rule that deletes by default, because leaving a fake support comment up is itself the harm.
See it runA comment made entirely of ordinary words cannot be abusive, a scam or an impersonation, so it never reaches a model. Links, phone numbers, emoji floods, hashtag stuffing and tagging your buyers are all caught by pattern rules that cost nothing. One unrecognised word anywhere and the whole comment goes to the AI — the error always falls in the direction of looking more closely, and this is why an AI credit buys far more comments than it buys model calls.
You choose how sure the AI has to be before it acts, on a slider that starts at 80%. At or above your line it acts on its own. Within 30 points below it the comment goes to a review queue for you rather than being guessed at. Below that it is left alone. Every AI rule is scored before anything is decided, so the strongest call wins instead of whichever rule happened to be checked first.
Beyond the built-ins you can write your own rule as a question in your own words — up to 600 characters — and choose whether a match is hidden, deleted or sent to you. Use it for the thing that is specific to your account: a competitor's name, a claim you keep having to correct, a question that always means a refund. All your enabled rules are answered in one call, so a custom rule does not cost more than the built-in ones.
The backtest replays your existing comments through your current rules and shows you what would have happened, on your own audience in your own languages. It reaches back 90 days by default and you can widen or narrow that. It reads and scores; it never hides, deletes or replies to anything. It also lists the comments the AI was least sure about, which are the ones worth your attention before you switch it on for real.
Run Comment Guard in watch-only for up to a fortnight and every verdict is recorded while nothing is actually touched on Instagram. It is the honest way to answer "would this have deleted a customer?" — you read the log for a week first, then turn it loose.
The free rules run before the campaign, so a comment that is plainly a scam link is gone before anything is sent. The AI rules run behind the DM, a few seconds later, and hide or delete then if they decide to. That ordering is deliberate: somebody who comments your keyword expects a DM within seconds, and no model is fast enough to sit in front of that.
Fifty people commenting "price" under one reel is one AI check, not fifty. Identical comments are served from a verdict that is cached for a day, so a viral post costs a fraction of what its comment count suggests. There is also a ceiling of 2,000 AI comment checks per account per hour, so a post that takes off overnight cannot spend your whole month before you have seen it happen.
Spam links, contact poaching, emoji floods, hashtag stuffing, tagging your buyers and every word you typed yourself are pattern rules. They cost nothing and keep working whatever your balance says. Only the rules that need a model stop, and the dashboard tells you so rather than leaving you to notice.
Guard watches from the moment it is on, which does nothing about the spam under a reel from six months ago. The clean-up tool loads an old post, scores its comments and shows you what would go. You pick what actually goes, and nothing reaches Instagram until you do.
Alongside your rules the AI is always asked one more question: is this a real customer with a real problem — a late order, a damaged item, a refund they are chasing? It is told explicitly that such a person is not abuse however angry they sound, and that a criticism sitting next to a compliment is not an attack. You can also protect anyone you have captured as a lead or messaged before, so a rule can never reach them.
The model is told to judge the comment in whatever language it arrived in and never to assume English, so a Hinglish insult and a German one get the same treatment. Romanised Hindi, Tamil, Telugu and Bengali are read as those languages rather than as misspelled English. Sarcasm is treated as genuinely hard: when the AI is unsure it is told to give a middle score, which sends the comment to you.
Two jobs a creator does by hand and would rather not: writing the DM, and setting up a campaign from a reel that is already doing well. Neither one publishes anything on its own — both hand you a draft to read, edit and approve.
Every message a campaign sends can be written or rewritten by the AI: the main DM with your link, the opening greeting, the delayed follow-up, the follow-gate ask, the email request, the thank-you afterwards and the public reply under the comment. Leave the box empty and it writes from scratch; leave your own words in it and it rewrites those. It is charged once per campaign a month, so tweaking and regenerating inside that campaign is free after the first time.
One press gives you four alternates built from your own wording, and HookSend rotates between them and your original so the same message does not go out identically to everybody. There is no tone dropdown, on purpose: the AI is told to keep your tone, your emoji style and your level of formality rather than to pick a personality. Which variation a person receives is worked out from who they are, so a retry never reads as two different people writing.
The AI writes in the same language and script you wrote in. More usefully, the placeholders are protected in code rather than by asking nicely: a variation that lost your {link}, came back empty or ran over Instagram's message limit is discarded and the next model is asked instead. Near-identical variations are thrown out too, so four variations are actually four.
Paste a link to one of your own posts or reels and HookSend reads the caption and drafts the campaign: a name, the keywords to watch for, the DM to send and the public reply. The draft is never saved — it opens in the builder for you to approve, because AI misreads captions and picks keywords that would fire on every comment. It is told never to invent a discount, a price or a deadline the caption does not support, and it can only read captions from an account you have connected.
A DM flow is a guided conversation with buttons and questions. Two of its steps are AI, and both are built so that a model being slow or a balance running out never leaves somebody waiting in your DMs.
Drop an AI answer step into a flow and that point in the conversation is answered from your documents instead of from a script. Use it where you cannot predict the question — after the link has been sent, or when the person types something the buttons do not cover. If the flow ends on one, the chat stays open with that knowledge base, so whatever they ask next is still answered.
See it runDescribe up to eight paths in your own words and an AI routing step reads one reply and decides which the message belongs on, in whatever language it was written. It answers with a path number rather than a label, because models paraphrase labels. Every failure — no credits, no model, an answer that does not parse — takes the Otherwise path you chose, so a routing step can never leave a person hanging.
You can describe a flow to your own AI client and have HookSend build it. It is checked before it is saved and refused if it could not run, naming what is wrong, and a flow built this way is saved unpublished — nothing answers a real person until you publish it yourself.
A creator with a busy inbox gets two thousand messages and ten of them are brands offering money. The AI reads incoming DMs as they arrive, works out which are genuine approaches, and pulls the terms out onto their own screen.
Paid offers, barter and gifting, affiliate and commission arrangements, and invitations to paid events are all recognised — including the opening approach that has not named a brand or a budget yet, because a large share of real pitches are written that way. A fan admiring a collab you already did, and someone asking you to promote them for free, are not.
See it runThis is the part that makes the feature worth having. "Minimalist, 18,000 rupees, one reel and two stories, by the 28th" is something you can act on from a list; a message snippet is not. The budget is copied exactly as the brand wrote it — 18k, ₹18,000, 600 euros — and never converted, so you are negotiating against their words rather than our arithmetic. Each offer is also sorted into paid, barter, affiliate or event, and marked as coming from a brand or an agency.
Brands write deadlines the way people talk — Oct 5, 5th October, by the 28th — with no year in sight. HookSend works the year out, so a brand writing in December about "Jan 10" means the coming January rather than one that has gone. A date that makes no sense is left blank instead of guessed, because a wrong deadline on this screen is worse than an empty one.
The clearest sign of a fake offer is that it asks you to pay: a registration fee, a refundable deposit, a shipping charge, a link to pay on. A real brand pays the creator. That gets its own score alongside the collab score, and a message that scores high on it goes to a separate Scams tab, so the list you actually read is only real offers.
One digest a day, at the hour you choose in your own timezone, listing who wrote, what they offered and how many are still waiting on a reply. On a day nobody approached you, nothing is sent — a morning email saying "nothing today" is one you stop opening, and silence is information. It also tells you how many went to the Scams tab.
A brand pitch is never a dozen words, so anything shorter than that is filtered out for free before a model is asked. The handle is looked at too. It means your credits go on the messages that might actually be offers, not on every "hii" in your inbox.
It reads, files and summarises. It does not answer, because a collaboration is a negotiation and the first number you send is the one you are held to. If the AI cannot make sense of a message it stays in your inbox exactly as it arrived — a missed offer is a message you still have, whereas a wrongly filed one costs you attention on every screen.
Everything above is AI working inside HookSend. This is the other direction. Connect HookSend to Claude, Codex, Antigravity or any client that speaks the Model Context Protocol, and you can run your Instagram automation by asking for it. Your AI does the thinking on your own subscription; HookSend carries out the actions. It is included in the free trial and on every paid plan.
Say which reel, which keyword and what the DM should send, and it is set up: the link, the product carousel, the follow gate, the email question before the link, the thank-you afterwards.
Say what the conversation should ask and where each answer leads. HookSend validates the flow before saving and refuses one that could not run, naming what is wrong. It is saved unpublished, so nothing answers a real person until you say so.
Views, reach, saves and shares per post, your sends and clicks, the leads you captured, your Comment Guard decisions, and how much of your DM allowance and AI credits are left.
There are 60 tools in all, covering what you would otherwise do in the dashboard. Every one runs through the same checks the dashboard uses, so your plan limits and your allowance apply exactly as they do on screen — the connector is not a way around your plan. Buying anything, changing your billing and inviting a colleague all refuse an AI client outright.
How to connect it, and the full tool listThese are the reasons to choose this over something cheaper, so they are on the page rather than in the small print. All of them are enforced by HookSend, not by asking a model nicely.
The AI is given passages from your own files and told to use nothing else. When they do not cover the question it returns nothing rather than filling the gap, and the code checks that before a DM can be sent.
Answer somebody by hand and the AI goes quiet in that conversation for as long as you chose, and closes the chat so it cannot resume mid-thread. It also waits while a flow or a campaign is expecting an answer.
No connector tool deletes a campaign, a flow or a knowledge base — where the dashboard deletes, the connector pauses or archives instead. The few actions that genuinely cannot be undone refuse the first request and make your AI check with you.
A flow built by your AI client is saved unpublished. A new Comment Guard rule is saved switched off. A campaign drafted from a link is handed to you in the builder and never saved on its own.
The AI rules in Comment Guard, answering every DM, and collab detection are all off on a new account. You turn on the ones you want, when you want them, and the backtest and watch-only mode are there so you can see what a setting would have done before it does it.
Every AI call returns nothing rather than throwing. If no model answers, a comment is left alone, a routing step takes your Otherwise path, and saving a campaign or delivering a DM is unaffected.
Everything runs on Instagram's official API from an account you connect on Instagram's own screen. HookSend is an approved Meta Tech Provider. Nothing here drives a browser pretending to be you.
AI work is charged in credits, and a credit is only ever spent on a real model call. Comment Guard’s pattern rules cost nothing, and the free rules handle most comments before the AI is asked, which is why a credit buys far more comments than it buys model calls.
The copy assistant and the campaign builder are charged once per campaign a month, so regenerating and refining inside that campaign after the first time is free.
How AI credits work, and what each plan includesHookSend is not wired to one model or one company. A request goes down a chain of models across several providers and the first usable answer wins. A provider that is rate limited, timing out or returning errors is skipped and the next is tried; one that keeps failing is stood down for a while and then given another go. Two more things follow from that, and both matter more than the chain itself.
Every reply is checked in code before it is used. A variation that lost your link, a moderation verdict that answered a question nobody asked, a routing step that named a path that does not exist — each is discarded and the next model is asked instead. Falling back is not only for a provider being down; it is also for an answer that does not hold up.
Every AI call returns nothing rather than failing loudly. If no model answers at all, a comment is left where it is, a flow takes the Otherwise path you chose, and the campaign still saves and the DM still sends. The AI degrades a button; it never blocks the product.
Meta reviewed HookSend and approved it. Everything here runs on Instagram's official API, inside the limits Instagram sets, from an account you connect on Instagram's own screen.
You connect on Instagram's own screen. We never see your password and never sign in as you.
A viral night is queued and answered in order, at the rate Instagram allows.
Disconnect from Instagram or from here, and everything stops at once.
HookSend is built and run by HOOKSEND SOFTWARE, an approved Meta Tech Provider. Meta reviewed the app and approved every permission it uses. Every message and every hidden comment goes through Instagram’s official API, paced inside Instagram’s own limits. We never ask for your password, never log in as you, and never automate a browser pretending to be you, which is what the tools that do get accounts restricted are actually doing.
About HOOKSEND SOFTWAREFifteen days of all of it, no card. Then a free plan that does not expire.
